[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7C4VS9NTZP6P29ZZR3DM70S":3,"skill-same:01M7C4VS9NTZP6P29ZZR3DM70S":106},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":14,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":6,"updatedAt":31,"install":32,"resources":45},"01M7C4VS9NTZP6P29ZZR3DM70S","architectural-conformance-audit",null,"Pre-R0 diff of implementation vs SOTA research. Triggers: before R0, architectural audit. NOT per-PR review.","EtanHey\u002Fgolems","EtanHey","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F93338119?v=4","01M7C4SKC9HJC3B9TAX0D9MZ1E","Etan Heyman","01M7C4SKCT59Z49TTN4WWDDEYD",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: architectural-conformance-audit\ndescription: \"Pre-R0 diff of implementation vs SOTA research. Triggers: before R0, architectural audit. NOT per-PR review.\"\nargument-hint: \"\u003Csprint-name | research-output-path>\"\ndisable-model-invocation: false\n---\n\n# Skill: architectural-conformance-audit\n\n> **Wave evidence (severity 10, 4 corroborating digests):** AP1 root cause was *architectural ground-truth blindness*. Researcher's SOTA output at idx [380] LITERALLY cited Letta-on-FastAPI as a counter-example, yet R0→R5 sprint hooks optimized within the daemon assumption anyway. ~35h misdirected work + 2.5h explicit correction. PR #312 fixed it in code (FastAPI daemon deleted, socket-direct CLI added, merged May 22 11:39Z). See `pain-points\u002F_consolidated.md` Pattern 4 for the full chronology. **This skill prevents AP1-class recurrence procedurally.**\n\n---\n\n## WHEN TO ACTIVATE\n\nThis skill fires at a **specific sprint moment:** the kickoff of a new R0→R5 sprint OR a `large-plan` phase that builds atop a research output (whichever fires first in a given build arc). It is **NOT** a per-PR check; it is a **per-sprint gate**.\n\n### Tier 1 — Mandatory triggers (always invoke)\n- New sprint kickoff that references a prior researcher SOTA output (`research\u002F*.md` or similar).\n- New sprint that modifies \u002F extends architecture established in an earlier PR (PR-α, PR-β, etc.).\n- After any researcher dispatch whose output explicitly cites counter-examples.\n- When a sprint is rebooting after an architectural correction (e.g., resuming from an AP1-class failure).\n\n### Tier 2 — Recommended triggers (invoke unless explicitly skipped)\n- New skill development that wraps or extends an MCP\u002Fdaemon\u002Fservice.\n- When R5 evaluator score \u003C parent-objective threshold (suggests local-optimum trap).\n- When implementation language \u002F framework choice was inherited (not first-principles).\n\n### Tier 3 — Manual invocation\n- User asks \"is the architecture correct?\" \u002F \"verify against research\" \u002F \"before we go further, let's audit\"\n\n---\n\n## THE AUDIT CONTRACT\n\nThe audit produces **three artifacts** before any R0 work begins:\n\n### Artifact 1 — SOTA Excerpt (verbatim)\n\nFor each cited research output:\n- Pull the SOTA output file (`research\u002F\u003Ctopic>-research.md` or equivalent).\n- Extract every section that mentions architecture, framework choice, OR counter-examples.\n- Quote the relevant passages **verbatim** with source indices.\n\nOutput → `docs.local\u002Faudits\u002F\u003Csprint>\u002F\u003Cdate>-sota-excerpt.md`.\n\n### Artifact 2 — Implementation Map (concrete)\n\nFor the current implementation:\n- List every architectural primitive (daemon, service, MCP, queue, socket, HTTP layer, persistence backend).\n- For each primitive: cite file path + line range that defines it.\n- For each primitive: note whether it was first-principles-derived in this sprint, inherited from a prior PR, or scaffolded by an external generator.\n\nOutput → `docs.local\u002Faudits\u002F\u003Csprint>\u002F\u003Cdate>-impl-map.md`.\n\n### Artifact 3 — Conformance Verdict\n\nFor each `(SOTA excerpt × impl primitive)` pair:\n- **MATCH** — implementation follows SOTA recommendation.\n- **DIVERGE — UNJUSTIFIED** — implementation contradicts SOTA without documented rationale.\n- **DIVERGE — JUSTIFIED** — implementation contradicts SOTA with documented rationale (cite the rationale).\n- **N\u002FA** — SOTA silent on this primitive.\n\n**The gate rule:** if ANY `DIVERGE — UNJUSTIFIED` exists → **SPRINT R0 IS BLOCKED** until the divergence is either reconciled (impl changed) or justified (rationale documented + `brain_store`'d).\n\nOutput → `docs.local\u002Faudits\u002F\u003Csprint>\u002F\u003Cdate>-conformance-verdict.md`.\n\n---\n\n## WORKFLOW\n\n### Step 1 — Locate the SOTA research output\n\nCanonical scan order (most-recent on tie via `ls -lat`):\n\n```bash\nls -lat research\u002F*.md 2>\u002Fdev\u002Fnull\nls -lat docs.local\u002Fresearch\u002F*.md 2>\u002Fdev\u002Fnull\nls -lat $ORCHESTRATOR_ROOT\u002Fdocs.local\u002Fresearch\u002F*.md 2>\u002Fdev\u002Fnull\nls -lat $ORCHESTRATOR_ROOT\u002Fdocs.local\u002Fhandoffs\u002F**\u002Fresearch\u002F*.md 2>\u002Fdev\u002Fnull\n```\n\nIf multiple SOTA outputs conflict, the audit MUST list both and **require Etan to pick** the canonical source before proceeding. **Do NOT auto-canonicalize by date** — staleness is the AP1 root cause.\n\n### Step 2 — Extract SOTA architectural claims\n\nRead the SOTA output in full. For each architectural claim, extract:\n- The claim itself.\n- Source index (line number, section name, or `[N]` reference).\n- Whether the claim is positive (\"use X\"), negative (\"avoid Y\"), or comparative (\"X over Y because Z\").\n- Any cited **counter-examples** (the FastAPI case: SOTA cited Letta-on-FastAPI as the *thing not to do*).\n\nIf extraction grows large or repetitive, follow `workflows\u002Fextract-claims.md`.\n\n### Step 3 — Map the implementation\n\nPer architectural primitive in scope:\n\n```bash\nfind src packages -name \"daemon.py\" -o -name \"service.py\" -o -name \"*.service.ts\" | xargs wc -l\ngrep -rn \"^from fastapi\\|^import fastapi\\|from socketio\\|import asyncio\" src packages 2>\u002Fdev\u002Fnull\ngrep -rn \"mcp__server__\\|@server\\.tool\\|@server\\.resource\" src packages 2>\u002Fdev\u002Fnull\n```\n\nFor each primitive found: file path + line range; direct vs. transitive (inherited) authorship; first-principles vs. scaffolded. If mapping grows large, follow `workflows\u002Fmap-impl.md`.\n\n### Step 4 — Diff\n\nFor each `(SOTA claim, impl primitive)`:\n- SOTA says \"use X\" + impl uses X → **MATCH**.\n- SOTA says \"use X\" + impl uses Y → **DIVERGE** (look for documented rationale via `git log -p --follow \u003Cimpl-file>` and `brain_search \"\u003Cprimitive> chosen over \u003Calternative>\"`).\n- SOTA cites X as **counter-example** + impl uses X → **DIVERGE — UNJUSTIFIED** unless explicitly documented (this IS the AP1 pattern; treat as severity-10).\n- SOTA silent → **N\u002FA**.\n\n### Step 5 — Gate decision\n\n- ANY `DIVERGE — UNJUSTIFIED` → **R0 BLOCKED**. Surface to Etan + sprint LEAD with verbatim SOTA cite + impl divergence + proposed reconciliation path (change impl OR document rationale).\n- ALL `MATCH | DIVERGE — JUSTIFIED | N\u002FA` → **R0 CLEARED**. `brain_store` the audit verdict at importance ≥8 with tags `[architectural-audit, \u003Csprint>, R0-cleared]`. Composes with `\u002Fbrain-store-fallback` for transport failures.\n\n**There is no `--override` flag.** Per gen-8 decision: document or change impl — those are the only two paths. Footgun risk too high. If override is later deemed necessary, it must `brain_store` at importance 10 with tag `[audit-override]` + verbatim rationale.\n\n---\n\n## ANTI-PATTERNS\n\n### AP1 — Reading research output but not diffing against impl\nThe historical case. Researcher's output at idx [380] existed; multiple agents READ it (researcher, R5 evaluator, Codex workers, Cursor auditors). Nobody DIFFED it against `daemon.py`. The audit MUST produce a literal pairwise diff, not \"I read the research.\"\n\n### AP2 — Skipping audit because \"we already audited this last sprint\"\nArchitectural assumptions decay. Each sprint must re-audit. If nothing changed, the audit is fast (re-cite the prior verdict). If something changed, the audit catches it.\n\n### AP3 — Treating \"the researcher said use X\" as gospel\nSOTA outputs themselves can be wrong or stale. The audit's job is **conformance** — does impl match SOTA? — NOT validation that SOTA is correct. If SOTA is wrong, that's a separate research-correction sprint (and worth flagging).\n\n### AP4 — Confusing this with code review\nCode review reads the diff and checks craft. This audit reads the architecture and checks first-principles alignment. They compose; this fires BEFORE R0, code review fires DURING R3.\n\n### AP5 — Letting the audit become a paperwork exercise\nIf audits routinely come back MATCH for everything with no friction, suspect false-pass. The audit's value comes from catching real DIVERGE cases. If 3 consecutive sprints show all-MATCH, run a meta-audit on the auditor (was it reading the right SOTA? was it checking the right primitives?).\n\n---\n\n## COMPOSITION\n\n- **Research artifacts** — Claude Desktop\u002FGemini\u002Fweb research outputs feed the SOTA output this audit reads.\n- **`\u002Fnever-fabricate`** — audit verdicts cite specific file paths + line ranges; never-fabricate enforces that those citations are real. This skill is the architectural-level fabrication guard; never-fabricate is the file-level guard.\n- **`\u002Fbrain-store-fallback`** (SHIP-2, merged) — audit verdicts get stored at importance ≥8; brain-store-fallback handles transport failures during storage. Mandatory composition.\n- **`\u002Fcoderabbit`** — composes downstream; coderabbit fires per-PR, this skill fires per-sprint.\n- **`\u002Fplan-council`** assumption check ([references\u002Fassumption-check.md](..\u002Fplan-council\u002Freferences\u002Fassumption-check.md), formerly `\u002Fplan-validate`): the general assumption check; this skill is the architecture-specific one.\n- **`\u002Flarge-plan\u002Fworkflows\u002Fscaffold`** — the audit is a pre-R0 step in scaffold.md so it isn't skipped by oversight.\n- **`\u002Forc`** — orc invokes this skill at sprint kickoff when Tier-1 triggers fire.\n\n---\n\n## EVALS (summary — full scenarios in `evals\u002Fevals.json`)\n\n| # | Scenario | Without skill (baseline) | With skill (target) | Assertion |\n|---|---|---|---|---|\n| 1 | SOTA recommends socket-direct; impl uses FastAPI HTTP (AP1 re-creation) | R5 graded local-optimum 8.85\u002F10; mismatch slips through | DIVERGE — UNJUSTIFIED; R0 blocks until reconciled | Verdict file lists FastAPI primitive with counter-example cite |\n| 2 | SOTA recommends X; impl uses X | No-op | MATCH; R0 clears | Verdict shows MATCH; brain_store fires |\n| 3 | SOTA silent on primitive Z; impl uses Z | No-op | N\u002FA for Z (doesn't block) | Verdict file shows N\u002FA for Z |\n| 4 | Two SOTA outputs conflict | Stale SOTA used silently | Audit lists both; gate held pending Etan pick | Both files referenced; no auto-canonicalize |\n| 5 | Mixed MATCH\u002FDIVERGE across multiple primitives | Slips through | Lists all; ANY UNJUSTIFIED blocks | R0 blocked even if 9\u002F10 MATCH |\n\n**The AP1 re-creation eval (scenario 1) is load-bearing.** Fixture: real-world excerpt from the May 2026 brainlayer-readpath research output + synthetic FastAPI daemon snippet mimicking the deleted PR-α `daemon.py`.\n\n**Smoke test (retrospective):** run against the current brainlayer codebase POST-PR #312. Expected: MATCH on socket-direct primitive (the audit retrospectively confirms the fix held). Note: smoke is **read-only** against `$HOME\u002FGits\u002Fbrainlayer\u002F` per cross-repo constraint.\n\n---\n\n## DEFINITION OF DONE (per-invocation)\n\n- [ ] Audit produces 3 artifacts in `docs.local\u002Faudits\u002F\u003Csprint>\u002F\u003Cdate>-{sota-excerpt,impl-map,conformance-verdict}.md`.\n- [ ] R0 gate is enforced: ANY UNJUSTIFIED DIVERGE blocks proceeding.\n- [ ] Verdict is `brain_store`'d at importance ≥8 with tags `[architectural-audit, \u003Csprint>, R0-cleared|R0-blocked]`. Use `\u002Fbrain-store-fallback` if BL transport fails.\n- [ ] Skill composes cleanly with research artifacts, `\u002Fcoderabbit`, `\u002Flarge-plan\u002Fscaffold`.\n- [ ] AP1 re-creation eval passes (scenario 1 fixture).\n\n---\n\n## R5 EVALUATOR EXTENSION — OUT OF SCOPE HERE\n\nPer consolidated.md Pattern 4 system-fix:\n> \"R5 evaluator skill change: must include 'goal-envelope check' — score against the original parent objective, not the sprint's local optimization.\"\n\nThis skill does **NOT** modify the R5 evaluator (it's a separate skill change, tracked as a future SHIP-9 candidate). This skill surfaces the parent-objective in the audit so the R5 evaluator has a referenceable target. The two compose; they ship independently.\n\n---\n\n## ESCALATION\n\n- Multiple SOTA candidates conflict → list both, require Etan to canonicalize. Do NOT auto-pick by date.\n- AP1-class DIVERGE found → R0 BLOCKED message MUST include verbatim SOTA cite, file path of divergent impl, and explicit \"change impl OR document rationale\" path. No silent block.\n- BrainLayer transport fails during verdict storage → fall back via `\u002Fbrain-store-fallback` and report the fallback file path in the verdict.\n","# Skill: architectural-conformance-audit\n\n> **Wave evidence (severity 10, 4 corroborating digests):** AP1 root cause was *architectural ground-truth blindness*. Researcher's SOTA output at idx [380] LITERALLY cited Letta-on-FastAPI as a counter-example, yet R0→R5 sprint hooks optimized within the daemon assumption anyway. ~35h misdirected work + 2.5h explicit correction. PR #312 fixed it in code (FastAPI daemon deleted, socket-direct CLI added, merged May 22 11:39Z). See `pain-points\u002F_consolidated.md` Pattern 4 for the full chronology. **This skill prevents AP1-class recurrence procedurally.**\n\n---\n\n## WHEN TO ACTIVATE\n\nThis skill fires at a **specific sprint moment:** the kickoff of a new R0→R5 sprint OR a `large-plan` phase that builds atop a research output (whichever fires first in a given build arc). It is **NOT** a per-PR check; it is a **per-sprint gate**.\n\n### Tier 1 — Mandatory triggers (always invoke)\n- New sprint kickoff that references a prior researcher SOTA output (`research\u002F*.md` or similar).\n- New sprint that modifies \u002F extends architecture established in an earlier PR (PR-α, PR-β, etc.).\n- After any researcher dispatch whose output explicitly cites counter-examples.\n- When a sprint is rebooting after an architectural correction (e.g., resuming from an AP1-class failure).\n\n### Tier 2 — Recommended triggers (invoke unless explicitly skipped)\n- New skill development that wraps or extends an MCP\u002Fdaemon\u002Fservice.\n- When R5 evaluator score \u003C parent-objective threshold (suggests local-optimum trap).\n- When implementation language \u002F framework choice was inherited (not first-principles).\n\n### Tier 3 — Manual invocation\n- User asks \"is the architecture correct?\" \u002F \"verify against research\" \u002F \"before we go further, let's audit\"\n\n---\n\n## THE AUDIT CONTRACT\n\nThe audit produces **three artifacts** before any R0 work begins:\n\n### Artifact 1 — SOTA Excerpt (verbatim)\n\nFor each cited research output:\n- Pull the SOTA output file (`research\u002F\u003Ctopic>-research.md` or equivalent).\n- Extract every section that mentions architecture, framework choice, OR counter-examples.\n- Quote the relevant passages **verbatim** with source indices.\n\nOutput → `docs.local\u002Faudits\u002F\u003Csprint>\u002F\u003Cdate>-sota-excerpt.md`.\n\n### Artifact 2 — Implementation Map (concrete)\n\nFor the current implementation:\n- List every architectural primitive (daemon, service, MCP, queue, socket, HTTP layer, persistence backend).\n- For each primitive: cite file path + line range that defines it.\n- For each primitive: note whether it was first-principles-derived in this sprint, inherited from a prior PR, or scaffolded by an external generator.\n\nOutput → `docs.local\u002Faudits\u002F\u003Csprint>\u002F\u003Cdate>-impl-map.md`.\n\n### Artifact 3 — Conformance Verdict\n\nFor each `(SOTA excerpt × impl primitive)` pair:\n- **MATCH** — implementation follows SOTA recommendation.\n- **DIVERGE — UNJUSTIFIED** — implementation contradicts SOTA without documented rationale.\n- **DIVERGE — JUSTIFIED** — implementation contradicts SOTA with documented rationale (cite the rationale).\n- **N\u002FA** — SOTA silent on this primitive.\n\n**The gate rule:** if ANY `DIVERGE — UNJUSTIFIED` exists → **SPRINT R0 IS BLOCKED** until the divergence is either reconciled (impl changed) or justified (rationale documented + `brain_store`'d).\n\nOutput → `docs.local\u002Faudits\u002F\u003Csprint>\u002F\u003Cdate>-conformance-verdict.md`.\n\n---\n\n## WORKFLOW\n\n### Step 1 — Locate the SOTA research output\n\nCanonical scan order (most-recent on tie via `ls -lat`):\n\n```bash\nls -lat research\u002F*.md 2>\u002Fdev\u002Fnull\nls -lat docs.local\u002Fresearch\u002F*.md 2>\u002Fdev\u002Fnull\nls -lat $ORCHESTRATOR_ROOT\u002Fdocs.local\u002Fresearch\u002F*.md 2>\u002Fdev\u002Fnull\nls -lat $ORCHESTRATOR_ROOT\u002Fdocs.local\u002Fhandoffs\u002F**\u002Fresearch\u002F*.md 2>\u002Fdev\u002Fnull\n```\n\nIf multiple SOTA outputs conflict, the audit MUST list both and **require Etan to pick** the canonical source before proceeding. **Do NOT auto-canonicalize by date** — staleness is the AP1 root cause.\n\n### Step 2 — Extract SOTA architectural claims\n\nRead the SOTA output in full. For each architectural claim, extract:\n- The claim itself.\n- Source index (line number, section name, or `[N]` reference).\n- Whether the claim is positive (\"use X\"), negative (\"avoid Y\"), or comparative (\"X over Y because Z\").\n- Any cited **counter-examples** (the FastAPI case: SOTA cited Letta-on-FastAPI as the *thing not to do*).\n\nIf extraction grows large or repetitive, follow `workflows\u002Fextract-claims.md`.\n\n### Step 3 — Map the implementation\n\nPer architectural primitive in scope:\n\n```bash\nfind src packages -name \"daemon.py\" -o -name \"service.py\" -o -name \"*.service.ts\" | xargs wc -l\ngrep -rn \"^from fastapi\\|^import fastapi\\|from socketio\\|import asyncio\" src packages 2>\u002Fdev\u002Fnull\ngrep -rn \"mcp__server__\\|@server\\.tool\\|@server\\.resource\" src packages 2>\u002Fdev\u002Fnull\n```\n\nFor each primitive found: file path + line range; direct vs. transitive (inherited) authorship; first-principles vs. scaffolded. If mapping grows large, follow `workflows\u002Fmap-impl.md`.\n\n### Step 4 — Diff\n\nFor each `(SOTA claim, impl primitive)`:\n- SOTA says \"use X\" + impl uses X → **MATCH**.\n- SOTA says \"use X\" + impl uses Y → **DIVERGE** (look for documented rationale via `git log -p --follow \u003Cimpl-file>` and `brain_search \"\u003Cprimitive> chosen over \u003Calternative>\"`).\n- SOTA cites X as **counter-example** + impl uses X → **DIVERGE — UNJUSTIFIED** unless explicitly documented (this IS the AP1 pattern; treat as severity-10).\n- SOTA silent → **N\u002FA**.\n\n### Step 5 — Gate decision\n\n- ANY `DIVERGE — UNJUSTIFIED` → **R0 BLOCKED**. Surface to Etan + sprint LEAD with verbatim SOTA cite + impl divergence + proposed reconciliation path (change impl OR document rationale).\n- ALL `MATCH | DIVERGE — JUSTIFIED | N\u002FA` → **R0 CLEARED**. `brain_store` the audit verdict at importance ≥8 with tags `[architectural-audit, \u003Csprint>, R0-cleared]`. Composes with `\u002Fbrain-store-fallback` for transport failures.\n\n**There is no `--override` flag.** Per gen-8 decision: document or change impl — those are the only two paths. Footgun risk too high. If override is later deemed necessary, it must `brain_store` at importance 10 with tag `[audit-override]` + verbatim rationale.\n\n---\n\n## ANTI-PATTERNS\n\n### AP1 — Reading research output but not diffing against impl\nThe historical case. Researcher's output at idx [380] existed; multiple agents READ it (researcher, R5 evaluator, Codex workers, Cursor auditors). Nobody DIFFED it against `daemon.py`. The audit MUST produce a literal pairwise diff, not \"I read the research.\"\n\n### AP2 — Skipping audit because \"we already audited this last sprint\"\nArchitectural assumptions decay. Each sprint must re-audit. If nothing changed, the audit is fast (re-cite the prior verdict). If something changed, the audit catches it.\n\n### AP3 — Treating \"the researcher said use X\" as gospel\nSOTA outputs themselves can be wrong or stale. The audit's job is **conformance** — does impl match SOTA? — NOT validation that SOTA is correct. If SOTA is wrong, that's a separate research-correction sprint (and worth flagging).\n\n### AP4 — Confusing this with code review\nCode review reads the diff and checks craft. This audit reads the architecture and checks first-principles alignment. They compose; this fires BEFORE R0, code review fires DURING R3.\n\n### AP5 — Letting the audit become a paperwork exercise\nIf audits routinely come back MATCH for everything with no friction, suspect false-pass. The audit's value comes from catching real DIVERGE cases. If 3 consecutive sprints show all-MATCH, run a meta-audit on the auditor (was it reading the right SOTA? was it checking the right primitives?).\n\n---\n\n## COMPOSITION\n\n- **Research artifacts** — Claude Desktop\u002FGemini\u002Fweb research outputs feed the SOTA output this audit reads.\n- **`\u002Fnever-fabricate`** — audit verdicts cite specific file paths + line ranges; never-fabricate enforces that those citations are real. This skill is the architectural-level fabrication guard; never-fabricate is the file-level guard.\n- **`\u002Fbrain-store-fallback`** (SHIP-2, merged) — audit verdicts get stored at importance ≥8; brain-store-fallback handles transport failures during storage. Mandatory composition.\n- **`\u002Fcoderabbit`** — composes downstream; coderabbit fires per-PR, this skill fires per-sprint.\n- **`\u002Fplan-council`** assumption check ([references\u002Fassumption-check.md](..\u002Fplan-council\u002Freferences\u002Fassumption-check.md), formerly `\u002Fplan-validate`): the general assumption check; this skill is the architecture-specific one.\n- **`\u002Flarge-plan\u002Fworkflows\u002Fscaffold`** — the audit is a pre-R0 step in scaffold.md so it isn't skipped by oversight.\n- **`\u002Forc`** — orc invokes this skill at sprint kickoff when Tier-1 triggers fire.\n\n---\n\n## EVALS (summary — full scenarios in `evals\u002Fevals.json`)\n\n| # | Scenario | Without skill (baseline) | With skill (target) | Assertion |\n|---|---|---|---|---|\n| 1 | SOTA recommends socket-direct; impl uses FastAPI HTTP (AP1 re-creation) | R5 graded local-optimum 8.85\u002F10; mismatch slips through | DIVERGE — UNJUSTIFIED; R0 blocks until reconciled | Verdict file lists FastAPI primitive with counter-example cite |\n| 2 | SOTA recommends X; impl uses X | No-op | MATCH; R0 clears | Verdict shows MATCH; brain_store fires |\n| 3 | SOTA silent on primitive Z; impl uses Z | No-op | N\u002FA for Z (doesn't block) | Verdict file shows N\u002FA for Z |\n| 4 | Two SOTA outputs conflict | Stale SOTA used silently | Audit lists both; gate held pending Etan pick | Both files referenced; no auto-canonicalize |\n| 5 | Mixed MATCH\u002FDIVERGE across multiple primitives | Slips through | Lists all; ANY UNJUSTIFIED blocks | R0 blocked even if 9\u002F10 MATCH |\n\n**The AP1 re-creation eval (scenario 1) is load-bearing.** Fixture: real-world excerpt from the May 2026 brainlayer-readpath research output + synthetic FastAPI daemon snippet mimicking the deleted PR-α `daemon.py`.\n\n**Smoke test (retrospective):** run against the current brainlayer codebase POST-PR #312. Expected: MATCH on socket-direct primitive (the audit retrospectively confirms the fix held). Note: smoke is **read-only** against `$HOME\u002FGits\u002Fbrainlayer\u002F` per cross-repo constraint.\n\n---\n\n## DEFINITION OF DONE (per-invocation)\n\n- [ ] Audit produces 3 artifacts in `docs.local\u002Faudits\u002F\u003Csprint>\u002F\u003Cdate>-{sota-excerpt,impl-map,conformance-verdict}.md`.\n- [ ] R0 gate is enforced: ANY UNJUSTIFIED DIVERGE blocks proceeding.\n- [ ] Verdict is `brain_store`'d at importance ≥8 with tags `[architectural-audit, \u003Csprint>, R0-cleared|R0-blocked]`. Use `\u002Fbrain-store-fallback` if BL transport fails.\n- [ ] Skill composes cleanly with research artifacts, `\u002Fcoderabbit`, `\u002Flarge-plan\u002Fscaffold`.\n- [ ] AP1 re-creation eval passes (scenario 1 fixture).\n\n---\n\n## R5 EVALUATOR EXTENSION — OUT OF SCOPE HERE\n\nPer consolidated.md Pattern 4 system-fix:\n> \"R5 evaluator skill change: must include 'goal-envelope check' — score against the original parent objective, not the sprint's local optimization.\"\n\nThis skill does **NOT** modify the R5 evaluator (it's a separate skill change, tracked as a future SHIP-9 candidate). This skill surfaces the parent-objective in the audit so the R5 evaluator has a referenceable target. The two compose; they ship independently.\n\n---\n\n## ESCALATION\n\n- Multiple SOTA candidates conflict → list both, require Etan to canonicalize. Do NOT auto-pick by date.\n- AP1-class DIVERGE found → R0 BLOCKED message MUST include verbatim SOTA cite, file path of divergent impl, and explicit \"change impl OR document rationale\" path. No silent block.\n- BrainLayer transport fails during verdict storage → fall back via `\u002Fbrain-store-fallback` and report the fallback file path in the verdict.\n",{"name":5,"description":7,"argument-hint":24,"disable-model-invocation":20},"\u003Csprint-name | research-output-path>","master","3400bc971628f00ab7bf1063993477e211a5508a","skills\u002Fgolem-powers\u002Farchitectural-conformance-audit","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems","https:\u002F\u002Fgithub.com\u002FEtanHey","etanheyman.com","2026-11-15 03:05:08",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add EtanHey\u002Fgolems --skill architectural-conformance-audit",{"tool":38,"command":39},"bunx","bunx skills add EtanHey\u002Fgolems --skill architectural-conformance-audit",{"tool":41,"command":42},"pnpm","pnpm dlx skills add EtanHey\u002Fgolems --skill architectural-conformance-audit","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002FSKILL.md",true,[46,52,58,62,66,70,74,78,82,86,90,94,98,102],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"evals\u002Fevals.json","REFERENCE","参考文档",8263,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Fevals.json",{"path":53,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":56,"sourceHtmlUrl":57},"evals\u002Ffixtures\u002Fap1-reproduction\u002Fimpl-daemon.py.snippet","OTHER","其他",1565,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fap1-reproduction\u002Fimpl-daemon.py.snippet",{"path":59,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":60,"sourceHtmlUrl":61},"evals\u002Ffixtures\u002Fap1-reproduction\u002Fsota-excerpt.md",2699,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fap1-reproduction\u002Fsota-excerpt.md",{"path":63,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":64,"sourceHtmlUrl":65},"evals\u002Ffixtures\u002Fconflict-case\u002Fsota-A.md",614,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fconflict-case\u002Fsota-A.md",{"path":67,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":68,"sourceHtmlUrl":69},"evals\u002Ffixtures\u002Fconflict-case\u002Fsota-B.md",851,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fconflict-case\u002Fsota-B.md",{"path":71,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":72,"sourceHtmlUrl":73},"evals\u002Ffixtures\u002Fmatch-case\u002Fimpl-socket.py.snippet",906,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fmatch-case\u002Fimpl-socket.py.snippet",{"path":75,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":76,"sourceHtmlUrl":77},"evals\u002Ffixtures\u002Fmatch-case\u002Fsota-excerpt.md",475,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fmatch-case\u002Fsota-excerpt.md",{"path":79,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":80,"sourceHtmlUrl":81},"evals\u002Ffixtures\u002Fmixed-case\u002Fimpl-summary.md",1730,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fmixed-case\u002Fimpl-summary.md",{"path":83,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":84,"sourceHtmlUrl":85},"evals\u002Ffixtures\u002Fmixed-case\u002Fsota-excerpt.md",1032,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fmixed-case\u002Fsota-excerpt.md",{"path":87,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":88,"sourceHtmlUrl":89},"evals\u002Ffixtures\u002Fsilent-case\u002Fimpl-persistence.py.snippet",920,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fsilent-case\u002Fimpl-persistence.py.snippet",{"path":91,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":92,"sourceHtmlUrl":93},"evals\u002Ffixtures\u002Fsilent-case\u002Fsota-excerpt.md",582,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Ffixtures\u002Fsilent-case\u002Fsota-excerpt.md",{"path":95,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":96,"sourceHtmlUrl":97},"evals\u002Fgreen-with-skill.log",4906,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Fgreen-with-skill.log",{"path":99,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":100,"sourceHtmlUrl":101},"evals\u002Fred-baseline.log",4496,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Fred-baseline.log",{"path":103,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":104,"sourceHtmlUrl":105},"evals\u002Fsmoke-real-env.log",3745,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Farchitectural-conformance-audit\u002Fevals\u002Fsmoke-real-env.log",[107,114,121,128,135,142],{"skillNo":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":113,"publishedAt":19},"01M7C53JEF6A6SS20Q0D2YJYM8","yash","Push dependency bugs upstream. Triggers: dependency bug, workaround, monkey-patch, fork\u002Ffix. NOT app bugs.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fyash\u002FSKILL.md",[],[],{"skillNo":115,"name":116,"nameZh":6,"description":117,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":118,"categories":119,"keywords":120,"publishedAt":19},"01M7C53EPMNSZKHK47VY7WQDP0","whats-new","Diff Claude\u002FCodex\u002FCursor\u002FWispr\u002Fagy changelogs vs local setup. Triggers: whats new, changelog, version bump.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fwhats-new\u002FSKILL.md",[],[],{"skillNo":122,"name":123,"nameZh":6,"description":124,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":125,"categories":126,"keywords":127,"publishedAt":19},"01M7C53C4TVQ9G7AB3X57DACPB","weave","Orchestrator convergence: mine JSONLs, cite findings, route actions. Triggers: weave, run weave.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fweave\u002FSKILL.md",[],[],{"skillNo":129,"name":130,"nameZh":6,"description":131,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":132,"categories":133,"keywords":134,"publishedAt":19},"01M7C5342EDDN7YH53KKW511V2","unslop","Cut AI tells while keeping every receipt, number, path, and hedge intact. Triggers: unslop, cut the slop, wall of text, too verbose, rambly, make this readable. NOT for delivery cadence (see i-have-adhd), channel choice, or artifact length governance.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Funslop\u002FSKILL.md",[],[],{"skillNo":136,"name":137,"nameZh":6,"description":138,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":139,"categories":140,"keywords":141,"publishedAt":19},"01M7C52M0CTDW5DDP4452FJV0G","tmp-block","Guard denying durable writes to \u002Ftmp, \u002Fprivate\u002Ftmp, \u002Fvar\u002Ffolders, $TMPDIR; enforces \u003Crepo>\u002F.worktrees\u002F\u003Cname>. Triggers: tmp guard, TMP-BLOCK deny, WORKTREE-CONVENTION deny, bypass ledger. NOT for storage policy.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Ftmp-block\u002FSKILL.md",[],[],{"skillNo":143,"name":144,"nameZh":6,"description":145,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":146,"categories":147,"keywords":148,"publishedAt":19},"01M7C52HN4Q9E1Z74Z11K23E2H","t3-visuals","Show\u002Fvisualize\u002Fchart\u002Fdiagram\u002Fcompare\u002Fstatus board in T3 with compact interactive HTML. Not standalone dashboards.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Ft3-visuals\u002FSKILL.md",[],[]]