Directory Update Log

2026-10-15 22:58

• New concept: generation prior conflict — pretraining priors override system prompt rules (knowledge/context-engineering/generation-prior-conflict.md)
• New concept: CNL authoring rules for weak models — ASD-STE100 principles applied to agent instructions (knowledge/context-engineering/cnl-authoring-for-weak-models.md)
• New concept: assistant prefix continuation framing — empirical finding on Qwen3-14B (knowledge/context-engineering/assistant-prefix-continuation.md)

2026-10-15 22:57

• New concept: Goose message role mapping — how goose maps to OpenAI API roles, monolithic system prompt structure (knowledge/context-engineering/goose-message-role-mapping.md)

2026-10-15 21:50

• Session Canary rule: rewrite to CNL-style numbered Do: list — one instruction per sentence, active imperative voice (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-10-15 21:45

• Sync frontmatter version to match CHANGELOG latest (1.1.0 → 7.2.0) (skills/skill-harvest/SKILL.md)
• Sync frontmatter version to match CHANGELOG latest (1.2.0 → 7.2.0) (skills/okf-bundle-harvest/SKILL.md)

2026-10-15 21:44

• Sync frontmatter version to match CHANGELOG latest (2.0.0 → 2.0.1) (skills/agentfs-git-push/SKILL.md)
• Extend rule format mandate: Do: block with 2+ actions must use numbered list, no multi-action prose paragraphs (skills/agentfs-setup/references/design-spec.md)

2026-10-15 21:43

• Add CNL authoring checklist items: one instruction per sentence, active imperative voice, single vocabulary (skills/skill-gen/SKILL.md)
• Add CNL authoring checklist items: one instruction per sentence, active imperative voice, single vocabulary (skills/skill-gen/SKILL.md)

2026-10-15 20:32

• Update orchestration: deterministic model delegation, name resolution instructions, remove agent semantic triggering (skills/skupper-model-provider/SKILL.md)
• Add validate_host() and validate_profile() strict validation functions (skills/skupper-model-provider/scripts/common.sh)
• Add –profile, deterministic model stop before VAN teardown (skills/skupper-model-provider/scripts/down.sh)
• Add –profile, –list-sites, deterministic model delegation via hosted-model-ctl (skills/skupper-model-provider/scripts/up.sh)
• Add –names-only flag for script-consumable profile name listing (skills/hosted-model-ctl/scripts/list.sh)

2026-10-15 20:04

• Display composite active profile with sub-model names (skills/hosted-model-ctl/scripts/list.sh)
• Add composite profile handling — stop all sub-profiles, clear active (skills/hosted-model-ctl/scripts/stop.sh)
• Add composite profile handling — start all sub-profiles, set composite as active (skills/hosted-model-ctl/scripts/start.sh)

2026-10-15 19:54

• Add decision listener status display (skills/skupper-model-provider/scripts/status.sh)
• Add decision connector on rhel-ai and decision listener on localhost for port 9001 (skills/skupper-model-provider/scripts/setup.sh)
• Add RHEL_AI_DECISION_PORT and RHEL_AI_DECISION_ROUTING_KEY (skills/skupper-model-provider/topology.env)
• New concept: D3 decision model deployment on L4 — CUDA graph OOM, FP8 KV crash, optimization, NextToken vs SDM (knowledge/llm-inference-constrained-gpu/d3-decision-model-deployment.md)
• Add decision listener status display (skills/skupper-model-provider/scripts/status.sh)
• Add decision connector on rhel-ai and decision listener on localhost for port 9001 (skills/skupper-model-provider/scripts/setup.sh)
• Add RHEL_AI_DECISION_PORT and RHEL_AI_DECISION_ROUTING_KEY (skills/skupper-model-provider/topology.env)
• Add Section 17: D3 decision model benchmarks, CUDA graph OOM, optimization, side-by-side comparison (skills/hosted-model-ctl/references/benchmark-report.md)
• Add qw14b+d3 composite profile section, fitting score, speed comparison (skills/hosted-model-ctl/references/deployment-profiles.md)

2026-10-15 19:53

• Add composite profile deployment (+ prefix), per-profile env vars via CONTAINER_ENV (skills/hosted-model-ctl/scripts/setup.sh)
• Add composite profile qw14b+d3, CONTAINER_ENV for expandable_segments, d3-27b-8k and d3f-9b-32k profiles (skills/hosted-model-ctl/scripts/common.sh)

2026-10-15 15:57

• Remove spec decode from qw14b-32k profile: vllm-spec→vllm, clear spec config (skills/hosted-model-ctl/scripts/common.sh)
• Add Section 16: vLLM 0.32 upgrade + spec decode removal benchmarks for both qw14b profiles. Update Section 9.3 and 10 recommendations. (skills/hosted-model-ctl/references/benchmark-report.md)
• Update qw14b-32k: remove spec decode, speed 30.2 tok/s, fitting score 38→40.3, add decision endpoints (skills/hosted-model-ctl/references/deployment-profiles.md)

2026-10-15 15:39

• Fix vLLM 0.32 reasoning field: check both reasoning_content and reasoning in jq fallback chain (skills/hosted-model-ctl/scripts/test.sh)

2026-10-15 15:37

• Fix: regen-skill-index.py exit code 1 (version warnings) killed post-edit.sh via set -e before knowledge index regen could run. Added

  true to suppress.

2026-10-15 15:28

• New concept: speculative decoding tradeoffs — when it hurts on bandwidth-bound GPUs (L4 benchmark evidence, vLLM V1 issues) (knowledge/llm-inference-constrained-gpu/speculative-decoding-tradeoffs.md)

2026-10-15 15:00

• Bump version to 7.9.4 for spec decode removal (skills/hosted-model-ctl/SKILL.md)
• Update qw14b-40k: remove spec decode, speed 41.2→51.3 tok/s, add spec decode removal rationale, update fitting score 39→42 (skills/hosted-model-ctl/references/deployment-profiles.md)
• Remove spec decode from qw14b-40k profile; engine vllm-spec→vllm, clear spec config, update speed estimate (skills/hosted-model-ctl/scripts/common.sh)

2026-10-15 11:29

• Add deployment profiles to Quick Navigation table (README.md)
• Add fitting score summary table with all 6 profiles at top of doc (skills/hosted-model-ctl/references/deployment-profiles.md)

2026-10-15 11:20

• Add deployment-fitting-guide to knowledge bundles list (14 bundles now) (README.md)

2026-10-15 11:19

• Add doc-alignment check call at end of list output (skills/hosted-model-ctl/scripts/list.sh)
• Add doc-alignment check call at end of list output (skills/hosted-model-ctl/scripts/list.sh)

2026-10-15 11:16

• Add doc-alignment check: detects drift between deployment-profiles.md default marker and persistent state file; update builtin default for rhel-ai to qw14b-40k (skills/hosted-model-ctl/scripts/common.sh)

2026-10-15 11:08

• Update default profile marker: g8b-fp8-spec-128k → qw14b-40k to match actual persistent default (skills/hosted-model-ctl/references/deployment-profiles.md)
• Update default profile marker: g8b-fp8-spec-128k → qw14b-40k to match actual persistent default (skills/hosted-model-ctl/references/deployment-profiles.md)

2026-10-15 10:31

• Added qw14b-32k and qw14b-40k profiles, GPU_DEVICES map, spec configs, renamed 64k to 40k (skills/hosted-model-ctl/scripts/common.sh)

2026-10-15 10:27

• New concept: sizing rules of thumb for context and parameters (knowledge/deployment-fitting-guide/sizing-guidance.md)
• Added Qwen3-14B row, corrected Qwen3 dense context to 40K, added deployment notes (skills/hosted-model-ctl/references/model-landscape.md)
• qw14b-40k measured scores, renamed from 64k (skills/hosted-model-ctl/references/fitting-guide.md)
• Section 15: qw14b-40k benchmark, context discovery, agentic observations (skills/hosted-model-ctl/references/benchmark-report.md)
• Renamed qw14b-64k to qw14b-40k, added measured benchmark results (skills/hosted-model-ctl/references/deployment-profiles.md)
• Version bump to 7.9.0 for qw14b-40k docs (skills/hosted-model-ctl/SKILL.md)

2026-10-15 09:47

• Fix: setup.sh now restores active_provider after teardown+setup (recreate). Teardown removes active_provider but setup never set it back, leaving Goose Desktop with no provider on restart.

2026-10-14 22:41

• Added KGM registration step to Post-Creation Checklist — agent registers skill entity in KGM after creation so hey signal dispatch can find it (skills/skill-gen/SKILL.md)

2026-10-14 22:05

• Port-based mutual exclusion: stop any model container on target port before deploying (skills/hosted-model-ctl/scripts/setup.sh)

2026-10-14 21:57

• Version bump to 7.8.0 for Qwen3 14B profiles (skills/hosted-model-ctl/SKILL.md)
• Section 14: Qwen3 14B benchmark results, parser discovery, fitting score (skills/hosted-model-ctl/references/benchmark-report.md)
• Added qw14b-32k and qw14b-64k profiles, GPU_DEVICES map, spec configs (skills/hosted-model-ctl/scripts/common.sh)

2026-10-14 16:28

• v7.7.0: Added deployment fitting guide integration and scoring triggers (skills/hosted-model-ctl/SKILL.md)
• New reference: Applied fitting guide with per-profile scores (skills/hosted-model-ctl/references/fitting-guide.md)
• v7.7.0: Added deployment fitting guide integration and scoring triggers (skills/hosted-model-ctl/SKILL.md)
• New reference: Applied fitting guide with per-profile scores (skills/hosted-model-ctl/references/fitting-guide.md)
• New concept: Weighted scoring formula 0.30/0.30/0.30/0.10 with interpretation bands (knowledge/deployment-fitting-guide/weighted-scoring.md)
• New concept: Class-specific baselines - Gen AI (Opus) vs Decision (Jev) (knowledge/deployment-fitting-guide/class-baselines.md)
• New concept: 4x4 SUV positioning analogy for deployment traits (knowledge/deployment-fitting-guide/suv-model.md)
• New concept: The 4 deployment traits - Quality, Speed, Context, Size with correlations (knowledge/deployment-fitting-guide/four-traits.md)

2026-10-12 14:42

• v1.4.0: Added GPU/Wayland –ozone-platform=x11 fix (skills/goose-desktop-env-fix/SKILL.md)

2026-10-02 22:29

• New concept: confabulation-under-audit — model fabricates retroactive justifications for non-compliance. Evidence from Granite 8B canary audit. (knowledge/llm-behavioral-properties/confabulation-under-audit.md)
• P1: added weak-model compatibility checklist (formatting directives, concrete examples, context-aware references, wrapper scripts). P7: added script composition safety (SIGPIPE, set -euo pipefail patterns). (skills/skill-gen/SKILL.md)

2026-10-02 22:02

• v7.3.3 Rule 4: ‘start your response with’ replaces ‘Emit’ — unambiguous formatting instruction for canary placement (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-10-02 21:23

• v7.3.2 Rule 4: simplified — ‘These rules are already in your context. Treat as authoritative.’ Removed file-reference abstraction. (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-10-02 21:04

• v7.3.1 Rule 4: explicit ‘this document is AGENTS.md — already in context, do not search on disk’. Authoritative statement preserved. Canary format specified (two-word, ephemeral, with example). Anti-daydreaming: remember in-session, do not write anywhere. (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-10-02 16:22

• v7.3.0: Rule 2 cross-references Rule 3 (Post-Write) and Rule 6 (Checkpoint). Rule 4 noise removal — clearer obligations. Rule 1 step 2 trimmed redundancy. Quick Orientation and Scope Definitions signal-to-noise improvements. Harvest dispatch row trimmed. (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-10-02 15:19

• Bug fixes: S04 SIGPIPE, S07 regex false positives, S08 backtick stripping, S12 fence count, find

  true for missing scripts/ dir (skills/skill-eval/scripts/skill-check.sh)

2026-10-02 15:06

• Removed R4 (skill-accuracy) — skill evaluation moved to skill-eval. agentfs-eval now focuses on workspace health only. (skills/agentfs-eval/SKILL.md)
• New skill: skill-eval v1.0.0 — deterministic checks S01-S22 (P1-P7), LLM rubrics S23-S26. Refactored out of agentfs-eval R4. (skills/skill-eval/SKILL.md)

2026-10-02 13:35

• Added concrete tool call example to Signal Dispatch Rule 1 Path B: shows knowledgegraphmemory__search_nodes(query:…) for weaker model reliability (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-10-02 13:33

• Added –notes to merge-score-entry call (single recorder for both template + behavioral) (skills/agentfs-eval/scripts/template-eval.sh)
• Removed duplicate merge-score-entry call — template-eval.sh is the single score recorder (skills/agentfs-eval/scripts/template-behavioral.sh)
• Changed to append-only: never overwrites historical score entries. Removed update-in-place logic. (skills/agentfs-eval/scripts/merge-score-entry.sh)

2026-10-02 13:12

• Fixed B3 SIGPIPE crash: replaced ‘git log

  head -50’ with ‘git log –max-count=50’ to avoid pipe break with set -euo pipefail (skills/agentfs-eval/scripts/agentfs-behavior.sh)

2026-10-02 12:44

• Fixed set -e crash: template-check.sh non-zero exit killed template-eval.sh before score extraction. Used

  TEMPLATE_EXIT=$? pattern to survive failures and still record scores.

2026-10-02 12:35

• Fixed A3 EXPECTED_RULES 8→7: Rule 8 (Path Hygiene) was removed in v7.0.0 but check script was not updated (skills/agentfs-eval/scripts/template-check.sh)
• Fixed A3 EXPECTED_RULES 8→7: Rule 8 (Path Hygiene) was removed in v7.0.0 but check script was not updated (skills/agentfs-eval/scripts/template-check.sh)

2026-09-30 14:24

• Fixed context-engineering/index.md: removed blockquote markers from summary that broke GitHub Pages rendering (knowledge/index.md)

2026-09-29 22:36

• README drift fix: 8→7 rules, remove Path Hygiene row, add context-engineering bundle, version refs 7.2.1→7.2.2 (README.md)
• Version bump 7.2.1 → 7.2.2 (skills/agentfs-setup/SKILL.md)
• Checkpoint rule: explicit version-bump step for seed edits (skills/agentfs-setup/scripts/seed-agents-md.sh)
• Category 12: seed-version-bump guard (skills/agentfs-setup/scripts/pre-push-scan.sh)
• Remove Path Hygiene assertions, add Checkpoint seed-edit assertions (skills/agentfs-setup/scripts/test-agents-md-fidelity.sh)

2026-09-29 22:09

• Removed Path Hygiene from violation check — rule removed from AGENTS.md (skills/agentfs-violation-chk/SKILL.md)

2026-09-29 22:02

• Created agentfs-violation-chk skill — full behavioral self-audit for session rule/principle compliance (skills/agentfs-violation-chk/SKILL.md)
• Enhanced Session Canary rule with lightweight self-violation check at re-verification points (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-29 21:46

• Step 6: explicit instruction to render tool output in response message (skills/agentfs-git-push/SKILL.md)
• v7.2.1 — Step 6: explicit instruction to render tool output in response message (skills/agentfs-git-push/SKILL.md)

2026-09-29 21:42

• v7.2.1 — git push gate in Pre-Flight, versioning policy in design-spec (skills/agentfs-setup/SKILL.md)
• v7.2.1 — git push gate in Pre-Flight, versioning policy in design-spec (skills/agentfs-setup/scripts/test-agents-md-fidelity.sh)
• v7.2.1 — git push gate in Pre-Flight, versioning policy in design-spec (skills/agentfs-setup/scripts/seed-agents-md.sh)
• v7.2.1 — git push gate in Pre-Flight, versioning policy in design-spec (skills/agentfs-setup/references/design-spec.md)

2026-09-29 21:27

• v7.2.0 — rules consolidated 10→8, severity tags removed (knowledge/context-engineering/rule-ordering-by-frequency.md)
• v7.2.0 — rules consolidated 10→8, severity tags removed (README.md)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/agentfs-git-push/SKILL.md)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/skill-harvest/SKILL.md)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/okf-bundle-harvest/SKILL.md)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/agentfs-eval/SKILL.md)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/agentfs-eval/scripts/template-check.sh)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/agentfs-setup/SKILL.md)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/agentfs-setup/scripts/pre-push-scan.sh)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/agentfs-setup/scripts/test-agents-md-fidelity.sh)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/agentfs-setup/scripts/seed-agents-md.sh)
• v7.2.0 — rules consolidated 10→8, severity tags removed (skills/agentfs-setup/references/design-spec.md)

2026-09-29 19:05

• New concept in context-engineering bundle (knowledge/context-engineering/rule-ordering-by-frequency.md)
• New concept in context-engineering bundle (knowledge/context-engineering/jit-reminder-recency-pattern.md)
• New concept in context-engineering bundle (knowledge/context-engineering/moim-persistent-instructions-reality.md)
• New concept in context-engineering bundle (knowledge/context-engineering/instruction-override-semantics.md)
• New concept in context-engineering bundle (knowledge/context-engineering/llm-context-layer-model.md)

2026-09-29 17:55

• v7.1.2 — Removed hardcoded load_skill() syntax from harvest action for graceful degradation (plugins/signal-dispatch/patterns.txt)
• v7.1.2 — Added load_skill failure handling to Signal Dispatch Path B step 3 (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-29 17:07

• v7.1.1 — Renamed Post-Write Hook → Post-Write rule reference (skills/agentfs-setup/scripts/merge-changelog-entry.sh)
• v7.1.1 — Renamed Post-Write Hook → Post-Write rule reference (skills/agentfs-setup/scripts/post-edit.sh)
• v7.1.1 — Renamed Post-Write Hook → Post-Write rule reference (skills/okf-bundle-gen/SKILL.md)

2026-09-29 17:00

• Update guardrail reference to v7 rule name (skills/goose-agentfs-setup/scripts/setup.sh)
• Update guardrail reference to v7 rule name (skills/goose-agentfs-setup/references/memory-routing.md)
• Update post-edit scope detection (skills/agentfs-setup/scripts/post-edit.sh)
• Fix merge-changelog-entry formatting (skills/agentfs-setup/scripts/merge-changelog-entry.sh)

2026-09-29 16:59

• Version bump to 7.1.1 (skills/agentfs-setup/SKILL.md)
• Add Category 10: post-write coverage check — L3→L2 graduation for Rule 3 enforcement (skills/agentfs-setup/scripts/pre-push-scan.sh)

2026-09-29 16:26

• Version bump to 3.2.1 — guardrail ref text update (skills/okf-bundle-gen/SKILL.md)
• Version bump to 1.5.1 — guardrail ref text update (skills/goose-agentfs-setup/SKILL.md)

2026-09-29 16:25

• Updated Structural Rules section from v6.0.9 table format to v7.1.0 heading format (README.md)

2026-09-29 16:16

• v7.1.0 — Rule priority reorder: Pre-Flight and Post-Write to HIGH, Session/Conflict to NORMAL (skills/agentfs-setup/SKILL.md)
• v7.1.0 — Updated fidelity assertions for new rule ordering and priority tags (skills/agentfs-setup/scripts/test-agents-md-fidelity.sh)
• v7.1.0 — Reordered rules by frequency: Pre-Flight and Post-Write promoted to HIGH (position 2-3), Session Start/Canary/Conflict Resolution moved to NORMAL (position 4-6). Post-Write Hook renamed to Post-Write. (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-29 14:47

• Fix: never overwrite valid N/M behavioral score with dash (skills/agentfs-eval/scripts/merge-score-entry.sh)
• Fix: preserve behavioral score on skip instead of overwriting with dash (skills/agentfs-eval/scripts/template-eval.sh)

2026-09-29 14:44

• Version bump to 1.7.0 for v7 compat (skills/agentfs-eval/SKILL.md)
• Updated A3 and A8 assertions for v7.0.0 heading-format rules (skills/agentfs-eval/scripts/template-check.sh)
• Updated A3 and A8 assertions for v7.0.0 heading-format rules (skills/agentfs-eval/scripts/template-check.sh)

2026-09-29 14:23

• v7.0.0 — Major version bump for context landscape optimization (skills/agentfs-setup/SKILL.md)
• v7.0.0 — Rewritten for v7 structure (prose rules, named headers, dispatch hierarchy) (skills/agentfs-setup/scripts/test-agents-md-fidelity.sh)
• v7.0.0 — 8 principles (removed signal dispatch, revised ledger→data integrity), reordered by operational impact (skills/agentfs-setup/scripts/seed-soul.sh)
• v7.0.0 — Prose rules with priority tags, generated Signal Dispatch table from patterns.txt, named rule references (skills/agentfs-setup/scripts/seed-agents-md.sh)
• v2.0.0 — Plugin version bump, updated description (plugins/signal-dispatch/plugin.json)
• v2.0.0 — Named rule references (Signal Dispatch rule) replacing numbered (plugins/signal-dispatch/scripts/enforce-hey.sh)
• v2.0.0 — Reads patterns.txt dynamically instead of hardcoded case block (plugins/signal-dispatch/scripts/detect-hey.sh)
• v2.0.0 — Single source of truth for Signal Dispatch table-matched patterns (plugins/signal-dispatch/patterns.txt)

2026-09-28 14:45

• Rule 2 prose polish: -16 words, removed redundant clauses, tightened fallback chain (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-28 14:27

• Added Scope Definitions to PROJECT template heredoc — fixes sync/verify misalignment (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-28 14:02

• Tier 2 fixes: Rule 2 search_nodes fallback, Rules 4/7 script-not-found guards, Rule 10 adaptive canary cadence (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-28 13:48

• Rule 4: added template ownership guard — edit seed-agents-md.sh not project copy for template-owned content

2026-09-28 13:43

• Tier 1 clarity: added keyword definition+example to Rule 2, added multi-step threshold to Rule 8 in project template (skills/agentfs-setup/scripts/seed-agents-md.sh)
• Tier 1 clarity: added keyword definition+example to Rule 2, added multi-step threshold to Rule 8 in project template (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-28 11:33

• Skill Check overhaul v2.0.0: resolve-target.sh, spec/tests, error contract (skills/agentfs-git-push/SKILL.md)

2026-09-28 11:05

• Rule 2: added search_nodes query construction constraint (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-28 10:51

• Vendor-neutral SOUL.md fallback; moved context lookup fallback to Quick Orientation (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-28 10:11

• v3.2.0: Added safety rules, updated known limitations with manage_launcher_entry opt-out and wrapper path mismatch (skills/hermes-desktop-fixes/SKILL.md)

2026-09-28 09:48

• v3.2.0: Added Bug #6 (GPU process crash on Intel Arc + Wayland), fixed desktop entry overwrite (skills/hermes-desktop-fixes/SKILL.md)

2026-09-28 00:08

• Table-match discrimination: only set enforcement flag for unmatched hey signals (skill dispatch fallthrough). Table-matched signals (remember, forget, prefer, always, never) skip enforcement entirely. (plugins/signal-dispatch/scripts/detect-hey.sh)

2026-09-28 00:00

• Remove load_skill/read/write/edit from allow-list — only search_nodes passes through, enforcing Rule 2 mandatory first tool call (plugins/signal-dispatch/scripts/enforce-hey.sh)

2026-09-27 23:48

• JIT reminder now specifies mandatory search_nodes: ‘first tool call MUST be search_nodes’ (plugins/signal-dispatch/scripts/enforce-hey.sh)

2026-09-27 23:47

• Tighten behavioral tests: expected first tool is search_nodes only (not search_nodes OR load_skill), load_skill moved to anti-pattern list (skills/agentfs-eval/scripts/template-behavioral.sh)
• Rule 2: mandatory search_nodes first — ‘first tool call MUST be search_nodes’, closes smart-model shortcut (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-27 23:35

• Fix B2 expected tool order: search_nodes before load_skill, matching Rule 2 fallthrough sequence (skills/agentfs-eval/scripts/template-behavioral.sh)
• Fix B2 expected tool order: search_nodes before load_skill, matching Rule 2 fallthrough sequence (skills/agentfs-eval/scripts/template-behavioral.sh)

2026-09-27 23:22

• v6.0.1: Rule 2 now includes search_nodes fallthrough. Signal Dispatch table has specific patterns only. (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-27 23:16

• v1.5.0: JIT reminder includes search_nodes fallthrough as catch-all. Allowlist expanded with read/write/edit for memory routes. Signal Dispatch table no longer has overlapping default rule. (plugins/signal-dispatch/scripts/enforce-hey.sh)

2026-09-27 23:01

• v1.1.0: Fixed denied-call skip — check toolResult.error field, not just result (skills/agentfs-eval/scripts/template-behavioral.sh)

2026-09-27 22:44

• v6.0.1: All test inputs use hey prefix for signal-dispatch plugin enforcement. Tool extraction skips denied-by-policy tool calls, scores first successful call. (skills/agentfs-eval/scripts/template-behavioral.sh)

2026-09-27 22:31

• v6.0.1: Added KI-2 — session poisoning from PreToolUse block reasons, compaction as recovery pattern (skills/agentfs-setup/references/design-spec.md)

2026-09-27 22:25

• v1.4.0: Block reason references AGENTS.md dispatch table instead of embedding it — single source of truth (plugins/signal-dispatch/scripts/enforce-hey.sh)

2026-09-27 22:23

• v1.3.0: Block reason includes full dispatch table inline with user keywords — covers all signal routes, not just skill dispatch (plugins/signal-dispatch/scripts/enforce-hey.sh)

2026-09-27 22:19

• v1.2.0: Rewrote block reason — affirmative framing, no alarm language, no prohibition list, reduced session poisoning risk (plugins/signal-dispatch/scripts/enforce-hey.sh)

2026-09-27 22:11

• v6.0.1: Patch — SOUL.md sync via seed-soul.sh, plugin version detection in post-write, Rule 10 unconditional re-read (skills/agentfs-setup/SKILL.md)

2026-09-27 22:02

• v6.0.0: Added plugin directory detection — warns when plugin files modified without –version (skills/agentfs-setup/scripts/post-write.sh)
• test change (skills/agentfs-setup/scripts/post-write.sh)
• test change (plugins/signal-dispatch/scripts/enforce-hey.sh)
• test change (skills/agentfs-setup/scripts/post-write.sh)

2026-09-27 21:58

• v6.0.0: Rule 10 — unconditional re-read every ~5 turns instead of conditional on canary loss (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-27 21:25

• v1.1.0: Version bump and updated description (plugins/signal-dispatch/plugin.json)
• v1.1.0: Updated comment to reflect keyword file cleanup (plugins/signal-dispatch/scripts/cleanup-stale.sh)
• v1.1.0: Now saves extracted keywords to flag file for use in block reason (plugins/signal-dispatch/scripts/detect-hey.sh)
• v1.1.0: Block reason now includes step-by-step dispatch instructions with exact tool name, extracted keywords, KGM fallback to Skills section, and explicit tool prohibition (plugins/signal-dispatch/scripts/enforce-hey.sh)

2026-09-27 21:08

• v6.0.0: Added fingerprint-based filtering to strip template-covered content during migration (skills/agentfs-setup/scripts/seed-soul.sh)

2026-09-27 21:03

• v6.0.0: SOUL.md sync now runs in both version-change and up-to-date paths (skills/agentfs-setup/scripts/sync-agents-md.sh)

2026-09-27 21:02

• v6.0.0: Updated to use seed-soul.sh for template content, custom content goes below PROJECT-OWNED marker (skills/agentfs-setup/scripts/author-soul.sh)
• v6.0.0: Added SOUL.md sync via seed-soul.sh — detects old format, migrates, and syncs template Principles (skills/agentfs-setup/scripts/sync-agents-md.sh)
• v6.0.0: New script — generates and syncs template-owned SOUL.md with ownership markers (skills/agentfs-setup/scripts/seed-soul.sh)

2026-09-27 20:33

• v3.1.1: Added –check pre-flight to launcher to avoid unconditional rebuilds when already up to date (skills/hermes-desktop-fixes/SKILL.md)

2026-09-27 20:18

• Bump agentfs-eval version to 1.6.0 (skills/agentfs-eval/SKILL.md)
• Update template-check.sh for agentfs v6.0.0: A2 no Scopes section, A3 expects 10 rules, A5 checks template-wide
• Update template-check.sh for agentfs v6.0.0: A2 no Scopes section, A3 expects 10 rules, A5 checks template-wide

2026-09-27 20:10

• v3.1.0: Fixed orphaned skip-worktree cleanup in revert script, added orphan detection to health check (skills/hermes-desktop-fixes/SKILL.md)

2026-09-27 20:02

• v1.5.0: Added Plugin Management section documenting –plugin-check and –plugin-list (skills/goose-agentfs-setup/SKILL.md)
• v1.5.0: Added –plugin-check and –plugin-list commands for signal-dispatch plugin management (skills/goose-agentfs-setup/scripts/setup.sh)
• v6.0.0: Added plugins/ and scripts/ directory checks to USER scope verification (skills/agentfs-setup/scripts/verify-setup.sh)
• v6.0.0: Added plugins/ and scripts/ directories to USER scope scaffolding and index.md template (skills/agentfs-setup/scripts/scaffold-dotagents.sh)

2026-09-25 23:36

• v6.0.0: Synced with design spec — updated rules table (17→10), added plugins/ and scripts/ to USER scope structure, updated signal dispatch description (README.md)
• v6.0.0: Updated PROJECT scope template — 10 rules, Signal Dispatch table, scope tables removed (skills/agentfs-setup/scripts/seed-agents-md.sh)
• v6.0.0: Bumped version to 6.0.0 (skills/agentfs-setup/SKILL.md)
• v6.0.0: Updated SOUL/AGENTS layer descriptions, signal dispatch architecture, scope tables (added plugins/ scripts/), KI-1 mitigated status, added v6.0.0 changelog entry (skills/agentfs-setup/references/design-spec.md)

2026-09-25 13:34

• v3.0: Updated recovery script — new desktop entry, launcher watcher, retired stale patches (skills/hermes-desktop-fixes/recover.sh)
• v3.0: Rewrote Bug #4 (Wayland app-id), retired Bugs #3/#5, updated architecture (skills/hermes-desktop-fixes/SKILL.md)

2026-09-25 10:05

• Added orphaned namespace Known Issue entry, bumped to v8.15.0 (skills/skupper-model-provider/SKILL.md)
• Added orphaned namespace detection section (skills/skupper-model-provider/scripts/status.sh)

2026-09-25 01:46

• Update rule count 18→17, A5 checks Rule 7 for search_nodes+load_skill (skills/agentfs-eval/scripts/template-check.sh)
• v5.16.0 — Remove Tool Priority section and Rule 8; generalize Rule 7 to ‘hey <keywords>’ dispatch; renumber rules 8-17; -190 tokens (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-25 00:50

• v5.15.0 — Generic Rule 7: ‘hey <keywords>’ → search_nodes dispatch prefix for reliable skill invocation on weaker models (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-25 00:22

• Add Known Issues section: KI-1 command-shaped signal phrase vs shell tool dispatch ambiguity (skills/agentfs-setup/references/design-spec.md)
• B1 test: use ‘check headroom status’ (no CLI-name collision); document command-shape ambiguity (skills/agentfs-eval/scripts/template-behavioral.sh)

2026-09-24 23:26

• v5.14.0 — Add Tool Priority section, strengthen Rule 8 anti-shell clause for weaker models (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-24 23:10

• v1.1.0 — Add scope boundaries, hard gate on stopped cluster, decision-point table per skill design principles (skills/crc-status/SKILL.md)

2026-09-24 22:50

• Resilient session cleanup: trap EXIT handler, per-session cleanup fn, orphan sweep on updated_at, DB timeout=5s (skills/agentfs-eval/scripts/template-behavioral.sh)

2026-09-24 22:24

• Fix test expectations: accept search_nodes as valid first tool (Rule 8 dispatch), add dedup guard, exit 0 always, multi-tool expected support (skills/agentfs-eval/scripts/template-behavioral.sh)

2026-09-24 22:13

• Add 10-min dedup guard + exit 0 + anti-loop warning — prevents weaker models from re-running behavioral test (skills/agentfs-eval/scripts/template-behavioral.sh)
• Exit 0 always + anti-loop warning — prevents weaker models from re-running eval in a loop (skills/agentfs-eval/scripts/template-eval.sh)
• Add 10-min dedup guard + exit 0 + anti-loop warning — prevents weaker models from re-running behavioral test in a loop (skills/agentfs-eval/scripts/template-behavioral.sh)
• Exit 0 always + anti-loop warning — prevents weaker models from re-running eval in a loop (skills/agentfs-eval/scripts/template-eval.sh)

2026-09-24 21:21

• New knowledge bundle: KGM-Based Skill Dispatch (5 concepts) (knowledge/agentfs-kgm-skill-dispatch/architecture.md)

2026-09-24 21:14

• Update assertion docs: A2 Scopes before Rules, A4 context lookup fallback, A5 search_nodes + load_skill, remove stale A6 (skills/agentfs-eval/SKILL.md)

2026-09-24 21:09

• Add durable fix principle to default SOUL.md template (skills/agentfs-setup/scripts/author-soul.sh)

2026-09-24 21:06

• Add ledger integrity principle to default SOUL.md template (skills/agentfs-setup/scripts/author-soul.sh)

2026-09-24 20:46

• Document AgentSkill entity type, staleness guard, anti-pattern moved from AGENTS.md (skills/goose-kgm/SKILL.md)
• Delete Discovery Tiers section, add context lookup fallback note, Rule 8 uses search_nodes (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-24 19:25

• Version bump 1.5.2 (skills/agentfs-eval/SKILL.md)

2026-09-24 19:24

• SQLite-based test session cleanup instead of goose session remove (skills/agentfs-eval/scripts/template-behavioral.sh)
• Rule 8: KGM-based skill dispatch via search_nodes, fallback to # Skills (skills/agentfs-setup/scripts/seed-agents-md.sh)
• Separate knowledge/skills KGM indexes, merge into combined .kgm-index.jsonl (skills/goose-kgm/scripts/reindex-kgm.sh)

2026-09-24 19:11

• setup-kgm.sh: symlink server-memory default memory.jsonl to managed .kgm-index.jsonl (workaround for envs not passed to MCP process)

2026-09-24 17:43

• Rule 8: add anti-pattern clause and concrete example (audit context → agentfs-ctx-chk → load_skill) (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-24 17:25

• Fix double-run bug: capture output from single run instead of running checks twice (skills/agentfs-eval/scripts/template-eval.sh)

2026-09-24 17:17

• template-eval.sh reads AGENT_SESSION_ID to detect provider/model from current session. Zero args required. Fallback chain: session metadata → config.yaml (skills/agentfs-eval/SKILL.md)

2026-09-24 16:51

• template-eval.sh accepts –provider/–model args, falls back to config.yaml. Quick Dispatch passes session provider/model. (skills/agentfs-eval/SKILL.md)

2026-09-24 16:09

• Test without version (skills/agentfs-eval/SKILL.md)
• post-write.sh now fails (exit 1) instead of warning when skill file modified without –version

2026-09-24 16:05

• Version bump 3.6.0 (skills/skill-gen/SKILL.md)
• Version bump 1.3.0 (skills/agentfs-eval/SKILL.md)

2026-09-24 16:01

• Add corollary: lower cognitive bar for model callers — wrapper scripts, auto-detect context, avoid similar names (skills/skill-gen/SKILL.md)
• Rename agents-md-* to template-*, add template-eval.sh zero-arg wrapper, simplify Quick Dispatch (skills/agentfs-eval/SKILL.md)

2026-09-24 15:43

• Fix arg parsing bug: TARGET consumed first positional before while loop (skills/agentfs-eval/scripts/agents-md-behavioral.sh)

2026-09-24 15:34

• Add merge-score-entry.sh: deterministic score sheet management (insert/update/dedup)

2026-09-24 14:54

• v5.11.1: Rule 1 agent-agnostic, USER.md read at session start, fix USER.md orphan (skills/agentfs-setup/SKILL.md)

2026-09-24 14:48

• Rule 1: remove vendor-specific filenames, add USER.md read at session start (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-24 13:51

• Auto-append behavioral results to score sheet (skills/agentfs-eval/scripts/agents-md-behavioral.sh)

2026-09-24 13:42

• Single score table with model column, reverse chronological (skills/agentfs-eval/references/template-scores.md)

2026-09-24 13:39

• Split score sheet into deterministic + behavioral tables, reverse chronological, model column for behavioral (skills/agentfs-eval/references/template-scores.md)

2026-09-24 13:31

• Move template-scores.md from agentfs-setup to agentfs-eval, update all references (skills/agentfs-eval/SKILL.md)

2026-09-24 13:27

• Add agentfs score signal phrase (skills/agentfs-eval/SKILL.md)

2026-09-24 13:25

• Add agentfs status signal, add quick score dispatch row (skills/agentfs-eval/SKILL.md)

2026-09-24 13:21

• Simplify signals to eval agentfs/agentfs health/maturity check. Fold AGENTS.md check into full eval flow (skills/agentfs-eval/SKILL.md)

2026-09-24 13:16

• Add score/quality signal phrases and Quick Dispatch table mapping user requests to scripts (skills/agentfs-eval/SKILL.md)

2026-09-24 13:14

• Add template-scores.md: persistent quality score history for AGENTS.md template versions

2026-09-24 13:08

• Document agents-md-check.sh and agents-md-behavioral.sh in SKILL.md body and supporting files list

2026-09-24 13:03

• Add agents-md-behavioral.sh: LLM-based behavioral test for skill discovery compliance
• Add agents-md-check.sh: deterministic AGENTS.md template quality assertions (A1-A9)

2026-09-24 12:39

• Bump version to 5.11.0 (skills/agentfs-setup/SKILL.md)
• v5.11.0: Reorder sections (Scopes→Discovery→Rules→Orientation), merge Discovery tier table with Action column, remove redundant CRITICAL preamble and WHY prose from rules, add anti-pattern callout, trim ~22% tokens (skills/agentfs-setup/scripts/seed-agents-md.sh)

2026-09-24 11:02

• Rewrote Discovery Tiers and Rule 8 for small-model compatibility: explicit skill lookup procedure, direct load_skill references, anti-pattern callouts (skills/agentfs-setup/scripts/seed-agents-md.sh)

r# Directory Update Log

2026-09-18 12:12

• Added ocp-containerfile knowledge bundle to README knowledge list — fixes D3 drift (README.md)

2026-09-18 11:54

• v8.14.0: signal cleanup (skills/skupper-model-provider/SKILL.md)
• v7.6.0: new script for setting default profile (skills/hosted-model-ctl/scripts/set-default.sh)
• v7.6.0: persistent default profile support (skills/hosted-model-ctl/scripts/common.sh)
• v7.6.0: signal cleanup + set default profile (skills/hosted-model-ctl/SKILL.md)

2026-09-18 11:26

• v8.13.0: add S11/T11 specs, update operations docs (skills/skupper-model-provider/SKILL.md)
• v8.13.0: add localhost target fast path (skills/skupper-model-provider/scripts/down.sh)
• v8.13.0: add localhost target fast path (skills/skupper-model-provider/scripts/up.sh)

2026-09-18 10:19

• v8.12.0: add linger Known Issue entry (skills/skupper-model-provider/SKILL.md)
• v8.12.0: call enable_linger in Phase 5 (skills/skupper-model-provider/scripts/setup.sh)
• v8.12.0: add enable_linger() function (skills/skupper-model-provider/scripts/common.sh)

2026-09-16 12:11

• Added D9/P8 Coverage Gaps dimension — non-blocking signal for unmentioned content (skills/agentfs-readme-audit/SKILL.md)

2026-09-16 11:49

• Log entries now include relative file path for traceability in batch invocations (skills/agentfs-setup/scripts/post-write.sh)
• Test log entry (knowledge/ocp-containerfile/base-images-ubi.md)

2026-09-16 11:46

• Added regen_knowledge_index to post-edit checks — regenerates knowledge/index.md using rebuild-index.sh (reverse-chronological ordering) before auditing

2026-09-16 11:37

• OCP Containerfile skill v1.0.0 (skills/ocp-containerfile/SKILL.md)
• OCP Containerfile skill v1.0.0 (skills/ocp-containerfile/scripts/audit.sh)
• OCP Containerfile skill v1.0.0 (skills/ocp-containerfile/scripts/generate.sh)
• OCP Containerfile skill v1.0.0 (skills/ocp-containerfile/scripts/verify.sh)
• OCP Containerfile skill v1.0.0 (skills/ocp-containerfile/references/containerfile-templates.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/index.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/uid-model-restricted-v2.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/uid-model-user-namespaces.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/scc-and-security-context.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/file-permissions-gid0.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/base-images-ubi.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/image-metadata-labels.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/layer-optimization.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/runtime-constraints.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/certification-requirements.md)
• OCP Containerfile knowledge bundle (knowledge/ocp-containerfile/s2i-compatibility.md)

2026-09-16 09:12

• Discovery Tiers template updated: 2a/2b/2c split, KGM-preferred, staleness guard, freshness check
• v5.10.0 — Discovery Tiers restructured: 2a/2b/2c split, KGM-preferred knowledge discovery, staleness guard

2026-09-16 08:57

• Now reports config entry + session note + detailed JSONL breakdown (bundles/concepts/relations)
• Deprecated: prints session-scoped instructions instead of editing config.yaml
• Deprecated: prints session-scoped instructions instead of editing config.yaml
• v1.1.0 — session-scoped enable/disable, status session check, reindex signal, sync model docs

2026-09-11 11:03

• fix log coverage check to find log.md in both USER scope (repo root) and PROJECT scope (.agents/)

2026-09-11 10:56

• Fix template version reference 5.6.0 → 5.9.0 in README.md

2026-09-11 10:52

• v2.0.0 — Add PROJECT scope support: scope resolution, PROJECT dimensions P1–P7, dual report format

2026-09-10 21:32

• Added Pre-Flight Checklist as primary mitigation; restructured effectiveness section with preventive/detective/corrective layers

2026-09-10 21:31

• Added Rule 18 (Pre-Flight Checklist) to AGENTS.md template

2026-09-10 18:05

• Created knowledge bundle llm-behavioral-properties v1.0: 5 concepts documenting LLM behavioral properties and AgentFS mitigations. OKF v0.1 conformant (13/13 checks pass).

2026-09-10 17:39

• Updated agentfs-setup v5.8.1: post-edit.sh enhanced with log drift detection — compares skill file mtimes against latest log entry, warns on unlogged modifications. Catches missed Rule 13 obligations during iterative debugging.

2026-09-10 13:59

• Moved fedora-openshell from USER to PROJECT scope (./.agents/skills/) in agentbox. USER index: 58 skills (was 59).

2026-09-10 13:56

• Reversed default skill scope from USER to PROJECT in skill-gen v3.5.0, agentfs-setup v5.8.0, and AGENTS.md Rule 11. Scope governs discovery/context loading, not operational target. PROJECT avoids signal phrase bloat; promote to USER when cross-project demand is proven.

2026-09-10 13:13

• Created fedora-openshell v1.0.0: Install, start, stop, status, upgrade, teardown for NVIDIA OpenShell on Fedora. Includes verify.sh (S1–S9), privilege gate pattern (exit 3) for sudo operations. All 10 tests pass.

2026-09-10 12:59

• Updated skill-gen v3.4.0: Added Principle 6 (Context Economy) with soft 300-line threshold, two-step audit, references/ conventions. Added Principle 7 (Error Contract) with semantic exit codes 0-3, privilege gate pattern, idempotency/troubleshooting/recovery obligations. Made Skill Check mandatory in Post-Creation Checklist. Updated skill-schema.md v2.3.0 with directory structure and exit code conventions.

2026-09-09 22:25

• Fixed escaped backticks in Rule 13 template row (\)

2026-09-09 22:16

• agentfs-setup v5.7.0: Added post-write.sh orchestrator script. Redesigned Rule 13 — trigger moved from ‘after each write’ to ‘before sending response’ (end-of-turn audit); action collapsed from 4 manual steps to single post-write.sh call. Updated seed-agents-md.sh template, filesystem-integrity.md, and project AGENTS.md.

2026-09-09 22:10

• Fixed 7 log entries missing bullet prefix (plain text and [OVERRIDE] entries) — added ‘- ‘ prefix to match canonical format.

2026-09-09 22:08

• Fixed 42 malformatted log entries in ~/.agents/log.md: inserted missing blank lines between ## headings and bullet content to match merge-log-entry.sh canonical format.

2026-09-09 21:21

• Updated base_url from localhost to host.containers.internal in three provider skills: goose-litellm-provider v1.7.0 (restore.sh, verify.sh, SKILL.md); goose-skupper-provider v5.4.0 (setup.sh, SKILL.md); goose-headroom-provider v2.1.0 (SKILL.md). Enables dual-resolution: host resolves to 127.0.0.1, containers resolve to 169.254.1.2.

2026-09-09 14:32

• Fixed skupper-model-provider T6 test: localhost→127.0.0.1 for rootless Podman pasta IPv4-only listener

2026-09-09 14:25

• Updated skupper-model-provider v8.11.0: added LOCAL_LISTENER_HOST for configurable listener bind address, updated topology.env and topology.env.example, patched live Listener YAMLs to 0.0.0.0

2026-09-09 14:02

• Updated litellm-vertex-ai-proxy v3.1.0: added –host flag to setup.sh, verify.sh auto-detects bind address, Gotcha #8 for container access

2026-09-09 09:36

• README.md: clarified title to ‘AgentFS (agent-filesystem)’

2026-09-08 15:33

• Updated agentfs-setup v5.5.0→v5.6.0: added Anti-Action-Sycophancy (Rule 17) to seed-agents-md.sh, anti-assumption identity principle to author-soul.sh, updated SKILL.md rule inventory (12→13). Root cause: Agentbox weather session — model fabricated NYC coordinates instead of asking for location.

2026-09-04 16:55

• Log format cleanup: normalized all entries in ~/.agents/log.md (138 plain-text → bullet, 9 star-bullet → dash-bullet) and ~/.agents/knowledge/log.md (20 star-bullet → dash-bullet, 1 plain-text → bullet)

2026-09-04 15:01

• skill-gen v3.3.0: Added Script Comment SoC principle and cross-skill execution priority order. skill-schema v2.2.0: Added Script Comment Rules section.
• agentfs-setup v5.5.0: Added Script I/O Conventions section (bullet-prefixed log entries). Fixed merge-log-entry.sh comment (SoC). Added cross-skill execution rule to filesystem-integrity.md.

2026-09-04 14:13

• Updated README.md: expanded skill categories (added Diagramming, Shell Tooling, Spec Authoring, Data Services), updated structural rules from 12 to 16 with v5.4.0 Type/Stimulus/Action format, bumped template version ref 5.3.0 → 5.4.0

2026-09-04 14:01

• agentfs-setup v5.4.0: reordered dispatch group (signals before fallback); removed ~/.agents/scripts/ symlinks, paths point directly to skill scripts

2026-09-04 13:13

• AgentFS template v5.4.0: redesigned Rules system — merged Signal Routing + Rules into single typed table; added Discovery Tiers section; added enforcement preamble; phase-grouped rules; inlined script paths; created ~/.agents/scripts/ symlinks
• Updated fedora-obsidian-fix v1.1.0 and fedora-desktop-wmclass-fix v1.1.0 — fixed duplicate GNOME Dash icon caused by competing .desktop files and inconsistent Electron WM_CLASS under X11 mode

2026-09-01 18:58

• hosted-model-ctl v7.5.3: Decision — rhtevan-work g3b-16k stays on Granite 4.1. Rationale: llama.cpp has no reasoning parser (4.2 tags leak), 4.2 GGUF +145 MB (headroom drops to 0.14 GiB on 4 GB GPU), 3B reasoning quality marginal. Reverted GGUF preference to 4.1-first. Granite 4.2 upgrade is rhel-ai (vLLM) only.

2026-09-01 18:48

• hosted-model-ctl v7.5.2: Added Granite 4.2 model rows to model-landscape.md. Added Section 13 (4.2 benchmark results) to benchmark-report.md. Updated timestamps. Fixed 4.1 tool-call parser label from granite to hermes.

2026-09-01 18:36

• hosted-model-ctl v7.5.1: Granite 4.2 upgrade wrap-up. Updated all reference docs (deployment-profiles, memory-budget, model-landscape). Added reasoning overhead gotcha. Benchmark results: 7/7 pass (requires max_tokens>=8192 for code gen). Speed: ~50 tok/s total, ~47 content on long gen, 70-80% reasoning overhead on short responses. Quality improved on multi-step reasoning. Flipped GGUF preference to 4.2.

2026-09-01 18:07

• crc-ols v3.3.0: Updated skupper-model-rhel OLS provider model from granite-4.1-8b-fp8 to granite-4.2-8b-fp8 (Granite 4.2 upgrade dependency).

2026-09-01 17:39

• hosted-model-ctl v7.5.0: Upgraded rhel-ai profiles (g8b-fp8-spec-128k, g8b-spec-128k) from Granite 4.1 to 4.2. Parser changes: hermes→qwen3_coder (tool), added nemotron_v3 (reasoning). Draft models upgraded to 4.2-3b-fp8. All 4 tests pass. Speed: ~44 tok/s total, ~21 tok/s content-visible (50% reasoning overhead vs 4.1 baseline of 58-79 tok/s).

2026-08-31 20:51

• README.md: fixed D5 (Rule #1 KGM triage wording) and D6 (template version 5.0.0 → 5.3.0)

2026-08-31 20:45

• agentfs-setup v5.3.0: Rule #1 KGM triage — MUST read Source files (max 3) before answering

2026-08-31 20:32

• agentfs-setup v5.2.0: added post-sync Rule re-read requirement (KGM bootstrap fix)

2026-08-31 17:29

• agentfs-git-push v1.1.0: Add Step 0 (Resolve Target) — default to ~/.agents/ for bare ‘hey git’

2026-08-31 16:50

• v5.1.0: Rule #1 KGM-aware dispatch, LITE template fix, test-agents-md-fidelity.sh rewrite, KGM section in README.md, CHANGELOG format fixes (agentfs-setup, goose-kgm, agentfs-git-push)

2026-08-31 15:59

• A11: Archived v5 strategy/action-plan/migration-map to agentfs-setup/references/, bumped agentfs-setup to 5.0.0, goose-kgm to 1.0.1. AgentFS v5 rollout complete (agentfs-v5 A11)

2026-08-31 15:58

• A10: Updated README.md and agentfs-setup SKILL.md — v5 flat rule table (12 rules), removed Gate/Rule/Habit taxonomy, updated all guardrail references to ‘rule’ (agentfs-v5 A10)

2026-08-31 15:55

• A9: Wired conditional KGM reindex into post-edit.sh, added KGM integration section to agentfs-setup SKILL.md (agentfs-v5 A9)
• A8: Created KGM entity schema reference doc — KnowledgeBundle + KnowledgeConcept entity types, contains relation, observation conventions (agentfs-v5 A8)

2026-08-31 15:51

• A5: Rewrote seed-agents-md.sh PROJECT template — v5 flat table (12 rules, 6 signal rows, no Quick Reference, no inline guardrail prose), 69% token reduction (agentfs-v5 A5)

2026-08-31 15:48

• Created goose-kgm skill v1.0.0 — KGM lifecycle management with setup/teardown/enable/disable/status/reindex, OKF-primary design (agentfs-v5 A4)

2026-08-31 15:43

• A3: Created v5-migration-map.md — full content coverage audit, 12 rules mapped, zero true content loss (agentfs-v5 A3)

2026-08-31 15:42

• A2: Added references/filesystem-integrity.md to agentfs-setup — delegation table, scope logging rules, checkpoint protocol (agentfs-v5 A2)

2026-08-31 15:41

• Created agentfs-git-push skill v1.0.0 — absorbs Git Push Safety workflow from AGENTS.md Guardrail #10 (S5 externalization, agentfs-v5 A1)

2026-08-31 14:37

• Renamed skill ‘archify’ → ‘agentfs-archify’ (AgentFS collection prefix). Renamed skill ‘fuseki’ → ‘agentflow-fuseki’ (AgentFLOW collection prefix). Updated name: fields, cross-references in goose-nextai-drawio/SKILL.md, .agents/memories/USER.md. Signal phrases unchanged. Index rebuilt (56 skills).

2026-08-31 10:03

• dsh-setup v1.3.1 — added version check to status.sh; compares installed vs npm registry latest, shows upgrade recommendation

2026-08-31 09:59

• dsh-setup v1.3.0 — replaced ‘dsh launcher’/’dsh desktop’ signals with ‘dsh status’; added scripts/status.sh for backend state + desktop path query

2026-08-29 13:14

• Added van-topology.gif to skupper-model-provider/docs/ and added image link to SKILL.md Architecture section

2026-08-29 12:41

• Updated skupper-model-provider diagrams: VAN Topology page redesigned — removed RouterAccess component, removed Controllers for consistency, added 3 network zone bands (LAN/AWS/HOST), renamed CRC site to OpenShift Local, added flow animations, cleaned layout alignment. Operational Lifecycle and Startup Sequence pages unchanged.

2026-08-28 20:43

• goose-nextai-drawio v1.1.2: removed undocumented cwd/bundled null fields from SKILL.md, manage.sh, and both stdio entries in config.yaml (nextaidrawio + linuxmcpserver)

2026-08-28 20:41

• goose-nextai-drawio v1.1.1: fixed extension name ‘Next AI Draw.io’ → ‘Next AI Drawio’ — period in name broke Goose Desktop toggle. Updated manage.sh, SKILL.md, config.yaml.

2026-08-28 15:15

• goose-nextai-drawio v1.1.0 — added Specification (S1-S10) and Tests (T1-T9) per Principle 4 compliance

2026-08-28 15:03

• Created skill goose-nextai-drawio v1.0.0 — setup/teardown/upgrade/status for next-ai-draw-io MCP server as Goose stdio extension

2026-08-27 19:33

• crc-ols v3.2.0: Added Skupper provider liveness check in list.sh — detects model name mismatches and unreachable backends by probing /v1/models via the CRC router pod. Cross-skill validation between crc-ols and hosted-model-ctl without tight coupling.

2026-08-27 19:22

• crc-ols v3.1.0: Added model validation in switch-provider.sh — prevents silent runtime 404 when defaultModel doesn’t exist under defaultProvider. Added ‘switch ols model’ signal. Added gotcha for CRD-level model name mismatch.

2026-08-27 19:12

• crc-ols v3.0.0: Signal refactor per skill-gen principles. 5 unambiguous signals (install/list/add/switch/remove ols), zero cross-skill collisions. Moved list/switch/remove operations from prose to deterministic scripts. Added Gotchas (6 known issues), Specification (6 items), Tests (4 cases). Renamed switch-default to switch-provider to reflect provider+model pair semantics.

2026-08-27 12:59

• skupper-model-provider v8.8.0: Fixed delegation instructions — hosted-model-ctl takes profile names (not host names) due to multiple profiles per host with mutual exclusion. Added start/stop on PROVIDER signal rows with profile resolution order. Added 3 known issues: OLS duplicate volume mount with shared secrets, llama.cpp grammar incompatibility with OLS tool-use, provider naming convention.

2026-08-27 12:14

• skupper-model-provider v8.7.1: Documented combined status report format in SKILL.md — 5-section layout covering VAN infra, CRC links, localhost listeners, CRC listeners, and model containers

2026-08-27 11:39

• Updated README.md: expanded OpenShift/CRC category to include cluster lifecycle (start/stop/cleanup)

2026-08-27 11:30

• Updated crc-ctl skill to v2.0.0: added crc-cleanup.sh script for VM disk cleanup (prune images, vacuum journals, clean API logs, delete completed pods), added Gotchas/Specification/Tests sections, expanded signal phrases

2026-08-27 11:19

• skupper-model-provider v8.7.0: Refactored common.sh, up.sh, down.sh, status.sh for dual CRC links (link-hub-rhel-ai + link-local-ezhang). All scripts now manage both CRC links and both listeners (9000 + 10000). Precheck topology diagram updated. Verified: status.sh, setup.sh –check, bash -n on all scripts.

2026-08-27 10:57

• skupper-model-provider v8.6.0: Added RouterAccess on local-ezhang (port 55672) and CRC→local link; CRC now reaches both rhel-ai and rhtevan-work models. Updated SKILL.md architecture, site config, and topology.env. Scripts (up.sh/down.sh/setup.sh) need follow-up refactor to manage second CRC link lifecycle.

2026-08-27 10:29

• crc-ols v2.2.1: Fix skupper-model provider model name granite-4.1-8b → granite-4.1-8b-fp8 in SKILL.md template (matches g8b-fp8-spec-128k profile)

2026-08-27 09:12

• Updated README.md Desktop/System category to include Obsidian Snap fix

2026-08-27 09:08

• Created skill fedora-obsidian-fix v1.0.0 — captures Obsidian Snap portal fix for Fedora Wayland (wrapper + .desktop override with –ozone-platform=x11)

2026-08-26 23:21

• skills/skupper-model-provider v8.5.0 — Simplified signal routing: dropped redundant ‘model’ from all signals. Short forms (start skupper, stop skupper on crc) now primary. Signal table uses PROVIDER/CONSUMER placeholders.

2026-08-26 23:11

• skills/skupper-model-provider v8.4.0 — Added consumer site signal routing (start/stop skupper model on crc). Scoping rules now distinguish provider vs consumer sites. Added S10a/S10b specs, T10a/T10b/T10c tests. No script changes.

2026-08-26 21:55

• agentfs-setup: Added deterministic scope auto-detection to scaffold-dotagents.sh and seed-agents-md.sh. Scripts now compare ROOT_DIR against CWD — different path = LITE, same path or no path = PROJECT. Agent no longer needs to decide –scope flag. Updated SKILL.md to reflect simplified agent workflow (just pass the target dir).

2026-08-26 21:48

• agentfs-setup SKILL.md: added LITE scope usage section with inference rule table, updated Sync and Verification sections for LITE scope awareness. This was missing and caused the agent to create PROJECT scope when given an explicit remote path.

2026-08-26 21:34

• Scope consolidation (v4.19.0): renamed –mode to –scope across all agentfs-setup scripts. Standardized all prose from ‘USER/PROJECT mode’ to ‘USER/PROJECT scope’. Added LITE scope for small-context models — minimal AGENTS.md template (~850 tokens), no skills/profiles/knowledge. Updated scaffold-dotagents.sh, seed-agents-md.sh, sync-agents-md.sh, verify-setup.sh, test-soul-authoring.sh. Added LITE scope guards to agentfs-profile, agentfs-eval, agentfs-ctx-chk, skill-gen, skill-harvest, skill-index, okf-bundle-gen, okf-bundle-harvest, okf-bundle-setup. Updated SKILL.md and CHANGELOG.md for agentfs-setup, agentfs-profile, agentfs-eval, agentfs-ctx-chk. Updated design-spec.md, design-decisions.md, hermes-agentfs-setup, okf-bundle-gen, skill-merge, README.md terminology.

2026-08-26 18:42

• README.md: added Goose Desktop Operations to knowledge bundle list (README audit fix)

2026-08-26 17:49

• hosted-model-ctl v7.4.0 (cont): updated SKILL.md profile table and references/deployment-profiles.md to version-agnostic Granite 3B naming. All 4.2-specific model references removed from user-facing docs.

2026-08-26 17:31

• hosted-model-ctl v7.4.0: rollback g3b-16k profile to version-agnostic granite-3b (auto-detect GGUF, prefer 4.1). Container rolled back to granite-4.1-3b on rhtevan-work. Goose config updated.

2026-08-26 17:07

• goose-skupper-provider v5.3.0: Fix test.sh chat completion for Granite 4.2 reasoning models — increase max_tokens, add reasoning_content fallback

2026-08-26 17:06

• hosted-model-ctl v7.3.1: Fix test.sh for Granite 4.2 reasoning models — increase max_tokens, add reasoning_content fallback

2026-08-26 17:03

• hosted-model-ctl v7.3.0: Upgrade g3b-16k profile from Granite 4.1 to 4.2 — model ID, GGUF repo/filename, deployment-profiles.md, SKILL.md

2026-08-26 14:29

• hosted-model-ctl: Added Section 12 (Self-Hosted vs Cloud Provider Comparison) to benchmark-report.md — measured Granite 8B FP8 spec-decode vs Opus 4.6 via GCP Vertex. Granite 2.4-3.3× faster on short/tool-call workloads (0.34s vs 2.54s tool-call latency), Opus 1.1× faster on sustained 1K token generation. Updated Section 8 speed table and Section 10 recommendations with measured cloud data.

2026-08-26 14:08

• goose-skupper-provider v5.2.0: Set supports_streaming:false in setup.sh and PROVIDER.md to work around goose v1.47.0 streaming parser bug with vLLM hermes tool calls. Confirmed fix in Test rhel-ai v3 session — model correctly called load_skill(agentfs-setup) and completed sync.

2026-08-26 14:01

• Updated goose-desktop-operations knowledge bundle: documented vLLM hermes streaming bug in goose v1.47.0 (streaming tool-call accumulator drops initial argument fragment when vLLM sends name and args as separate entries in same chunk). Fix: set supports_streaming:false in custom_skupper provider JSON. Updated custom_skupper.json provider config.

2026-08-26 13:40

• agentfs-setup v4.18.0: Added Guardrail #0 (Signal-First Dispatch) — ⛔ GATE requiring models to scan skill description signals in system prompt before generic interpretation. Addresses signal-routing failures observed on smaller models (Granite 8B failed to match ‘sync agentfs’ despite exact match in skill descriptions). Updated Guardrail Quick Reference table. Bumped template version.

2026-08-26 12:36

• okf-bundle-index v1.4.0: rebuild-index.sh sorts newest-first by mtime. agentfs-setup v4.17.8: post-edit.sh audits knowledge indexes, Gate 3 MUST use rebuild-index.sh.

2026-08-26 12:22

• agentfs-setup v4.17.7: Guardrail #5 template — log.md entries MUST use merge-log-entry.sh; added knowledge root log to scope table

2026-08-26 12:09

• Created knowledge bundle goose-desktop-operations (toolshim-and-tool-calls.md, custom-provider-schema.md). Moved Incident 5 from skupper-vllm-deployment to new bundle. Updated knowledge/index.md.

2026-08-26 12:04

• Added Incident 5 (GOOSE_TOOLSHIM Desktop hang) to skupper-vllm-deployment/agentfs-process-lessons.md knowledge bundle

2026-08-26 11:07

• goose-skupper-provider v5.1.0: Skill check fixes — added Signal Routing Table (10 patterns), expanded opening paragraph, updated signal phrases (test/recreate/check), fixed heredoc injection (env vars + single-quoted PYEOF), removed redundant PROVIDER.md changelog, concrete T8 test.

2026-08-26 10:33

• goose-skupper-provider v5.0.0: Replaced static alias mapping with API-driven model discovery. setup.sh accepts host/profile names, discovers model ID + context from live API. Added poison-JSON safeguard (Python json.dumps + round-trip + post-write validation + backup restore). Rewrote SKILL.md and PROVIDER.md.

2026-08-26 09:52

• skupper-model-provider v8.3.0: Aligned with hosted-model-ctl v7.2.0. Replaced alias layer with host-based routing (common.sh, test-model.sh, up.sh, down.sh). Updated SKILL.md routing table, signal routing, scoping rules. Updated van-topology diagram and re-delivered HTML.

2026-08-25 22:43

• hosted-model-ctl sanity check complete: Fixed –tool-call-parser granite→hermes in report.sh (8 occurrences). Replaced co-hosting combos with speculative decoding recommendation in Large tier report. Fixed vLLM ≥0.9→nightly in model-landscape.md (3 occurrences). Fixed –profile→direct profile name in benchmark-report.md (3 occurrences). Rewrote memory-budget.md to match current 4-profile architecture. Updated co-hosting finding to historical context in benchmark-report.md.

2026-08-25 22:11

• hosted-model-ctl: Removed gemma4-128k and qwen38-128k profiles — 11.8-13.4 tok/s not usable vs 58-79 tok/s default. Now 4 profiles total (2 per host). Cleaned all scripts, SKILL.md, deployment-profiles.md.

2026-08-25 22:07

• v7.1.0 hosted-model-ctl: Profile-only architecture. Removed Model Registry and on-demand individual aliases. All deployments go through profiles. Added g350m-2k profile. Renamed gemma4-31b→gemma4-128k, qwen38-27b→qwen38-128k. Rewrote all scripts (setup.sh, start.sh, stop.sh, status.sh, test.sh, list.sh) to accept profile names directly. Rewrote deployment-profiles.md as canonical reference with full specs per profile. Updated SKILL.md to v7.1.0.

2026-08-25 21:51

• v7.0.0 hosted-model-ctl: Major simplification — removed all co-hosting and solo-* profiles. New naming: g3b-16k, g8b-spec-128k, g8b-fp8-spec-128k. Added vllm-spec engine with SPEC_CONFIGS for speculative decoding. Added FP8+CUDA graphs benchmark results (58-79 tok/s). Updated common.sh, setup.sh, SKILL.md, CHANGELOG.md, benchmark-report.md. Removed obsolete aliases (g350m, g1b, g8b-lc), legacy engines (vllm-ilab, vllm-bnb), co-host GPU pinning logic.

2026-08-25 21:04

• Created benchmark-report.md under hosted-model-ctl/references — comprehensive 7-test benchmark results across all models (Granite 350M/3B/8B, Qwen3.8-27B, Gemma 4 31B), hosts (rhtevan-work, rhel-ai), and speculative decoding strategies (ngram, gemma4_mtp, draft_model). Includes hardware analysis, failed experiments, and final deployment recommendations.

2026-08-25 16:23

• hosted-model-ctl: Removed g1b model (Granite 4.0 1B vLLM-BnB) from common.sh and rhtevan-work. Cleaned container. Conducted benchmark tests on rhtevan-work: solo-g3b (Granite 3B Q4, ~22 tok/s) and solo-g8b-lc (Granite 8B Q4, ~6.5 tok/s). Both pass all 7 benchmark tests (instruction following, reasoning, tool calling, code gen, multi-step, conciseness, speed). g3b is 3.4x faster with comparable quality — confirms default profile choice.

2026-08-25 16:05

• hosted-model-ctl: Switched ALL Qwen3.8 entries from vLLM v0.27.1 to nightly image. Added –reasoning-parser qwen3 flag to both qwen38-27b and qwen38-27b-40k aliases. This fixes the thinking token leak in content field (CoT now routed to reasoning_content). Solo profile (solo-qwen38) will now use nightly + reasoning parser automatically — same fix as validated in co-host testing.

2026-08-25 15:23

• hosted-model-ctl: Fixed tool calling for ALL models. Root cause: Granite 4.1 outputs JSON XML format but the ‘granite’ parser in vLLM 0.8.4 expects bare JSON arrays — switched to ‘hermes’ parser which matches. For Qwen3.8 on vLLM 0.27.1, switched from ‘hermes’ to ‘qwen3_xml’ parser (new in 0.27.1). Both models now return finish_reason=tool_calls with properly parsed tool_calls array. Updated common.sh (qwen3_xml parser for all Qwen entries), setup.sh (hermes parser for vllm-ilab and vllm-bnb engines). Qwen3.8 still leaks thinking tokens in content field — separate issue requiring reasoning parser support.

2026-08-25 14:40

• hosted-model-ctl: Phase 7 co-hosting validated. Deployed co-g8b-qwen38 profile — Granite 8B BF16 TP=2 @128K on GPUs 0,1 (port 9000) + Qwen3.8-27B FP8 TP=2 @40K on GPUs 2,3 (port 9001). Both models responding simultaneously, all 4 GPUs ~21.2/23 GB utilized. Key learnings: (1) CDI per-GPU selection works with –device nvidia.com/gpu=N syntax + –security-opt label=disable. (2) CUDA_VISIBLE_DEVICES env var does NOT work inside CDI containers. (3) Qwen3.8 FP8 TP=2 maxes out at ~40K context (not 64K) due to enforce-eager overhead. Updated common.sh aliases (64k→40k), deployment-profiles.md, SKILL.md. Implemented GPU pinning in setup.sh for vllm and vllm-ilab engines.

2026-08-25 14:10

• hosted-model-ctl: Updated deployment-profiles.md, memory-budget.md, SKILL.md with corrected VRAM numbers from Gemma 4 deployment. Key corrections: BF16 removed (OOM), solo-gemma4-fp8 profile removed, all Gemma 4 entries now use RedHatAI FP8-block + enforce-eager. Added VRAM reality check table showing CUDA graph overhead. Co-host profiles marked NOT YET TESTED with GPU pinning requirements noted.

2026-08-25 14:06

• hosted-model-ctl: Phase 5 complete. Gemma 4 31B deployed successfully on rhel-ai with critical learnings: (1) BF16 OOMs on 4x L4 even at 48K context — CUDA graphs + weights consume ~19.3 GB/GPU leaving no room for KV cache. (2) Must use pre-quantized FP8 model (RedHatAI/gemma-4-31B-it-FP8-block). (3) Must use –enforce-eager to skip CUDA graph capture (~10 GB/GPU savings). (4) Must use vLLM nightly (>v0.27.1) due to AmbiguousGlobalPerLayerAttributeError with heterogeneous Gemma 4 layers. (5) Removed solo-gemma4-fp8 profile (256K) and gemma4-31b-fp8 alias — consolidated to single gemma4-31b alias using RedHatAI FP8-block. Updated common.sh model registry. Both solo-gemma4 and solo-qwen38 profiles tested and working (4/4 tests each). Default restored to solo-qwen38.

2026-08-25 11:53

• hosted-model-ctl: Phase 3+4 complete. Pulled vLLM v0.27.1 (21.6 GB) on rhel-ai. Downloaded Granite 4.1 3B GGUF Q4_K_M (2.0 GB) on rhtevan-work. Deployed solo-g3b profile on rhtevan-work (4/4 tests pass, model ready in 15s). Deployed solo-qwen38 profile on rhel-ai — Qwen3.8-27B-FP8 TP=4 @128K on vLLM v0.27.1 (4/4 tests pass, model ready in 555s including 28.8 GB weight download). Fixed PROFILE_STATE_DIR escaping in common.sh. Updated setup.sh with –profile support, g3b llamacpp config, upstream vLLM engine type. Both default profiles active and serving.

2026-08-25 10:19

• hosted-model-ctl v6.0.0: Created references/deployment-profiles.md (deployment profile system with mutual exclusion, 12 profiles, defaults). Updated references/model-landscape.md (added Gemma 4 31B/26B-A4B/12B, Qwen3.8-27B, Qwen3-Coder-30B, GLM-4-32B, Nemotron entries, vLLM v0.27.1 image, gemma4 tool-call parser). Updated references/memory-budget.md (VRAM budgets for all new models at TP=2/4). Updated SKILL.md to v6.0.0 (deployment profiles section, new model registry, profile-aware specs S1b/S3b/S4b/S5c, cold start times). Updated scripts/common.sh (new model entries, DEPLOY_PROFILES registry, profile helper functions). Updated scripts/start.sh (–profile flag, mutual exclusion, multi-model startup). Updated scripts/stop.sh (–profile flag, active profile cleanup). Updated scripts/status.sh (active profile display). Updated scripts/list.sh (–profiles flag, new aliases). Updated CHANGELOG.md.

2026-08-24 19:29

• dsh-setup v1.2.1: separate Chrome profile for security/maximize/process tracking; replaced connection monitoring with wait '$CHROME_PID'; fixed WMClass to chrome-127.0.0.1__-Default for Wayland icon matching; updated SKILL.md, CHANGELOG.md, install-desktop.sh, launcher, teardown.sh, verify.sh

2026-08-24 17:58

• dsh-setup v1.2.0: replaced PID-based idle detection with systemd user service (dsh.service) + TCP connection monitoring; fixed ss header line false-positive; updated SKILL.md, install-desktop.sh, verify.sh, teardown.sh, launcher

2026-08-24 16:59

• dsh-litellm-provider v1.1.0: fixed apiKeyEnv (DSH rejects empty — now uses LITELLM_VERTEX_AI_API_KEY with dummy value), fixed verify.sh bash arithmetic and model grep patterns; dsh-setup: updated launcher and install-desktop.sh to export dummy API key env var; dsh icon whale size bumped 680→760px

2026-08-24 14:26

• Guardrail #5 catch-up: bumped dsh-setup v1.0.0→v1.1.0 (post-install fixes), dsh-litellm-provider v1.0.0→v1.0.1 (markdown fix), updated changelogs, regenerated skills/index.md

2026-08-24 14:24

• Updated goose-desktop-env-fix v1.2.0 → v1.3.0: added Gotchas section documenting hermit node wrapper PATH interference and per-skill workaround pattern

2026-08-24 14:01

• Fixed dsh-litellm-provider/SKILL.md markdown rendering (blank lines around code fence). Added official DSH whale icon (favicon.svg → dsh.png) to dsh-setup/assets/.

2026-08-24 13:52

• Created skills: dsh-setup v1.0.0 (pnpm-based DSH install, Chrome app-mode launcher, .desktop integration, update/teardown/verify) and dsh-litellm-provider v1.0.0 (auto-discover models from LiteLLM, write DSH settings.yaml with compat flags). Regenerated skills/index.md.

2026-08-21 00:04

• agentfs-setup v4.17.6: bump version for Guardrail #5 scope note addition
• agentfs-setup seed-agents-md.sh: add Guardrail #5 scope note — README.md at ~/.agents/ is AgentFS-managed and not exempt from post-edit.sh + log.md requirements

2026-08-20 23:52

• README.md: fix Guardrail #10 description — README audit trigger updated from ‘README.md staged + staleness Clean’ to ‘agentfs files staged’

2026-08-20 23:44

• agentfs-readme-audit v1.2.0 + pre-push-scan.sh: fix README audit trigger — runs whenever agentfs files staged (not gated on README.md being staged); Category 7 replaced with README_AUDIT_REQUIRED signal; seed-agents-md.sh Guardrail #10 updated to match

2026-08-20 23:27

• agentfs-setup: v4.17.5 — author-soul.sh is_stub() fixed to use awk multi-line comment stripping, consistent with sync-agents-md.sh

2026-08-20 23:24

• agentfs-setup: v4.17.4 — sync-agents-md.sh emits SOUL_ACTION_REQUIRED signal; SKILL.md documents agent post-sync guidance flow (apply default / customise / skip)

2026-08-20 23:17

• agentfs-setup: sync-agents-md.sh — added SOUL.md stub detection; warns with author-soul.sh command when SOUL is empty or missing, even when AGENTS.md is already up to date

2026-08-20 22:55

• agentfs-setup v4.17.2: fix merge-changelog-entry.sh timestamp format (HH:MM added); dedup now matches on date+version

2026-08-20 22:48

• agentfs-readme-audit v1.1.0: trigger condition corrected (staged AND Clean); flow diagram updated
• agentfs-setup v4.17.1: test-pre-push-scan.sh Cat 9+10 (25 pass); test-agents-md-fidelity.sh 84 checks; sync-agents-md.sh pre-versioning fallback; agentfs-readme-audit v1.1.0 trigger fix

2026-08-20 22:40

• agentfs-setup v4.17.0: add sync-agents-md.sh (deterministic idempotent sync, fixes Agent Profiles duplication bug); update SKILL.md docs

2026-08-20 22:26

• agentfs-setup v4.16.0: Guardrail #10 rendered Markdown report; Category 10 memory PII detection in pre-push-scan.sh; 5th violation added; project AGENTS.md synced

2026-08-20 22:07

• agentfs-setup v4.15.0: harden Guardrail #10 — numbered workflow, README audit trigger fix, same-turn confirmation, 4 explicit violations

2026-08-20 21:45

• agentfs-setup v4.14.0: add merge-changelog-entry.sh (deterministic CHANGELOG, repair, dedup); CHANGELOG coverage in pre-push-scan.sh (Cat 9); post-edit.sh reminder; seed-agents-md.sh delegation table; README Guardrail #5

2026-08-20 19:04

• README.md: fixed SOUL.md auto-load description (@import, not agent-initiated); updated profiles/ tree entry (recipe.yaml, output/); added author-soul.sh and gen-profile-recipe.sh to Agent Setup category; updated memory model table

2026-08-20 18:44

• agentfs-profile v1.9.0: refactor create-profile.sh (delegate SOUL to author-soul.sh), add recipe.yaml + output/ dir, add gen-profile-recipe.sh
• agentfs-setup v4.13.0: add author-soul.sh, @.agents/SOUL.md import, expanded Guardrail #7, SOUL verify checks, test suite (21 tests)

2026-08-20 10:30

• Updated fedora-nm-boot-slow to v1.1.0 — added Specification and Tests sections for P4 compliance
• Updated skill-gen to v3.2.0 — added explicit GATE requiring user permission before skipping P4/P5 principles

2026-08-20 10:16

• Created skill fedora-nm-boot-slow v1.0.0 — documents root cause and remediation of NM-wait-online boot delay on Fedora

2026-08-19 16:18

• Added .pre-push-allowlist for semantic false-positive filtering in Guardrail #10 Git Push Safety. Updated seed-agents-md.sh workflow and Quick Reference. Bumped agentfs-setup to v4.12.0. Removed hardcoded skill count from README.md.

2026-08-19 14:56

• hosted-model-ctl: simplified report signal phrases to ‘model hosting report’ and ‘hosting machine report’

2026-08-19 13:59

• hosted-model-ctl: strengthened report rendering instruction — agent MUST echo full stdout in response message since tool outputs may not be visible to user in many UIs

2026-08-19 13:22

• hosted-model-ctl v5.4.1 — Fixed report operation agent instruction (present output as-is). Fixed signal phrases with hosted model prefix for reliable routing.

2026-08-19 11:31

• hosted-model-ctl v5.4.0 — Added report operation (S8) with report.sh script generating 3-section platform report (basic specs, accelerator, model recommendations by VRAM tier). Added references/model-landscape.md curated model database. New signals: hosted model report, machine spec, host report. Updated rhtevan-work gotcha: nvidia-smi now available.

2026-08-18 23:29

• Added interactive diagram links to skupper-model-provider SKILL.md Architecture section

2026-08-18 23:23

• [OVERRIDE] Pre-push scan: hostname in index (intentional) and README staleness (false positive — skill count unchanged, only sort order). User confirmed.

2026-08-18 23:20

• Fixed regen-skill-index.py: now uses max mtime across all files in skill directory instead of SKILL.md only; ensures index date reflects any file change (docs, CHANGELOG, scripts)

2026-08-18 23:16

• Added CHANGELOG.md entry for skupper-model-provider: architecture and lifecycle diagrams added to docs/

2026-08-18 23:11

• [OVERRIDE] Pre-push scan: 5 false-positive categories overridden (archify test fixtures: mock emails, SVG coordinates as phone numbers, test URLs, variable name ‘authorization’, hostnames in diagram content). User confirmed.

2026-08-18 23:10

• Updated README.md skill count 50 → 51 (archify added)

2026-08-18 23:05

• Created architecture and lifecycle diagrams for skupper-model-provider skill using archify: van-topology.architecture.html and operations.lifecycle.html in docs/

2026-08-18 22:39

• Added Defaults section to archify SKILL.md: working directory (project root), output location (docs/).

2026-08-18 22:32

• Updated archify description to all Command-pattern signal phrases (verb+noun), 9 signals covering all 5 diagram types + mermaid conversion

2026-08-18 22:22

• Restored original archify description as first paragraph in SKILL.md body, preserving context for agents loading the skill

2026-08-18 22:20

• Updated archify skill frontmatter: added metadata.tags, converted description to concise signal phrases per AgentFS skill conventions

2026-08-18 22:11

• Installed archify skill (v2.15.0) via npx skills add tt-a1i/archify -g; doctor checks all passed

2026-08-18 22:00

• Added custom CSS override to reduce Cayman theme header height

2026-08-18 21:55

• Changed GitHub Pages theme from minimal to cayman for wider content area

2026-08-18 21:48

• Excluded index.md from Jekyll build in _config.yml so README.md serves as GitHub Pages homepage

2026-08-18 21:32

• Added _config.yml (Jekyll theme: minimal) and permalink front matter to README.md to enable GitHub Pages rendering with README.md as homepage

2026-08-17 11:22

• agentfs-setup v4.11.0 — Enhanced Gate guardrails (⛔ GATE with STOP language) in seed-agents-md.sh template and design-spec.md

2026-08-17 11:16

• Fixed 5 pre-existing version mismatches (fuseki 1.1.0, goose-agentfs-setup 1.4.0, hermes-desktop-fixes 2.1.0, hermes-headroom-provider 1.2.1, skupper-model-provider 8.1.1). Updated AGENTS.md Guardrails #5, #9, #10 from 🚧 to ⛔ GATE with STOP language and explicit violation definitions.

2026-08-17 10:08

• hosted-model-ctl v5.3.0 — Added tool-calling flags to vllm-ilab engine, fixed local keyword outside function in setup.sh

2026-08-17 10:01

• crc-ols v2.2.0 — Added self-hosted Skupper model provider documentation (setup pattern, Provider Types Reference, dependency notes)

2026-08-16 19:36

• agentfs-setup v4.10.0: merge-log-entry.sh duplicate heading fix; version bump

2026-08-16 19:32

• merge-log-entry.sh: fixed duplicate heading bug — replaced echo-pipe grep with direct file grep for heading detection

2026-08-16 19:27

• README.md: updated Guardrail #5 description to reflect edit-time logging rule and deterministic log coverage check
• Guardrail #5: added edit-time logging rule to seed-agents-md.sh; added log coverage check (Category 8) to pre-push-scan.sh; bumped agentfs-setup to 4.9.0

2026-08-16 19:20

• README.md: fixed skill count (48→50), added 2 missing knowledge bundles (llm-inference-constrained-gpu, skupper-vllm-deployment), updated Guardrail #5 description to match current template (completion gate pattern)

2026-08-16 19:11

• agentfs-setup v4.8.0 — integrated agentfs-readme-audit into Guardrail #10 (Git Push Safety): semantic README alignment check runs after pre-push-scan.sh returns Clean for README staleness

2026-08-16 19:02

• Created agentfs-readme-audit skill v1.0.0 — semantic README alignment check complementing pre-push-scan.sh deterministic staleness check

2026-08-16 18:41

• agentfs-setup v4.7.0 — pre-push-scan.sh: promoted README staleness check from interactive prompt to structured report row (Category 7); removed y/n notice that gets swallowed in non-interactive execution

2026-08-16 17:45

• agentfs-setup v4.6.0: Guardrail #5 Filesystem Integrity rewrite — separated WHAT (completion gate) from HOW (delegation table with Gate column); added scope rule table for log routing; scripts path declared once, removed inline code block

2026-08-16 16:54

• agentfs-setup v4.4.0: Template version now derived from metadata.version (single source of truth).

2026-08-16 16:49

• agentfs-setup v4.3.0: Made Guardrail #5 completion gate explicit (4-item checklist). Added version-CHANGELOG alignment check to regen-skill-index.py (v4.1.0). Synced AGENTS.md to v4.3.0.

2026-08-16 16:19

• AGENTS.md optimization (v3.14→v4.0): extracted regen-skill-index.py from post-edit.sh (DRY), created pre-push-scan.sh and checkpoint.sh (Strategy A), compressed guardrails prose (Strategy C), added completion gate to Guardrail #5, updated seed-agents-md.sh template, updated skill-index/SKILL.md to use canonical script, added test scripts for validation.

2026-08-16 13:38

• Moved merge-log-entry.sh from okf-bundle-gen to agentfs-setup (shared utility in foundational skill). Updated 17 references across 6 skills.
• Removed Root Log Format subsection from Guardrail #5 (mechanics now in script). Updated delegation table: root logs use merge-log-entry.sh, skills index uses post-edit.sh.
• agentfs-setup v3.13.0→v3.14.0, template version 3.13→3.14

2026-08-15 12:00

• Added post-edit.sh to agentfs-setup skill: automates skills index regeneration and log.md anchor validation. Fragile steps → code (skill-gen principle).
• Updated Guardrail #5 STOP block and Quick Reference row to reference post-edit.sh in both AGENTS.md and seed-agents-md.sh template
• agentfs-setup v3.12.0→v3.13.0, template version 3.12→3.13

2026-08-15 11:32

• Fixed agentfs-setup template version mismatch: seed-agents-md.sh had 3.10, should be 3.12 (matching SKILL.md and AGENTS.md). This blocked agentfs-setup --sync from updating project AGENTS.md files.

2026-08-14 19:40

• Guardrail #5 final trim: removed Post-Edit Completeness subsection (redundant with delegation table); STOP block moved to one-liner at top; 64→45 lines. agentfs-setup v3.11.0→v3.12.0, template version 3.11→3.12.

2026-08-14 19:19

• Rewrote Guardrail #5 in AGENTS.md and agentfs-setup template: replaced 4 bloated subsections (Link Integrity, Log & Changelog Currency, Index Currency, Post-Edit Completeness) with 4 focused sections (Editing Rules, Log & Index Delegation table, Root Log Format, Post-Edit Gate). Removed 26 lines of redundancy.
• Added Log & Index Delegation table mapping every log/index file to its owning skill
• Standardized bullet prefix * → - in merge-log-entry.sh, skill-harvest, okf-bundle-gen, okf-bundle-harvest, okf-bundle-setup
• Fixed ~/.agents/knowledge/log.md missing comment line
• Fixed skill-schema.md stale ## Changelog section → CHANGELOG.md reference
• Added insertion anchor reference to skill-gen log update checklist item
• Updated okf-bundle-setup log format rules to reference comment anchor and - bullet
• Regenerated ~/.agents/skills/index.md (49 skills)

2026-08-14 12:33

• Strengthened Guardrail #5 Post-Edit Completeness with hard STOP gate in AGENTS.md and agentfs-setup template: “Do NOT respond until every box is checked”; clarified index scope includes supporting scripts

2026-08-14 12:29

• Strengthened Guardrail #5 insertion anchor rule in AGENTS.md and agentfs-setup template: before anchor MUST be the comment line, NEVER a ## heading; always head -6 before editing

2026-08-14 12:13

• Updated skupper-model-provider v8.2.0: Partial Availability — up.sh/down.sh skip unreachable hosts instead of hard-failing; added Site Roles (Provider/Consumer); added Partial Availability agent protocol; added S9a-c specs and T9a-c tests; updated CHANGELOG

2026-08-14 09:30

• Created skill: fedora-desktop-wmclass-fix (USER scope)
• New files: SKILL.md, CHANGELOG.md, scripts/wmclass-audit.sh
• Audits and fixes StartupWMClass in .desktop files for Electron apps on Wayland
• Documents lowercase binary name convention for Electron Wayland app-ids
• Regenerated ~/.agents/skills/index.md (49 skills)

2026-08-13 19:50

• Fixed skupper-model-provider service auto-start bug: skupper system start re-enables services via default.target.wants/ symlinks on every run; added systemctl --user disable to install_router_auto_restart() and install_controller_auto_restart() in common.sh; applied fix to all 3 hosts; updated Known Issues, CHANGELOG (v8.1.1)

2026-08-13 19:23

• Externalized ## Changelog sections from all 48 skills’ SKILL.md → CHANGELOG.md files
• SKILL.md now contains only > See [CHANGELOG.md](./CHANGELOG.md) for version history. reference
• Updated skill-gen v3.0.0 → v3.1.0: template, Post-Creation Checklist, Skill Check P3, edit-existing guidance
• Updated skill-gen/references/skill-schema.md v2.0.0 → v2.1.0: Changelog Rules, Anti-Patterns table
• Rationale: changelogs grow unbounded and bloat model context window on load_skill

2026-08-13 18:11

• Skill skupper-model-provider v8.0.0 → v8.1.0 — Network Observer UI on CRC
• Modified 6 files: topology.env.example, topology.env, common.sh, setup.sh, teardown.sh, status.sh, SKILL.md
• Regenerated skills/index.md

2026-08-13 17:39

• Skill skupper-model-provider v7.4.1 → v8.0.0 — CRC as 4th VAN site
• Modified 10 files: topology.env.example, topology.env, common.sh, setup.sh, teardown.sh, up.sh, down.sh, status.sh, test-model.sh, SKILL.md
• Regenerated skills/index.md (48 skills)

2026-08-13 12:52

• Skill check skupper-model-provider v7.4.0 → v7.4.1: teardown.sh now removes router start-watch.sh; setup.sh uses mktemp+chmod 600 for link YAML temp files; POSTMORTEM.md Architecture Decision #4 updated for v6.1.1; down.sh Phase 1 comment clarified; regenerated skills/index.md

2026-08-13 12:32

• Signal Phrase Schema Migration (v2.0.0): description field redefined as signal phrases (Command: verb+noun(s), Query: noun(s)); metadata.signals removed from schema
• Updated skill-gen/references/skill-schema.md to v2.0.0: Signal Phrase Rules, Opening Paragraph requirement, Progressive Disclosure Model
• Updated skill-gen/SKILL.md to v3.0.0: new description format, updated template, checklist, Skill Check P1 items
• Updated skill-index/SKILL.md to v3.0.0: index columns Skill | Tags | Description | Updated; legacy metadata.signals warning
• Updated agentfs-setup/references/design-spec.md to v3.11: SKILL.md Frontmatter Schema, Signal Routing architecture
• Updated agentfs-setup/scripts/seed-agents-md.sh: Routing Rules reference description field instead of metadata.signals
• Migrated all 48 skills: converted description to signal phrases, removed metadata.signals from frontmatter
• Regenerated skills/index.md with new Skill | Tags | Description | Updated format

2026-08-13 11:37

• Updated skupper-model-provider (v7.3.0 → v7.4.0): Simplified start-watch.sh from 20 lines to 6 — removed bash SIGTERM trap, stop marker file, and podman inspect exit code check. Now uses direct podman wait exit code propagation + SuccessExitStatus=SIGTERM in systemd units. Fixes service marked failed (exit 143) after systemctl stop. Patched and verified on all 3 hosts.
• Regenerated ~/.agents/skills/index.md (48 skills).

2026-08-12 22:55

• Updated Guardrail #5 (Index Currency): consolidated skill-index requirement into MUST-stay-current bullet with explicit load_skill(name: "skill-index") call-out. Applied to agentfs-setup template (v3.10.1) and project AGENTS.md.
• Regenerated ~/.agents/skills/index.md (48 skills) — agentfs-setup now at top.

2026-08-12 22:48

• Updated Guardrail #5 (Log & Changelog Currency): added 24-hour format hint with date command to ISO 8601 timestamp rule. Applied to agentfs-setup template and project AGENTS.md.

2026-08-12 22:17

• Fixed skupper-model-provider/scripts/up.sh: removed recreate_router_with_tmpfs() from start path. All hosts now use identical systemctl --user start path. Tmpfs workaround is setup-only — container retains flags across stop/start cycles. Restores systemd auto-restart (start-watch.sh) on rhel-ai.
• Fixed skupper-model-provider/scripts/down.sh: added explicit podman stop (NOT rm) on tmpfs-workaround hosts as safety net after systemctl stop.
• Updated skupper-model-provider SKILL.md (v7.2.0 → v7.3.0): split S7→S7a/S7b for normal vs tmpfs-workaround host auto-restart. Added T7a/T7b. Added Gotchas for up.sh bypassing systemd on rhel-ai. T7b pending verification (rhel-ai unreachable).
• Regenerated ~/.agents/skills/index.md (48 skills).

2026-08-12 14:12

• Fixed critical bug in skupper-model-provider/scripts/down.sh: scoped mode (down.sh HOST) unconditionally stopped local router and controller, killing routes to other still-active hosts. Now checks if other remote hosts have active routers before stopping local infrastructure.
• Hardened skupper-model-provider/scripts/up.sh: scoped mode Phase 4 verification now checks that other routes are unaffected (not just the scoped host’s port).
• Updated skupper-model-provider SKILL.md (v7.1.0 → v7.2.0): split S2→S2a/S2b, S3→S3a/S3b/S3c with negative assertions for scoped operations. Added tests T2b/T2c/T3b/T3c/T3d/T3e covering all scoped start/stop scenarios. Added Gotcha for shared-infrastructure incident.
• Updated hosted-model-ctl SKILL.md (v5.1.0 → v5.2.0): split S5→S5a/S5b with negative assertion for single-alias stop. Updated tests T5a/T5b. Added Scoping Safety gotcha. Scripts confirmed correct (no code changes needed).
• Regenerated ~/.agents/skills/index.md (48 skills).

2026-08-12 11:35

• skupper-model-provider v7.1.0: Enhanced status.sh with dual-column reporting (Last-Known vs Live). Sites show controller+router systemd state with STALE flag. Links include TCP probe. Listeners check local port binding. Removed redundant systemd section.

2026-08-12 11:20

• skupper-model-provider v7.0.1: Clarified loose coupling semantics — agent semantically triggers hosted-model-ctl (no cross-script invocation)
• Stripped model container and e2e sections from status.sh (VAN-only)
• Added skupper model provider status/check skupper model provider signals
• Regenerated skills/index.md (48 skills)

2026-08-12 11:10

• skupper-model-provider v7.0.0: Decoupled model lifecycle from VAN scripts
  • up.sh/down.sh now manage Skupper infrastructure only (controllers + routers)
  • Model containers delegated to hosted-model-ctl at agent level (DRY + Loose Coupling)
  • Added Agent Orchestration section with signal routing, scoping rules, error handling
  • Updated signals, Specification (S1–S8), Tests (T1–T8), Relationship table
• hosted-model-ctl v5.1.0:
  • Changed default rhel-ai model from g30b-96k to g8b-128k
  • Removed ambiguous bare signals (model list/start/stop/status)
  • Added hosted model teardown/teardown hosted model signals
  • Added stop.sh --remove flag for container teardown
• Regenerated skills/index.md (48 skills)

2026-08-12 09:10

• skupper-model-provider v6.1.3: Fixed down.sh Phase 1 container filter to exclude router containers
• Regenerated skills/index.md (48 skills)

2026-08-12 09:04

• skupper-model-provider v6.1.2: Removed [Install]/WantedBy=default.target from systemd unit templates in common.sh
• Disabled skupper services on all 3 hosts to prevent auto-start on reboot with Linger=yes
• Updated SKILL.md: new known issue entry, updated auto-restart fix description, changelog, version bump
• Regenerated skills/index.md (48 skills)

2026-08-11 18:49

• skupper-model-provider v6.1.1: down.sh all now stops controllers on all 3 hosts (previously left running)
• Updated SKILL.md: version bump, script table, shutdown example, known issues table, changelog
• Regenerated skills/index.md (48 skills)

2026-08-11 18:28

• Sanitized POSTMORTEM.md: replaced site-specific hostnames (bastion.g7cpg.*) with *.example.com, IPs with placeholders, local-ezhang with local-site
• Sanitized SKILL.md changelog and log.md: replaced local-ezhang with local-site
• Regenerated USER skills index.md (48 skills)

2026-08-11 18:00

• skupper-model-provider v6.1.0: Extracted site-specific config (IPs, hostnames, SANs, usernames) from scripts to topology.env
• Created topology.env.example with placeholder values for safe git commit
• Added precheck capability (setup.sh --check): topology display + validation (podman, skupper, SSH, DNS, SANs, ports)
• Added signals: skupper model precheck, skupper model topology, show skupper topology
• Added topology.env and .rollback/ patterns to .gitignore
• Updated SKILL.md architecture, site config, and operations sections to reference topology.env variables

2026-08-11 17:17

• skupper-model-provider v6.0.1: Added Tests section (T1–T7 mapped to S1–S7) per skill-check Principle 4 finding
• Regenerated USER skills index.md (48 skills)

2026-08-11 17:12

• Fixed skill-gen metadata.signals: added Skill Check mode triggers (skill check, check skill, scan skill, audit skill, verify skill quality) that were missing from YAML frontmatter — only existed in prose body text, invisible to signal routing
• Regenerated USER skills index.md (48 skills)

2026-08-11 16:57

• skupper-model-provider: Added POSTMORTEM.md documenting 9 root causes, architecture decisions, and lessons learned

2026-08-11 16:50

• skupper-model-provider v6.0: Complete refactor — separated setup.sh/teardown.sh (one-time infrastructure) from up.sh/down.sh (daily start/stop)
• New scripts: setup.sh (416 lines), teardown.sh (85 lines); refactored up.sh (179 lines), down.sh (107 lines), status.sh (127 lines), common.sh (319 lines)
• Added auto-restart patches for router + controller (start-watch.sh + Restart=on-failure) on all 3 hosts
• Fixed: unique site names (hub-rhel-ai, hub-rhtevan-work, local-site), podman 4.x /tmp workaround, cert perms, SANs on RouterAccess, manual link building
• VAN fully operational: localhost:9000 → granite-4.1-8b (rhel-ai), localhost:10000 → granite-4.0-350m (rhtevan-work)
• Regenerated USER skills index.md (48 skills)

2026-08-11 11:50

• Added Guardrail Type System to AgentFS: Gate 🚧, Rule ⚖️, Habit 🔄
• Updated agentfs-setup/references/design-spec.md: added type badges to guardrail list, added Guardrail Type System sub-section with rationale and Trigger/Invariant collapse analysis
• Updated agentfs-setup/scripts/seed-agents-md.sh: Type column in Quick Reference table, verb-chain Key Actions, type badges on all 10 headings, gate blockquotes on #5/#9/#10, template version 3.9→3.10
• Updated agentfs-setup/SKILL.md changelog with v3.10.0 entry
• Synced AGENTS.md in project goofing-around to template v3.10, re-injected project-owned profiles (verifier, watchdog)
• Fixed skupper-model-provider/scripts/status.sh: replaced || echo "0" with || true on pipefail-sensitive pipelines
• Updated skupper-model-provider/SKILL.md changelog with v5.1 entry

2026-08-11 10:40

• Fixed skupper-model-provider/scripts/status.sh: replaced || echo "0" with || true on ss | grep | grep -c pipelines to prevent set -euo pipefail abort and 0\n0 arithmetic parse errors
• Updated skupper-model-provider/SKILL.md changelog with v5.1 entry

2026-08-10 16:45

• Updated litellm-vertex-ai-proxy v3.0.0: extracted all inline execution code to 4 scripts (setup.sh, detect-sa.sh, test-sa.sh, probe-models.sh); SKILL.md reduced from 308→202 lines; added Gotcha #7; added T8–T11 tests; workflow 8→6 steps
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-10 16:25

• Added `

2026-08-13 12:32

• Signal Phrase Schema Migration (v2.0.0): description field redefined as signal phrases (Command: verb+noun(s), Query: noun(s)); metadata.signals removed from schema
• Updated skill-gen/references/skill-schema.md to v2.0.0: Signal Phrase Rules, Opening Paragraph requirement, Progressive Disclosure Model
• Updated skill-gen/SKILL.md to v3.0.0: new description format, updated template, checklist, Skill Check P1 items
• Updated skill-index/SKILL.md to v3.0.0: index columns Skill | Tags | Description | Updated; legacy metadata.signals warning
• Updated agentfs-setup/references/design-spec.md to v3.11: SKILL.md Frontmatter Schema, Signal Routing architecture
• Updated agentfs-setup/scripts/seed-agents-md.sh: Routing Rules reference description field instead of metadata.signals
• Migrated all 48 skills: converted description to signal phrases, removed metadata.signals from frontmatter
• Regenerated skills/index.md with new Skill | Tags | Description | Updated format ` comment anchor to ~/.agents/log.md (was missing)
• Strengthened Guardrail #5 insertion anchor in AGENTS.md: added “read current file head before inserting” to prevent stale-context log misordering

2026-08-10 16:15

• Updated skill-gen v2.1.1: strengthened Gotchas from suggestion to requirement; added Gotchas section to SKILL.md template; added P1 checklist item for Gotchas/Known Issues
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-10 15:58

• Updated skill-gen v2.1.0: P1 inline code boundary guidance — reference vs execution code distinction; added checklist item
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-10 15:50

• Updated litellm-vertex-ai-proxy v2.0.0: updated model list; documented us-east5 as explicit region; added scripts/verify.sh (11 checks) and scripts/probe-models.sh (model discovery); added Specification and Tests; added SA key safety warning; fixed frontmatter
• Updated goose-litellm-provider v1.6.0: removed sonnet-4-5; fixed context_limits (opus/sonnet: 1M, haiku: 200k)
• Updated hermes-litellm-provider v2.2.0: removed sonnet-4-5 from reference config
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-10 15:26

• Updated goose-litellm-provider v1.5.0: added claude-haiku-4-5 to models; changed fast_model to claude-haiku-4-5; updated Vertex AI region from global to us-east5
• Updated hermes-litellm-provider v2.1.0: added claude-haiku-4-5 to reference config; updated region note
• Updated LiteLLM proxy config: all 4 models pinned to us-east5
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-10 14:10

• Updated hermes-litellm-provider v2.0.0: rewrote to match live Hermes config schema; added scripts/verify.sh and scripts/restore.sh; added Specification and Tests; removed hardcoded dummy api_key; documented hermes model persistence bug
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-10 12:55

• Updated goose-litellm-provider v1.4.0: removed pyyaml dependency from verify.sh (replaced with awk); consolidated Steps 1-2 into single pre-flight check; renumbered steps (6→5)
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-10 12:02

• Updated skill-gen v2.0.0: added Principle 5 (Security & Trust Boundary) with 6-item checklist; “Four Principles” → “Five Principles”; added “Real Expertise” anti-pattern and Gotchas guidance to Step 1; added “loose steps → instructions, fragile steps → code” aphorism to Agent-as-Orchestrator; updated Skill Check Procedure to 9 steps
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-10 11:23

• Updated skill-gen v1.9.0: strengthened P1 to “Accuracy, Consistency & Testability (Code-First)” — skills as prescriptive SOPs verified by execution; code-first for accuracy, consistency, conciseness, testability; added could-be-script 🟡 flag and knowledge-extraction checklist items; moved Concise from P3 to P1; renamed P3 to “Traceable & Well-Formatted”
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-10 10:38

• Updated goose-litellm-provider v1.3.0: set fast_model to claude-sonnet-4-6; added Model Selection Architecture section; added scripts/verify.sh and scripts/restore.sh; added Specification and Tests sections; consolidated workflow steps; added defensive template warning; normalized changelog versions
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-08 14:59

• Complete rewrite of hosted-model-ctl v5.0: 7 operational scripts, Specification, Tests, compact tables (659→229 lines)
• Complete rewrite of skupper-model-provider v5.0: 5 scripts (up/down/status/test/common), podman platform, interior mode (334→194 lines)
• Complete rewrite of goose-skupper-provider v4.0: 3 scripts (setup/teardown/test), JSON schema enforcement (127→126 lines)
• Updated skill-gen v1.8: Skill Check mode with 4 principles (Accuracy, Currency, Concise, Verifiable Spec)
• Fixed model ID: granite-4.1-8b-instruct → granite-4.1-8b in PROVIDER.md
• All scripts syntax-verified and live-tested
• Regenerated skills/index.md (48 skills)

2026-08-08 12:15

• Created OKF knowledge bundle skupper-vllm-deployment with 5 concept documents
• Added bundle to ~/.agents/knowledge/index.md

2026-08-08 10:55

• Updated agentfs-setup v3.8→v3.9: added “Never improvise when a skill exists” routing rule; added “Backup untracked files” to Guardrail #9
• Synced AGENTS.md to template v3.9
• Updated skill-gen v1.5→v1.6: added “Defensive file templates” writing guidance
• Updated skill-gen/references/skill-schema.md v1.0→v1.1: added writes-files optional field
• Regenerated skills/index.md (48 skills)

2026-08-08 10:09

• Fixed goose-skupper-provider/PROVIDER.md v3.0→v3.1: added schema warning, templated JSON placeholders; root cause: agent bypassed skill and wrote invalid custom_skupper.json with non-Goose fields
• Fixed custom_skupper.json to use correct Goose custom provider schema
• Regenerated skills/index.md (48 skills)

2026-08-08 01:43

• Updated skupper-model-provider/SKILL.md v3.0→v4.0: complete rewrite for interior mode (not edge), podman platform (not linux/systemd), routing keys model-api-rhtevan-work and model-api-rhel-ai, ports 10000 (rhtevan-work) and 9000 (rhel-ai), inter-router links on 55671 and 8000, documented podman gotchas
• Updated hosted-model-ctl/SKILL.md v3.2→v4.0: rhtevan-work model port 8000→10000, updated all container commands and test commands
• Updated goose-skupper-provider/SKILL.md v2.0→v3.0 and PROVIDER.md: port 8000→10000 for rhtevan-work, routing key model-api-rhtevan-work
• Regenerated skills/index.md (48 skills)
• Deployed full Skupper VAN interior mesh: localhost (outbound only) ↔ rhtevan-work:55671 ↔ rhel-ai:8000, all podman platform with official skupper-router:3.5.2 image
• Recreated rhtevan-work model containers (g350m, g1b, g8b) on port 10000
• Masked old linux-platform skupper-model-provider.service on rhel-ai

2026-08-07 00:38

• Updated goose-skupper-provider/SKILL.md v1.2→v2.0: model-to-port routing table (8000 for rhtevan-work, 9000 for rhel-ai), default g350m, signal parsing for model alias
• Updated goose-skupper-provider/PROVIDER.md v1.1→v2.0: multi-port base_url, added g30b-96k and g8b-128k to model reference
• Updated skupper-model-provider/SKILL.md v2.0→v3.0: two routing keys (model-api, model-api-rhel-ai), two local ports (8000, 9000), rhel-ai edge port 8000 (AWS 45671 blocked), multi-hub architecture
• Updated hosted-model-ctl/SKILL.md v3.1→v3.2: rhel-ai models serve on port 9000 (was 8000), port 8000 reserved for Skupper edge on rhel-ai
• Regenerated skills/index.md (48 skills)

2026-08-06 21:14

• Renamed local-model-ctl → hosted-model-ctl: directory, SKILL.md name, description, signals (hosted model * patterns), tags (hosted, self-hosted), version 3.0→3.1
• Updated skupper-model-provider/SKILL.md: all 5 references to local-model-ctl → hosted-model-ctl
• Regenerated skills/index.md (48 skills)

2026-08-06 21:03

• Updated goose-skupper-provider/SKILL.md v1.1→v1.2: Added recreate capability (teardown + setup); useful when remote model changes
• Regenerated skills/index.md (48 skills)

2026-08-06 20:58

• Fixed skill-index/SKILL.md v2.3→v2.4: Replaced fragile metadata block regex with line-by-line state machine for signals extraction; added inline bracket format support; signals now correctly populate for all 48 skills (was 4/48)
• Updated local-model-ctl/SKILL.md: All rhel-ai models now use port 8000 (single endpoint, one model at a time)
• Updated skupper-model-provider/SKILL.md: Model alias auto-routes to remote host; single localhost:8000 endpoint; “bring up”/”shutdown” invocation patterns; added signals
• Regenerated skills/index.md (48 skills, 48 with signals)

2026-08-06 20:04

• Updated local-model-ctl/SKILL.md v2.3→v3.0: Added rhel-ai host profile (4× NVIDIA L4, 92 GB VRAM); added g30b-96k (Granite 4.1 30B, BF16, tp=4, 96K context) and g8b-128k (Granite 4.1 8B, BF16, tp=2, 128K context) models; InstructLab container image support; rhel-ai-specific gotchas and memory budgets
• Updated skupper-model-provider/SKILL.md v1.2→v2.0: Added rhel-ai as supported remote host; updated model selection prompts, container check commands, and prerequisites for multi-host support
• Regenerated skills/index.md (48 skills)

2026-08-04 23:58

• Migrated all 48 skills to canonical frontmatter schema (skill-gen/references/skill-schema.md):
  • 14 skills: moved version: from top-level into metadata: block
  • 19 skills: added metadata.version (extracted from changelog)
  • 13 skills: normalized version format to quoted 3-part semver
  • 2 skills: removed YAML changelog: from frontmatter (goose-desktop-env-fix, goose-skupper-provider)
  • 1 skill: restructured non-standard frontmatter (goose-skupper-provider: added description:, wrapped tags/signals in metadata:)
  • 1 skill: added missing ## Changelog Markdown section (goose-skupper-provider)
• Result: 48/48 skills now have metadata.version as quoted 3-part semver

2026-08-04 23:52

• Created skills/skill-gen/references/skill-schema.md — canonical SKILL.md frontmatter schema (single source of truth for version location, format, changelog rules)
• Updated skills/skill-gen/SKILL.md v1.5.0 — template uses quoted 3-part semver ("1.0.0"); post-creation checklist requires metadata.version; writing guidance references schema doc
• Updated skills/skill-harvest/SKILL.md v1.1.0 — scaffolded template uses "1.0.0" with metadata.signals; quality checklist references canonical schema
• Updated skills/skill-index/SKILL.md v2.3.0 — added metadata.version presence validation warning; references canonical schema

2026-08-04 22:20

• Reverted local-model-ctl to v2.3 — g8b back to 16K/18 GPU layers (32K too slow for Goose prompt processing)
• Recreated model-g8b container on rhtevan-work with --ctx-size 16384 --n-gpu-layers 18 -v ...:/models:ro,z
• Updated custom_skupper.json context_limit to 16384
• Validated: 3/3 model tests passed via Skupper VAN

2026-08-04 20:24

• Updated local-model-ctl to v2.2 — g8b context increased from 16K to 32K, GPU layers reduced from 18 to 12, SELinux volume mount fix
• Recreated model-g8b container on rhtevan-work with --ctx-size 32768 --n-gpu-layers 12
• Updated skupper-linux-two-site to v2.0 — swapped site roles: localhost=edge, remote=interior/hub
• Updated skupper-model-provider to v1.1 — same role swap, firewall check moved to remote host
• Validated skupper-linux-two-site: create → link → nc test ✅ → teardown ✅

2026-08-04 19:37

• Updated skills/agentfs-setup/scripts/seed-agents-md.sh § Index Currency guardrail: strengthened skill-index mandate — explicitly prohibits ad-hoc scripts for index generation; requires load_skill and following skill instructions

2026-08-04 19:33

• Updated skills/goose-skupper-provider/SKILL.md to v1.1.0 — added setup/teardown capabilities with setup as default
• Rewrote skills/goose-skupper-provider/PROVIDER.md — merged original loaded-skill reference config into PROVIDER.md; added full Teardown section (remove custom_skupper.json + config.yaml entry)
• Regenerated skills/index.md

2026-08-04 19:20

• Updated skill-index skill (v2.1→v2.2): documented metadata block regex last-line bug fix ((?:\n|$)) and table generation blank-line bug fix
• Regenerated ~/.agents/skills/index.md — all 48/48 skills now have signals populated

2026-08-04 19:16

• Created skill goose-skupper-provider at ~/.agents/skills/goose-skupper-provider/SKILL.md — Goose custom provider for Skupper VAN model endpoint (localhost:8000)
• Created ~/.config/goose/custom_providers/custom_skupper.json — Skupper provider definition
• Added custom_skupper provider entry to ~/.config/goose/config.yaml
• Updated terminology in skupper-model-provider (v1.1→v1.2): local=edge, remote=hub/interior
• Updated terminology in skupper-linux-two-site (v2.0→v2.1): replaced my-hub/my-edge examples with neutral placeholders
• Regenerated ~/.agents/skills/index.md (48 skills)

2026-08-04 19:05

• Fixed missing signals in skills/spec-kit-setup/SKILL.md frontmatter — added 5 signal phrases
• Regenerated skills/index.md (48 skills) — spec-kit-setup Signals column now populated

2026-08-04 19:00

• Fixed fuseki skill v1.1: stop script now disables auto-start (systemctl --user disable); start script re-enables it. Prevents Fuseki from restarting on login after explicit stop.
• Updated skills/fuseki/scripts/stop-fuseki.sh — added disable step after stop
• Updated skills/fuseki/scripts/start-fuseki.sh — added enable step before start
• Updated skills/fuseki/SKILL.md changelog

2026-08-04 18:35

• Updated skupper-linux-two-site to v2.0 — swapped site roles: localhost=edge (outbound, no firewall needed), remote=interior/hub (accepts inbound links)
• Updated skupper-model-provider to v1.1 — same role swap, all scripts rewritten
• Updated link-sites.sh, test-nc.sh, teardown.sh in skupper-linux-two-site
• Updated up.sh, down.sh, status.sh in skupper-model-provider
• Validated skupper-linux-two-site with full cycle: create sites → link → nc test ✅ → teardown ✅
• Regenerated ~/.agents/skills/index.md

2026-08-04 17:38

• Created skupper-model-provider skill under ~/.agents/skills/skupper-model-provider/
• Scripts: up.sh, down.sh, test-model.sh, status.sh
• Replaces skupper-linux-two-site with integrated model runtime lifecycle
• Idempotent up/down with full Skupper VAN + model runtime management
• Regenerated ~/.agents/skills/index.md (47 skills)

2026-08-04 15:58

• Harvested knowledge bundle llm-inference-constrained-gpu at ~/.agents/knowledge/llm-inference-constrained-gpu/
• 6 concept documents: vLLM vs llama.cpp philosophy, containerized inference pattern, GPU layer tuning, AgentFS context requirements, Fedora NVIDIA gotchas, vLLM quantization pitfalls
• Updated ~/.agents/knowledge/index.md and ~/.agents/knowledge/log.md

2026-08-04 15:36

• Updated local-model-ctl to v2.1 — model-specific container names (model-g350m, model-g1b, model-g8b)
• Enables true start/stop switching without redeployment
• Renamed live container on rhtevan-work: llama-granite → model-g8b

2026-08-04 15:19

• Renamed skill granite-4.0-1b-model-ctl → local-model-ctl (v2.0)
• Added multi-model support: g350m (default), g1b, g8b with automatic engine selection
• Added llama.cpp engine support for g8b (GGUF Q4_K_M, hybrid CPU+GPU offload)
• Added list operation to show all supported models
• Added model alias parameter to start and setup operations
• Captured GPU layer tuning data (4K ctx → 25 layers, 16K ctx → 18 layers)
• Removed old granite-4.0-1b-model-ctl skill directory
• Regenerated ~/.agents/skills/index.md (46 skills)

2026-08-04 12:46

• Created skill granite-4.0-1b-model-ctl at ~/.agents/skills/granite-4.0-1b-model-ctl/SKILL.md
• Skill provides pre-check, setup, test, status, start, stop operations for containerized vLLM + Granite 4.0-1B inference server
• Captures all lessons learned from deployment session on rhtevan-work (Fedora/RPMFusion nvidia-smi absence, Podman pasta networking, bitsandbytes quantization, dtype auto-detect, tool calling flags)
• Regenerated ~/.agents/skills/index.md (46 skills)

2026-07-31 21:57

• Updated Index Currency sub-guardrail in seed-agents-md.sh template: clarified that metadata-only changes also trigger index regeneration
• Regenerated skills/index.md (45 skills)

2026-07-31 21:52

• Updated seed-agents-md.sh template script: added Guardrail #8 Anti-Daydreaming section, renumbered #9/#10, updated Quick Reference table, bumped template version to 3.8
• Bumped SKILL.md Version property from 3.7 to 3.8

2026-07-31 21:44

• Added Guardrail #8 Anti-Daydreaming to agentfs-setup skill (SKILL.md + design-spec.md) and README.md
• Renumbered Checkpoints → #9, Git Push Safety → #10 across all files
• Updated guardrail count from 9 to 10 in SKILL.md, design-spec.md, README.md
• Added v3.8 changelog entries to SKILL.md and design-spec.md

2026-07-30 14:14

• Updated goose-recipe-session-cleanup to v3.0 — Desktop sessions now eligible for cleanup on explicit request; Hidden sessions deletable; added SKIP_TRACKED safety for projects.json sessions; signal routing table for category selection; “clean all sessions” support
• Regenerated ~/.agents/skills/index.md (45 skills)

2026-07-30 13:54

• Updated goose-recipe-session-cleanup to v2.0 — report now shows 4 distinct sections (Recipe, Terminal, Hidden, Desktop reference); added optional before_date filter for targeted cleanup; improved deletion script with configurable flags; VACUUM only on bulk deletes
• Regenerated ~/.agents/skills/index.md (45 skills)

2026-07-30 13:39

• Created USER skill goose-recipe-session-cleanup at ~/.agents/skills/goose-recipe-session-cleanup/SKILL.md
• Identifies and removes orphaned session records from ad-hoc goose session run --recipe commands in SQLite DB
• Regenerated ~/.agents/skills/index.md (44 → 45 skills)

2026-07-29 10:28

• Created USER skill fedora-dns-cache at ~/.agents/skills/fedora-dns-cache/
• Added SKILL.md and scripts/setup.sh for systemd-resolved stale DNS caching
• Regenerated ~/.agents/skills/index.md (44 skills)

2026-07-28 16:45

• Updated hermes-desktop-fixes/SKILL.md: v2.1 — added fix #4 (duplicate taskbar icon on Fedora/GNOME Wayland, StartupWMClass case mismatch); renumbered electron-builder fix to #5; added .desktop file to “Files outside git” table
• Regenerated ~/.agents/skills/index.md (43 skills)

2026-07-27 18:40

• Added metadata.signals to all 43 skills (38 new, 5 existing)
• Updated skill-gen/SKILL.md: v1.4 — added metadata.signals to template and post-creation checklist, concise description guidance
• Updated skill-index/SKILL.md: v2.0 — removed Description column, Signals is primary discovery column
• Regenerated ~/.agents/skills/index.md with new format (no Description, with Signals)
• Updated design-spec.md: v3.7 — signal routing architecture, template versioning, --sync, README sync rule
• Updated seed-agents-md.sh: added Skills index to Quick Orientation, Skill signal resolution routing rule, README sync in Post-Edit step 4
• Updated README.md: skills frontmatter schema, signal routing flow, template versioning, README sync rule, 43 skills count

2026-07-27 17:20

• Updated agentfs-setup SKILL.md: v3.7 — added metadata.signals frontmatter field, --sync mode, template version stamping, project-owned section markers
• Updated agentfs-setup/references/design-spec.md: added SKILL.md Frontmatter Schema section with metadata.signals field spec
• Updated agentfs-setup/scripts/seed-agents-md.sh: slimmed Signal Routing table (11→9 rows), added template version stamp, added project-owned marker, added skill discovery note
• Updated skill-index SKILL.md: v1.9 — added Signals column extraction and display
• Added metadata.signals to: agentfs-setup, agentfs-eval, agentfs-ctx-chk, skill-harvest, okf-bundle-harvest
• Regenerated ~/.agents/skills/index.md with Signals column (43 skills, 5 with signals)

2026-07-27 15:49

• Updated seed-agents-md.sh template: refined “hey git” signal row — shortened keywords to "hey git", "git", clarified route as cd ~/.agents (or explicit path), stage, commit, trigger Guardrail #9

2026-07-27 15:10

• Created skill: fuseki — Install and manage Apache Jena Fuseki as a systemd user service
• Scripts: setup-fuseki.sh, start-fuseki.sh, stop-fuseki.sh, status-fuseki.sh
• Regenerated skills/index.md (43 skills)

2026-07-22 23:51

• Merged ## Modes into ## Directory Structure — eliminated terminology duplication (“scope” vs “mode”); Directory Structure now shows both USER and PROJECT scope trees with a Scope Boundaries note; replaced remaining “mode” references with “scope” in Getting Started
• Final section order: Scope Definitions → Getting Started → Directory Structure → Structural Guardrails → Memory Architecture → Skill Design Principles → Evaluation → License

2026-07-22 23:37

• Reordered README.md top-level sections; promoted ## Skill Design Principles from subsection under Skills (#### to ###); moved after Memory Architecture

2026-07-22 23:13

• Updated README.md — Memory Architecture: added Working Memory as a first-class memory type (runtime-managed, SESSION-scoped); added Memory Lifecycle vertical ASCII diagram showing auto-load → on-demand capture → on-demand graduation flow; all transitions human-gated by design; MEMORY.md is NOT auto-loaded but readable on-demand for recall or graduation (Option B); documented what gets loaded at session start (auto-load, first-use, on-demand tiers); acknowledged external systems (Hindsite, Cognee) as accelerators
• Updated skills/agentfs-setup/scripts/seed-agents-md.sh — Signal Routing: added “reflect” keyword row, routed to harvest triage (skill-harvest or okf-bundle-harvest)

2026-07-22 21:49

• Updated agentfs-setup skill template Guardrail #9 — added README Staleness Check (soft gate, step 4) to pre-push flow

2026-07-22 21:34

• Updated USER skill skupper-linux-two-site to v1.3 — replaced environment-specific examples (ezhang-work, rhtevan-work) with generic placeholders (my-hub, my-edge)
• Updated agentfs-setup skill template Guardrail #9 — expanded scan patterns: username leakage, SSH host aliases, local IPs, PII (email, phone, real names)
• Updated project ./AGENTS.md Guardrail #9 with same expanded scan patterns

2026-07-22 21:22

• Updated USER skill skupper-linux-two-site to v1.2 — added Invocation Example section
• Created knowledge bundle agentfs-skill-patterns at ~/.agents/knowledge/agentfs-skill-patterns/
• Added concept doc parameter-binding-pattern.md — reusable pattern for multi-parameter skill binding
• Updated ~/.agents/knowledge/index.md with new bundle entry

2026-07-22 21:11

• Updated USER skill skupper-linux-two-site to v1.1
• Added structured parameters: block in YAML frontmatter with binding-cues for semantic argument resolution
• Added argument-hint with CLI-style usage syntax
• Added Agent Binding Rules section: cue matching, missing-param prompts, default application, confirmation flow, script argument mapping table
• Regenerated ~/.agents/skills/index.md (42 skills)

2026-07-22 19:57

• Created USER skill skupper-linux-two-site at ~/.agents/skills/skupper-linux-two-site/
• SKILL.md: Two-site Skupper V2 Linux/systemd setup with parameterized site names, IPs, namespace
• Scripts: verify-prerequisites.sh, create-site.sh, link-sites.sh, test-nc.sh, teardown.sh
• Regenerated ~/.agents/skills/index.md (42 skills)

2026-07-22 19:06

• Update: Updated knowledge/skupper-v2-concepts/ — added migration (V1→V2), platform-details (Linux architecture, site bundles, multi-site, skrouterd install), and firewall rules. 6 new concept docs, 2 new sub-bundles.

2026-07-22 15:41

• Update: Updated knowledge/skupper-v2-concepts/security/ — added TLS passthrough, inter-router mTLS enforcement, and link port documentation to existing concepts.

2026-07-20 17:36

• Update: Regenerated skills/index.md (41 skills) — okf-bundle-gen timestamp updated to 2026-07-20 17:33 after v3.2 edit.

2026-07-20 17:33

• Update: Updated skills/okf-bundle-gen/SKILL.md v3.1→v3.2: Added Phase 7c (USER scope log.md entry) to fix Guardrail #5 compliance gap.

2026-07-20 17:31

• Creation: Generated OKF knowledge bundle knowledge/skupper-v2-concepts/ with 11 concept documents across 4 sub-bundles (core-concepts, security, advanced-features, operations).
• Update: Updated knowledge/index.md with new bundle entry.

2026-07-16 11:12

• Created skill crc-aap-demo-config at ~/.agents/skills/crc-aap-demo-config/SKILL.md — clone aap-demo repo, pre-flight checks against local CRC, apply protective config to prevent preset/resource override
• Regenerated ~/.agents/skills/index.md (41 skills)

2026-07-15 16:50

• Updated skills/agentfs-setup/scripts/seed-agents-md.sh: added Guardrail Quick Reference table to AGENTS.md seed template
• Updated skills/agentfs-setup/SKILL.md v3.5→v3.6: documented Guardrail Quick Reference addition

2026-07-15 15:18

• Updated skills/agentfs-setup/SKILL.md v3.4→v3.5: documented all seed template changes (Signal Routing promotion, Memory Scope rename, “hey git” signal, Post-Edit Completeness, log insertion anchor, knowledge index backtick fix)
• Updated skills/agentfs-setup/scripts/seed-agents-md.sh: knowledge index link changed to backtick format

2026-07-15 15:00

• Updated skills/agentfs-setup/scripts/seed-agents-md.sh: moved Signal Routing section to top (after Quick Orientation); removed duplicate at bottom; added log insertion anchor rule to Guardrail #5

2026-07-15 14:45

• Updated skills/agentfs-setup/scripts/seed-agents-md.sh: restructured Signal Routing as standalone section after guardrails; renamed Guardrail #2 to “Memory Scope”; added “hey git” signal

2026-07-15 14:10

• Fixed skills/litellm-vertex-ai-proxy/SKILL.md: added metadata.tags alongside vendor-nested metadata.hermes.tags so index generator can discover tags
• Regenerated skills/index.md

2026-07-15 13:57

• Updated skills/agentfs-setup/scripts/seed-agents-md.sh: added Post-Edit Completeness sub-section to Guardrail #5 in AGENTS.md seed template
• Updated skills/goose-desktop-env-fix/SKILL.md: restored rendered Changelog section at bottom of file (was dropped during v1.2 rewrite)

2026-07-15 13:00

• Updated skills/goose-desktop-env-fix/SKILL.md v1.1→v1.2: changed devbox shellenv guard from exported __DEVBOX_SHELLENV_LOADED to shell-local __devbox_shellenv_done — fixes refresh-global alias missing in devbox shell sessions while preserving fork-bomb protection

2026-07-14 19:26

• Updated skills/agentfs-setup/scripts/seed-agents-md.sh — compacted AGENTS.md template from 277→207 lines (25%): removed Resolves To column, dropped Executor/Scope columns from routing table, collapsed Skill Resolution Chain, merged Content File/Log Currency, replaced verbose Git Push Safety with 5-step list; updated sed insertion block
• Updated skills/agentfs-setup/SKILL.md v3.3→v3.4 — documented compaction
• Regenerated skills/index.md — 40 skills

2026-07-14 19:18

• Created skills/goose-agentfs-setup/references/memory-routing.md — full Goose memory routing table moved to on-demand reference file (74 lines)
• Updated skills/goose-agentfs-setup/scripts/setup.sh — replaced 76-line inline MEMORY_INSTRUCTIONS with 19-line compact stub that directs agent to load references/memory-routing.md on-demand
• Updated skills/goose-agentfs-setup/SKILL.md v1.3→v1.4 — documented on-demand memory routing architecture
• Updated ~/.config/goose/instructions.md — 89→32 lines; replaced full memory routing table with compact stub referencing on-demand file

2026-07-14 18:35

• Updated skills/agentfs-ctx-chk/SKILL.md v1.0→v1.1: added Directionality Rule (USER scope is canonical, PROJECT is fix target); updated Step 7 root cause table with Fix Direction column
• Updated skills/agentfs-ctx-chk/references/checklist.md v1.1: added Directionality Rule section before Phase 2; updated Phase 7 Decision Table

2026-07-14 18:14

• Created skills/agentfs-ctx-chk/SKILL.md v1.0 — context engineering audit skill: 8-phase methodology (inventory, redundancy, ambiguity, conflicts, effectiveness, cross-references, root causes, report+fix)
• Created skills/agentfs-ctx-chk/references/checklist.md — detailed audit checklist with patterns from first real audit
• Updated skills/goose-setup/SKILL.md: <username>→<user> (2 occurrences)
• Updated skills/skill-harvest/SKILL.md: <username>→<user> (1 occurrence)
• Regenerated skills/index.md — 40 skills indexed

2026-07-14 17:49

• Updated skills/agentfs-setup/scripts/seed-agents-md.sh: AGENTS.md template consolidated from 13 to 9 guardrails; Quick Orientation now includes SOUL.md and knowledge index rows for agent-agnostic progressive loading
• Updated skills/agentfs-setup/SKILL.md v3.2→v3.3: guardrail list and description updated to 9 guardrails
• Updated skills/agentfs-setup/references/design-spec.md: guardrail list updated, eval-driven guardrails section renumbered
• Updated skills/goose-agentfs-setup/SKILL.md: Guardrail #9→#2 cross-references
• Updated skills/goose-agentfs-setup/scripts/setup.sh: Guardrail #9→#2 cross-references
• Updated skills/okf-bundle-harvest/SKILL.md: Guardrail #8→#2 cross-references
• Updated skills/skill-gen/SKILL.md: Guardrail #10→#6, #6→#5 cross-references
• Updated skills/skill-harvest/SKILL.md: Guardrail #8→#2 cross-references; <username>→<user> placeholder
• Updated skills/goose-setup/SKILL.md: <username>→<user> placeholder (2 occurrences)
• Removed guardrails section from index.md — replaced with pointer to project-level AGENTS.md

2026-07-14 15:48

• Updated skills/agentfs-setup/scripts/seed-agents-md.sh: added entry relevancy rule to Guardrail #2 (Log Currency) in template
• Removed 4 cross-scope log entries that described PROJECT-scope changes (project AGENTS.md, project MEMORY.md)

2026-07-14 15:22

• Updated skills/agentfs-setup/SKILL.md v3.1→v3.2: AGENTS.md template now includes thirteen guardrails (added #10–#13)
• Updated skills/agentfs-setup/scripts/seed-agents-md.sh: added guardrails #10–#13 to heredoc template

2026-07-14 14:56

• Updated skills/skill-index/SKILL.md v1.7→v1.8: added name-directory consistency validation step and verification check per Agent Skills open standard

2026-07-14 14:51

• Fixed skills/crc-ols/SKILL.md name field: openshift-lightspeed-crc → crc-ols to match directory name per Agent Skills open standard
• Fixed skills/agentfs-eval/SKILL.md frontmatter: moved name and description to top-level (were nested under metadata:)
• Updated skills/skill-gen/SKILL.md v1.2→v1.3: added “Name consistency” check to post-creation checklist enforcing directory-name match rule
• Regenerated skills/index.md (39 skills, 0 empty names)

2026-07-14 13:55

• Updated skills/crc-ols/SKILL.md v2.0→v2.1: credential security hardening — replaced --from-literal with file-based secret creation, added security warnings and 401 troubleshooting guidance. Validated full add-provider + switch-default workflow against live MaaS LiteLLM endpoint.
• Regenerated skills/index.md (39 skills)

2026-07-13 17:06

• Updated skills/goose-setup/SKILL.md v1.3→v1.4: strengthened Git Push Safety from bullet list to 5-step procedural checklist

2026-07-13 16:44

• Added LICENSE file: Apache License 2.0
• Updated README.md: replaced placeholder license disclaimer with Apache 2.0 boilerplate and link to LICENSE file

2026-07-13 16:11

• Updated README.md: added “Skill Design Principles” section (non-interactive scripts, agent-as-orchestrator pattern, business process modeling) between Skills and Knowledge subsections
• Updated skills/skill-gen/SKILL.md v1.1→v1.2: added “Skill Design Principles” section before Simple Mode with same three principles

2026-07-13 15:46

• Created agentfs-eval skill — three-layer maturity assessment (structural, behavioral, semantic) with L0–L5 maturity levels
• Created scripts/agentfs-check.sh (Layer 1: 7 structural assertions)
• Created scripts/agentfs-behavior.sh (Layer 2: 5 behavioral assertions)
• Created rubrics/ directory with 4 semantic rubrics (memory-classification, reference-verification, sycophancy-detection, skill-accuracy)
• Created templates/report.md (eval output format)
• Created references/design-decisions.md (design rationale from design session)
• Updated agentfs-setup — scaffold-dotagents.sh now calls init-git.sh in PROJECT mode by default
• Updated agentfs-setup/scripts/init-git.sh — .gitignore no longer excludes .agents/memories/
• Updated agentfs-setup/references/design-spec.md — added Evaluation section, guardrails #10–12, git audit infrastructure
• Updated README.md — added Evaluation section with three-layer model, maturity levels, design decisions

2026-07-13 13:45

• Backfilled metadata.tags for 21 skills that were missing tags (including adding YAML frontmatter to hermes-headroom-provider)
• Regenerated skills/index.md — 38 skills, all with tags
• Added mandatory tags guardrail to Guardrail #6 in AGENTS.md and agentfs-setup seed template
• Updated skill-gen SKILL.md post-creation checklist — frontmatter validation now explains tags requirement and links to Guardrail #6

2026-07-13 13:33

• Updated skill-index SKILL.md v1.7 — added Tags column extraction from metadata.tags frontmatter; updated index template, verification checklist, and changelog
• Regenerated skills/index.md with Tags column — 38 skills indexed (17 with tags, 21 without)
• Updated agentfs-setup/scripts/seed-agents-md.sh — added skill resolution chain routing rule; renamed skill-creator → skill-gen in decision table

2026-07-13 11:20

• Renamed skills/skill-creator/ → skills/skill-gen/ for naming consistency with okf-bundle-gen, bash-completion-gen
• Updated SKILL.md: name, title, all internal path references, changelog (v1.1)
• Regenerated skills/index.md

2026-07-13 10:49

• Created bash-completion-gen skill under USER scope (~/.agents/skills/bash-completion-gen/SKILL.md) — generates bash completion scripts for any CLI command via systematic subcommand/option discovery, build, and validation
• Regenerated ~/.agents/skills/index.md — 38 skills indexed

2026-07-10 18:08

• agentfs-setup v3.0 — PROJECT is now the default mode; added canonical Scope Definitions section (USER=~/.agents/, PROJECT=./.agents/) to SKILL.md, AGENTS.md template, design-spec, and README; documented two USER setup paths (full clone vs minimal install); added Prerequisites section; nine guardrails (was eight); verify-setup.sh now checks for Scope Definitions in AGENTS.md; all user-facing instructions now say “ask your agent to run the skill” instead of directing users to execute bash scripts; Path B (minimal install) references the skill with USER scope hint; README reordered: Scope Definitions → Getting Started → Directory Structure → Modes → Guardrails → Memory Architecture
• Updated README.md — Scope Definitions and Getting Started moved to top; all instructions agent-centric (not bash-centric); step 3 needs no scope hint since PROJECT is default

2026-07-10 17:05

• Created skill-creator proxy skill — two modes (simple scaffold + advanced Anthropic upstream), AgentFS post-creation checklist, agent compatibility notes, fetch-upstream.sh for caching complete upstream file structure
• Updated AGENTS.md Guardrail #9 decision table — “create a skill” row now routes to skill-creator skill instead of LLM intrinsic
• Updated agentfs-setup seed template — same Guardrail #9 update

2026-07-10 16:15

• Updated goose-agentfs-setup skill (scripts/setup.sh) — replaced flat signal→action memory override with priority-based decision table: Cognee (pri 1) > Memory (pri 2) > Chat Recall (pri 3); runtime resolution via tool existence check; aligns with AGENTS.md Guardrail #9 Layer 2
• Updated goose-agentfs-setup SKILL.md — rewrote Memory Collision Avoidance section as Memory Signal Routing (Layer 2); added v1.3 changelog entry

2026-07-10 16:10

• Added Guardrail #9 (Memory Signal Routing) to agentfs-setup skill seed template (scripts/seed-agents-md.sh)
• Updated agentfs-setup design spec (references/design-spec.md) — v2.11: guardrail count 8→9, added §9 description and changelog entry
• Updated README.md — guardrail count 8→9, added Memory Signal Routing section under Memory Architecture with two-layer decision table architecture

2026-07-09 20:07

• Skill Harvest: Created crc-ctl skill v1.0 from 3 MEMORY.md entries (goofing-around project)

2026-07-09 19:55

• Created skill-harvest skill v1.0 — procedural memory-to-skill graduation (complements okf-bundle-harvest)

2026-07-09 19:41

• Updated okf-bundle-harvest SKILL.md v1.2 — added system/environment-specific non-graduation criterion

2026-07-09 17:43

• Added Memory Architecture section to ~/.agents/README.md

2026-07-09 02:09

• Updated okf-bundle-index/scripts/rebuild-index.sh — sub-bundle entries now include descriptions extracted from their index.md (first paragraph after heading, truncated to 120 chars)
• Updated okf-bundle-index/SKILL.md v1.3 — documented sub-bundle description extraction
• Rebuilt ~/.agents/knowledge/index.md with sub-bundle descriptions

2026-07-09 01:55

• Updated okf-bundle-index/SKILL.md changelog — added v1.2 entry for default path fix
• Updated okf-bundle-setup/SKILL.md changelog — added v1.1 entry for default path fix

2026-07-09 01:43

• Updated USER skill goose-agentfs-setup v1.2: added global goosehints for knowledge discovery (–hints-check, –hints-install, –hints-remove)
• Updated USER skill okf-bundle-gen v3.1: removed Phase 9 (SOUL.md pattern link injection), removed update-soul-links.sh script
• Updated USER skill okf-bundle-harvest v1.1: removed Phase 9 (SOUL.md update), removed update-soul-links.sh dependency
• Created ~/.config/goose/.goosehints with knowledge index reference for progressive loading
• Reverted SOUL.md from CONTEXT_FILE_NAMES (knowledge discovery now via global goosehints)
• Regenerated ~/.agents/skills/index.md — 34 skills indexed

2026-07-09 00:56

• Created USER skill okf-bundle-harvest: multi-project memory-to-knowledge distillation with graduation criteria, MEMORY.md pruning, OKF-compliant output
• Created ~/.agents/skills/okf-bundle-harvest/scripts/prune-memory.sh — removes graduated §-delimited entries from MEMORY.md
• Created ~/.agents/skills/okf-bundle-harvest/scripts/harvest-summary.sh — cross-project harvest candidate analysis
• Updated USER skill goose-agentfs-setup v1.1: added memory collision avoidance (–memory-check, –memory-install, –memory-remove)
• Regenerated ~/.agents/skills/index.md — 34 skills indexed

2026-07-08 23:39

• Updated README.md: removed knowledge/ from PROJECT mode tree, added scope notes (knowledge=USER-only, memories=PROJECT-only), expanded guardrails list from 4→8, updated skill count 31→33, clarified knowledge path in Getting Started
• Updated skills/agentfs-setup/references/design-spec.md: clarified prompt stacking item 4 (knowledge is USER-scoped, shared across projects)

2026-07-08 23:23

• Regenerated ~/.agents/skills/index.md — 33 skills indexed (completing workflow from stuck session 20260709_21)

2026-07-08 23:17

• Updated ~/.agents/skills/goose-desktop-env-fix/SKILL.md to v1.1: added root cause #4 (devbox fork bomb), detailed analysis section, updated .bashrc example with recursion guard, added verification checklist items, updated tags

2026-07-08 22:42

• Updated ~/.agents/skills/skill-index/SKILL.md to v1.6: clarified multi-line YAML scalar handling and improved fallback description extraction
• Regenerated ~/.agents/skills/index.md (33 skills) — all descriptions now correctly populated

2026-07-08 22:34

• Updated ~/.config/goose/instructions.md: added Path Hygiene guardrail (never use explicit home directory paths)
• Updated ~/.agents/skills/goose-setup/SKILL.md to v1.3: added Path Hygiene to example instructions and changelog

2026-07-08 22:26

• Created ~/.agents/skills/goose-desktop-env-fix/SKILL.md v1.0: captures the full Goose Desktop shell environment fix (goose-shell wrapper, .desktop entry, environment.d, bashrc restructuring)
• Regenerated ~/.agents/skills/index.md (33 skills)

2026-07-08 22:16

• Updated ~/.agents/skills/goose-setup/SKILL.md to v1.2: removed obsolete Tool Discovery instructions (replaced by goose-shell wrapper fix), updated description and examples
• Regenerated ~/.agents/skills/index.md (32 skills)

2026-07-08 18:06

• Fixed broken OKF spec links in README.md and skills/okf-bundle-setup/references/okf-spec-summary.md — now point to https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md

2026-07-08 17:49

• Updated skills/goose-setup/SKILL.md — v1.1: added Git Push Safety guardrail
• Updated ~/.config/goose/instructions.md — added Git Push Safety section

2026-07-08 17:41

• Created skills/goose-setup/SKILL.md — new skill for configuring Goose global persistent instructions and tool discovery
• Regenerated skills/index.md — 32 skills indexed
• Created ~/.config/goose/instructions.md — persistent instructions file for cross-session tool discovery
• Added GOOSE_MOIM_MESSAGE_FILE to ~/.config/goose/config.yaml

2026-07-08 14:34

• Deleted okf-bundle-merge skill — obsolete since okf-bundle-gen now writes directly to ~/.agents/knowledge/
• Regenerated skills/index.md (31 skills)

2026-07-08 14:19

• Modified okf-bundle-gen/SKILL.md: v3.0 — bundle root changed to ~/.agents/knowledge/ (user-level), removed project-local staging, SOUL.md links use absolute paths, memory scan PROJECT-only
• Modified okf-bundle-merge/SKILL.md: marked OBSOLETE — no longer needed since okf-bundle-gen writes directly to user-level knowledge

2026-07-08 13:38

• Recreated agentfs-setup/SKILL.md — was missing; reflects v2.10 with memory redesign (8 guardrails, knowledge USER-only, memories PROJECT-only)
• Modified agentfs-setup/scripts/seed-agents-md.sh: added guardrail §8 (Memory Scope) with NL-signal routing and graduation path
• Modified agentfs-setup/scripts/verify-setup.sh: knowledge checks USER-only, MEMORY.md template updated to “Project Experiences”, PROJECT mode checks for stale knowledge/
• Modified agentfs-setup/scripts/scaffold-dotagents.sh: knowledge/ USER-only, MEMORY.md “Experiences” template (done in prior session)
• Modified agentfs-setup/references/design-spec.md: removed knowledge from PROJECT tree, 8 guardrails, updated layer descriptions
• Modified agentfs-profile/SKILL.md: v1.8 — updated description, removed knowledge references
• Modified agentfs-profile/scripts/create-profile.sh: MEMORY.md template updated to “Project Experiences” with scope/NL-signal guidance

2026-07-08 10:11

• Memory: Added OKF non-concept type guideline to user-level MEMORY.md — distinguishes concept bundles from reference dataset bundles, documents valid use of Dataset/Script/Ground Truth types for companion data metadata

2026-07-08 09:46

• Memory: Added two OKF guidelines to user-level MEMORY.md — (1) keep bundle roots clean of concept files, (2) distill true concepts rather than raw documentation
• Reorganize: Moved claude-compat-analysis.md from user-global knowledge root into sub-bundle agentfs-claude-compat/

2026-07-08 08:54

• Created knowledge bundle knowledge/headroom-openai-compression-analysis/ with 3 concept documents: problem-analysis (root cause of zero compression), configuration-history (v1→v2 timeline), options-assessment (4 options, recommended passthrough + watch)
• Updated knowledge/index.md and knowledge/log.md

2026-07-07 16:52

• Created new skill goose-agentfs-setup — configures Goose CONTEXT_FILE_NAMES for cross-agent context file discovery (CLAUDE.md, .cursorrules, .windsurfrules)
• Updated agentfs-setup skill v2.9 — added Cross-Agent Context Discovery guardrail (§7) to AGENTS.md template in seed-agents-md.sh

2026-07-07 16:08

• Updated skills/headroom-litellm-proxy/SKILL.md to v1.1 — removed --lossless, added --target-ratio 0.5 and --intercept-tool-results; added Compression Tuning section, Flags NOT to Use section, expanded health/stats verification and troubleshooting
• Regenerated skills/index.md via skill-index (31 skills)

2026-07-07 16:04

• Updated skills/agentfs-setup/SKILL.md to v2.8 — added guardrails §6 bullets: mandatory skill-index invocation, scope-aware log.md updates; clarified skill-index requirement in Maintaining the Layers section
• Updated skills/agentfs-setup/scripts/seed-agents-md.sh AGENTS.md template with new §6 guardrail bullets
• Regenerated skills/index.md via skill-index (31 skills)

2026-07-07 15:58

• Created memories/MEMORY.md at USER level (~/.agents/memories/) with cross-project agent workflow guardrail: always run skill-index after modifying any skill
• Regenerated skills/index.md via skill-index skill (31 skills indexed)

2026-07-07 15:54

• Updated skills/headroom-proxy-status/SKILL.md to v1.1 — added Kompress ML, target ratio, uncompressed reasons to report format; expanded compression stats extraction fields; added compression troubleshooting section; removed --lossless from key flags example
• Updated skills/index.md timestamp for headroom-proxy-status

2026-07-07 00:00

• Removed USER skill claude-skills-link — redundant; Goose natively scans .claude/skills/ at both project and global scope
• Regenerated ~/.agents/skills/index.md — 31 skills indexed

2026-07-06 23:28

• Updated skills/claude-skills-link/SKILL.md — confirmed v2.0 (no content change); verified skill execution with CWD test: symlink creation, idempotency, stale cleanup, and PROJECT skill index generation all pass

2026-07-06 22:01

• Created USER skill hermes-headroom-provider at ~/.agents/skills/hermes-headroom-provider/SKILL.md — configure Hermes Agent to use the local Headroom proxy as its custom LLM provider
• Regenerated ~/.agents/skills/index.md — 31 skills indexed

2026-07-06 21:48

• Created skill headroom-litellm-proxy at ~/.agents/skills/headroom-litellm-proxy/SKILL.md — Headroom installation and systemd setup chained to LiteLLM
• Refactored skill goose-headroom-provider (v2.0) — now covers Goose custom provider config only; installation/systemd content moved to headroom-litellm-proxy
• Regenerated ~/.agents/skills/index.md — 30 skills indexed

2026-07-06 21:39

• Created skill goose-headroom-provider at ~/.agents/skills/goose-headroom-provider/SKILL.md — configure Goose to use the Headroom context-optimization proxy as a custom provider
• Regenerated ~/.agents/skills/index.md — 29 skills indexed

2026-07-06 21:32

• Created skill headroom-proxy-status at ~/.agents/skills/headroom-proxy-status/SKILL.md — check health, config, and runtime status of the local Headroom context-optimization proxy
• Regenerated ~/.agents/skills/index.md — 28 skills indexed

2026-07-06 20:06

• Updated skills/goose-maas-provider/SKILL.md to v1.3 — Goose Desktop v1.41 is incompatible with MaaS for tool-calling tasks (fails under all tested configurations: streaming on/off, toolshim on/off); CLI with GOOSE_TOOLSHIM: true is the only working approach; updated Desktop section, troubleshooting

2026-07-06 20:00

• Updated skills/goose-maas-provider/SKILL.md to v1.2 — added GOOSE_TOOLSHIM: true as required config (smaller models strip namespace prefixes from tool names); added supports_streaming: false as required for Desktop (streaming responses lost due to goose Desktop bug); documented Desktop vs CLI behavioral differences; updated checklist, troubleshooting, recovery

2026-07-06 19:39

• Updated skills/goose-maas-provider/SKILL.md to v1.1 — reasoning models (gpt-oss-120b, qwen3-14b, deepseek-r1-*) are fundamentally incompatible with Goose v1.41 streaming parser; changed default model to llama-scout-17b; added model compatibility matrix; updated recovery script, checklist, troubleshooting

2026-07-06 19:25

• Created skills/goose-maas-provider/SKILL.md v1.0 — new dedicated skill for MaaS (remote LiteLLM) provider setup; covers API key keyring storage, critical reasoning model fixes (reasoning: false, preserves_thinking: false), documented failure modes with evidence from real sessions, diagnostic tests, recovery script
• Updated skills/goose-litellm-provider/SKILL.md to v1.2 — removed all MaaS-related content (moved to goose-maas-provider); restored as local-proxy-only skill; updated description, tags, and related_skills to reference new skill
• Updated skills/index.md — added goose-maas-provider, bumped count to 27, refreshed goose-litellm-provider description

2026-07-06 19:14

• Updated skills/goose-litellm-provider/SKILL.md to v1.1 — added MaaS remote provider configuration, reasoning model gotcha (reasoning: true for thinking models), API key GNOME Keyring storage, available model discovery, expanded troubleshooting and recovery procedures
• Updated skills/index.md — refreshed description and timestamp for goose-litellm-provider

2026-07-06 18:04

• Created skills/goose-litellm-provider/SKILL.md v1.0 — skill to configure Goose with local LiteLLM proxy as ‘RedHat’ custom provider, includes reference JSON, config.yaml entries, recovery script, and troubleshooting
• Updated skills/index.md — added goose-litellm-provider entry, bumped count to 26

2026-07-06 14:38

• Updated skills/litellm-vertex-ai-proxy/SKILL.md v1.1 — made agent-agnostic by removing Hermes-specific Step 8, updated description and troubleshooting
• Updated skills/index.md — refreshed description for litellm-vertex-ai-proxy

2026-07-06 12:27

• Updated agentfs-setup skill to fully align all scripts, flags, and documentation from the legacy ‘SYSTEM’ terminology to ‘USER’ mode.

2026-07-06 11:37

• Strengthened guardrail §6 (Index Currency) in AGENTS.md and seed-agents-md.sh: skills/index.md must now be regenerated after any content modification to skill files (SKILL.md, scripts, references), not just structural changes (create/rename/move/delete)

2026-07-06 11:31

• Renamed AgentFS mode label SYSTEM → USER across all skills, scripts, design specs, AGENTS.md, and seed templates
• Updated script variable names: AGENTS_SKILLS_SYSTEM → AGENTS_SKILLS_USER, AGENTS_SKILLS_SYSTEM_EXPANDED → AGENTS_SKILLS_USER_EXPANDED
• Preserved SYSTEM_RESERVED_* kubelet variables in crc-post-setup-config (unrelated to AgentFS modes)
• Historical log.md entries left unchanged per append-only guardrail

2026-07-06 11:18

• Renamed agent-fs-profile → agentfs-profile and agent-fs-setup → agentfs-setup for consistent agentfs naming
• Renamed 6 skills to replace -configuration suffix with -config: crc-coo-config, crc-nad-dynamic-plugin-config, crc-nmstate-config, crc-noo-config, crc-ovn-frr-metallb-config, crc-post-setup-config
• Updated all internal references across SKILL.md files, scripts, index.md, and project-level files

2026-07-06 11:00

• Updated hermes-agentfs-setup to v1.1 — added PROJECT scope support (--project, --undo-project, --list)
• PROJECT scope registers a project’s .agents/skills/ as an absolute path in skills.external_dirs (per-project action)
• Regenerated skills/index.md (25 skills)

2026-07-06 10:54

• Created hermes-agentfs-setup skill — configures Hermes Agent to discover AgentFS SYSTEM skills from ~/.agents/skills/ via skills.external_dirs
• Includes scripts/setup.sh with --check, --undo, and idempotent setup modes
• Regenerated skills/index.md (25 skills) with Python-based YAML frontmatter parser to fix folded-scalar description truncation

2026-07-01 00:07

• Added total count to skills/index.md and profiles/index.md summary lines (e.g., > 24 skills | Sorted by…).
• Updated scaffold-dotagents.sh, verify-setup.sh, create-profile.sh, and skill-index/SKILL.md to emit/maintain the count.
• Regenerated ~/.agents/skills/index.md (24 skills).

2026-07-01 00:00

• Added missing YAML frontmatter (name + description) to crc-ovn-frr-metallb-config/SKILL.md; was the only skill without frontmatter, causing empty description in skills/index.md.
• Regenerated skills/index.md.

2026-06-30 23:54

• Added ## Changelog section to 19 SYSTEM skills that were missing it; all 24 skills now have consistent | Updated | Change | tables with YYYY-MM-DD HH:MM timestamps.

2026-06-30 23:49

• Expanded guardrail §2 (Log Currency): explicit SYSTEM/PROJECT/sub-bundle scope coverage; mandatory logging when skills or concept files change; standardized log.md format (title, comment, heading precision, entry style).
• Fixed okf-bundle-setup/scripts/scaffold-bundle.sh and okf-bundle-gen/scripts/merge-log-entry.sh to use YYYY-MM-DD HH:MM timestamps, <!-- Append-only --> comment, and - entry style.
• Standardized this file to use consistent format.

2026-06-30 23:36

• Updated guardrail §3 (Content File Currency): Changelog tables now require YYYY-MM-DD HH:MM timestamps and Updated column header.
• Updated Changelog tables in 6 files: agentfs-setup/SKILL.md, agentfs-setup/references/design-spec.md, agentfs-profile/SKILL.md, skill-index/SKILL.md, skill-merge/SKILL.md, okf-bundle-gen/SKILL.md.

2026-06-30 23:31

• Renamed index column Added → Updated across all skills/index.md and profiles/index.md templates and live files.
• Increased timestamp precision to YYYY-MM-DD HH:MM in all index.md seeds, log.md seeds, and script date calls.
• Updated guardrails §2 and §6 to use timestamp headings.
• Regenerated ~/.agents/skills/index.md (30 skills, Updated column, YYYY-MM-DD HH:MM).

2026-06-30 23:16

• Added Index Currency guardrail (§6) to AGENTS.md template in seed-agents-md.sh.
• Updated profiles/index.md schema: Identity + Memories + Updated columns, sorted newest-first.
• Updated skills/index.md schema: Updated column, sorted newest-first.
• Expanded profiles/ narrative in agentfs-setup/SKILL.md and design-spec.md with dual-purpose (multi-agent hub + ROLE-based specialization) and Hermes OOTB compatibility.
• Updated create-profile.sh to insert entries newest-first with memories link.
• Reinforced mandatory skills/index.md update in skill-merge/SKILL.md.
• Regenerated ~/.agents/skills/index.md.

2026-06-30 18:30

• Added profiles/index.md to scaffold and verify scripts; all profiles/ links now point to profiles/index.md; all memories/ links now point to memories/MEMORY.md.

2026-06-30 17:45

• All mutating scripts (seed-agents-md.sh, init-speckit.sh, rename-agent-context.sh) now append entries to .agents/log.md per the Log Currency guardrail.

2026-06-30 17:30

• Idempotent re-run: agent detects existing .agents/ and skips creation phases; verify-setup.sh --fix repairs missing files/dirs without overwriting; link integrity checks; profile completeness checks; skills/index.md seeded instead of .gitkeep.

2026-06-30 16:30

• Fixed index.md link convention: all relative links now use ./ prefix for consistent rendering across GitHub, VS Code, and other markdown viewers.

2026-06-30 16:00

• Fixed rename-agent-context.sh sed bug: replaced sed -i c\ with awk block replacement for SPECKIT marker merging.

2026-06-30 15:30

• Added Agent Profiles table to AGENTS.md; seed-agents-md.sh creates default row and retrofits existing files; create-profile.sh auto-registers new profiles.

2026-06-30 14:00

• v2.0 redesign of agentfs-setup: USER → SYSTEM mode rename; memory/ → memories/; roles/ → profiles/; added SOUL.md, USER.md, MEMORY.md; removed constitution.md (Spec-kit owns it); multi-agent collaboration design; prompt stacking order.
• Created companion skill agentfs-profile: scaffolds named agent profiles under .agents/profiles/ with SOUL.md + memories/.

2026-06-26 22:00

• Fixed verify-setup.sh to use opt-in --with-git / --with-spec flags instead of auto-detecting git/spec-kit on disk.

2026-06-26 14:00

• Initialized .agents/ directory structure (mode: system). m). ).