@chrono-meta/fh-gate 1.0.3 → 1.2.0

package/plugins/fh-meta/skills/sim-conductor/SKILL_detail.md

@@ -0,0 +1,337 @@
+---
+name: sim-conductor-detail
+description: Detail reference for sim-conductor — bash scripts, multi-team panel, Area B baseline methods, D-consumer execution, Area E execution, report template, PR bash, Path B. Load when executing a specific step.
+load: on-demand
+---
+
+# sim-conductor — Detail Reference
+
+> Load when executing a specific step. SKILL.md contains triggers, target profile analysis, precondition table, area summaries with persona routing, human gate, and Done When.
+
+---
+
+## §Step0-Bash — Environment Sync Scripts
+
+```bash
+HARNESS_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
+SIM_CLONE=~/sim/your-hub-sim/forge-harness-via-clone
+[ -d "$SIM_CLONE" ] && (cd "$SIM_CLONE" && git pull origin main) || SIM_CLONE="$HARNESS_ROOT"
+
+# Report storage path
+REPORT_DIR=$(find "$(dirname "$HARNESS_ROOT")" -maxdepth 2 -name "tracks" -type d 2>/dev/null | head -1)
+REPORT_DIR="${REPORT_DIR:-$HARNESS_ROOT/tracks}/_meta"
+mkdir -p "$REPORT_DIR"
+```
+
+**Area B frequency check bash**:
+```bash
+ls "$REPORT_DIR"/sim_*_area_B*.md 2>/dev/null | sort | tail -1
+# If date within 7 days → output warning + stop
+```
+
+---
+
+## §PersonaDiscovery — Plugin-Recommender Integration
+
+### Query format for plugin-recommender
+
+When a GAP is detected for a high-weight perspective:
+
+```
+/plugin-recommender query:
+  context: "sim-conductor persona gap"
+  needed_perspective: [perspective name, e.g. "security-auditor"]
+  artifact_type: [from profile, e.g. "Python auth code"]
+  audience: [from profile]
+  constraints: "must work as Agent sub-agent dispatch (no interactive CLI)"
+```
+
+plugin-recommender returns: ranked candidates with name · source (FH native / marketplace / external) · install command · token cost estimate · fit rationale.
+
+### Gap resolution decision tree
+
+```
+GAP detected for [perspective X]:
+  Is it high-weight (profile score ≥ 0.7)?
+    YES → present plugin-recommender result → user: install / skip / substitute
+          install: run install cmd → register → continue Step 0.3
+          skip: auto-fill with ② built-in fallback, flag as "degraded coverage"
+          substitute: user names alternative agent → use it
+    NO  → auto-fill with ② built-in fallback, no prompt
+```
+
+### Persona map by artifact type (typical optimal)
+
+| Artifact type | Optimal personas | Likely GAP (not in FH native) |
+|---|---|---|
+| SKILL.md / governance doc | challenger · newcomer · governance-expert | governance-expert → query plugin-recommender |
+| README / marketing copy | newcomer · devil-advocate · domain-expert | domain-expert (field-specific) → query |
+| Python / JS code | challenger/Code · power-user · security-auditor | security-auditor → query if auth/data |
+| Auth / security-sensitive code | security-auditor · challenger/Code · power-user | security-auditor → block if GAP (high-weight) |
+| Design doc + citations | challenger · domain-expert · skeptic | domain-expert (field-specific) → query |
+
+### Degraded coverage flag
+
+When any high-weight persona is substituted with built-in fallback (user chose "skip"):
+```
+⚠️ Degraded coverage: [perspective X] substituted with built-in [fallback name].
+   Risk: [what this perspective would have caught that built-in may miss].
+   Recommendation: install [plugin name] before external publish.
+```
+
+Degraded coverage persists in Step 3 report under "Persona composition used."
+
+---
+
+## §Profile-Examples — Target Profile Worked Examples
+
+### Example 1: SKILL.md (governance doc)
+
+```
+Target Profile:
+  artifact_type: SKILL.md (governance doc)
+  audience: mixed (internal FH users + potential external installers)
+  claim_density: medium (2 stated benefits, 1 "autonomous" claim)
+  risk_level: medium (not yet externally published)
+
+Recommendation:
+  Areas: B (internal meta audit) + D-skill (cold-start validation)
+  Persona composition: challenger (high — claim verification), newcomer (medium — onboarding), governance-expert (medium)
+  Scale: Minimum (3)
+  Prerequisites: none (not yet external publish)
+```
+
+### Example 2: README (pre-publish)
+
+```
+Target Profile:
+  artifact_type: README (external entry point)
+  audience: external installer (first-time user)
+  claim_density: high (5+ capability claims, "automatic", "zero config")
+  risk_level: high (marketplace / external publish imminent)
+
+Recommendation:
+  Areas: A (external user perspective — primary) + C (naming gap scan)
+  Persona composition: newcomer (high), devil-advocate (high — claim density), power-user (medium)
+  Scale: Extended (4–5)
+  Prerequisites: steel-quench REQUIRED before Area A proceeds
+```
+
+### Example 3: Bash script / Python code
+
+```
+Target Profile:
+  artifact_type: bash script (50 lines, 3 external commands)
+  audience: internal developer
+  claim_density: low (technical spec only)
+  risk_level: low (pre-PR, internal)
+
+Recommendation:
+  Areas: D-code (primary)
+  Persona composition: challenger/Code (edge cases, security surface), power-user (performance)
+  Scale: Minimum (2 — third persona adds minimal value for bash)
+  Prerequisites: none
+```
+
+### Example 4: Design doc with arXiv citations
+
+```
+Target Profile:
+  artifact_type: design doc (contains arXiv references + code blocks)
+  audience: internal team
+  claim_density: high (3 arXiv claims, 2 quantitative benchmarks cited)
+  risk_level: medium
+  novelty: high (references recent paper)
+
+Recommendation:
+  Areas: D-code + source-grounding-audit (quantitative claims)
+  Persona composition: challenger (claim-evidence), domain-expert (arXiv validity)
+  Scale: Minimum (3)
+  Prerequisites: source-grounding-audit recommended (novelty + citations)
+```
+
+---
+
+## §MultiTeam — Multi-Team Panel Execution
+
+### Pre-entry confirmation dialog
+
+```
+Area A — Multi-Team Mode available.
+Detected external CLIs: [list or "none"]
+Estimated additional token cost: ~2K–5K per external team (billed to that CLI's quota).
+Run multi-team? (a) Full panel  (b) Claude sub-agents only  (c) Skip to Area B
+```
+
+### Team → Persona mapping
+
+| Team | CLI | Personas | Dispatch method |
+|---|---|---|---|
+| T0 Claude | Agent sub-agent | hub-persona-auditor · challenger · domain-expert | Agent() call |
+| T1 Gemini | `gemini` pipe | newcomer · power-user · devil | `echo PROMPT \| gemini` |
+| T2 Copilot | `gh copilot suggest` | devil · domain-expert | `gh copilot suggest -t shell` |
+| T3 Ollama | `ollama run` | devil | `ollama run llama3 PROMPT` |
+| T4 Codex | `npx @openai/codex exec` | devil · edge-case-hunter | `echo PROMPT \| npx @openai/codex exec -m gpt-5 -` |
+
+### CLI detection bash
+
+```bash
+# Detect available external CLIs
+AVAILABLE_CLIS=""
+command -v gemini   &>/dev/null && AVAILABLE_CLIS="$AVAILABLE_CLIS gemini"
+command -v gh       &>/dev/null && gh copilot --version &>/dev/null && AVAILABLE_CLIS="$AVAILABLE_CLIS copilot"
+command -v ollama   &>/dev/null && AVAILABLE_CLIS="$AVAILABLE_CLIS ollama"
+command -v npx      &>/dev/null && npx @openai/codex --version &>/dev/null 2>&1 && AVAILABLE_CLIS="$AVAILABLE_CLIS codex"
+echo "Available: ${AVAILABLE_CLIS:-none}"
+```
+
+### Cross-team synthesis format
+
+```
+Cross-Team Synthesis:
+| Issue | Teams | Grade |
+|---|---|---|
+| [issue description] | [T0+T1 / T1+T2 / etc.] | S-confirmed / A-confirmed |
+
+Claude blind spots (external-only findings):
+- [T1 found X, T0 missed → Claude blind spot: ...]
+```
+
+**Path B fallback**: No external CLIs → cross-session `claude --print` as T0 cold-read variant.
+
+---
+
+## §AreaB-Baseline — External Baseline Injection Methods
+
+Structural methods to reduce self-reference risk in Area B:
+
+1. **Regular devil attacks**: Area B once/month + devil attack (challenger) once/quarter. Route devil → defense results directly into SKILL.md via steel-quench handoff after Area B ends.
+2. **Direct external user validation**: Non-owner attempts install + invocation → collect reactions. (cascade β validated: first autonomous external run confirmed.)
+3. **steel-quench integration**: After Area B ends, hand off challenger findings to `/steel-quench` for deeper adversarial review + SKILL.md inscription.
+4. **Dual validation principle**: Internal validation (Area B) alone is insufficient — minimized only when combined with external install reaction collection or cross-model validation.
+
+**Dispatch template for Area B parallel**:
+```
+Agent A (hub-persona-auditor): target=[files], task="3+ persona simulation, 4-axis review, 3-tier suggestions"
+Agent B (persona-innovator, Mode I): target=[files], task="naming gaps + structural gaps + 3-5 candidates"
+→ both complete → 
+Agent C (challenger, artifact_type="SKILL"): input=[A output + B output], task="what was missed? angles U2, D3, U5"
+```
+
+---
+
+## §AreaD-Consumer — D-consumer Execution Detail
+
+### D-session — Session Card Cold-Start
+
+Auto-detect target (when `--target` not specified):
+MEMORY.md → check `reference_next_session_starter.md` → if absent: search latest session-starter in `tracks/_meta/` or `memory/`.
+
+Provide only that file to consumer agent (no context). Ask:
+1. The most important thing to do today
+2. Current work-in-progress status
+3. Events/deadlines that must not be missed
+4. Parts that are unclear or lack context
+
+**Verdict**: Top priority matches + context reconstructable → F / Context reconstructed but gaps → P / Wrong top priority or key item missing → B.
+B/P items: fix directly in session card + commit.
+
+### D-skill — Skill Trigger Utterance Consumer Simulation
+
+Provide only one SKILL.md to consumer agent (no context, no SKILL_detail.md).
+Assume first trigger sentence from Trigger Phrases was input → attempt to complete from Step 1.
+Record: where it stalled, why, whether core task completes, what needs fixing.
+
+**Verdict**: Core task completes → F / Partial completion (stalled mid-step) → P / Blocked at Step 1 → B.
+
+### D-memory — Memory File Cold-Start Validation
+
+Provide only one memory file to consumer agent. Check:
+① When should this be applied?
+② Can current validity be determined from this file alone?
+③ Can the behavior guideline be executed from this file alone?
+④ Any stale or contradictory information?
+
+**Verdict**: ③ executable + ④ none → F / Some unclear parts → P / Stale/contradiction found → B.
+
+---
+
+## §AreaE-Detail — Area E Execution
+
+### E-1 — Domain Expert Objection
+
+Agent (subagent_type="challenger", artifact_type="Design"): false positive patterns (clearly wrong judgments) + false negative patterns (should-have-caught-but-didn't). Angles: U3 (evidence grounding), D4 (failure mode coverage).
+
+Findings format: `[judgment type · pattern · root cause · fix direction]`
+
+### E-2 — Practitioner Confusion
+
+Agent (Newcomer brief): confusing items, fix suggestions more awkward than original, classification criteria consistency breaks.
+
+Findings format: `[item · confusion cause · improvement direction]`
+
+### E-3 — Pattern Structuring (direct execution)
+
+Integrate E-1/E-2 → group false positives by root cause → assign pattern name → pinpoint fix location → M/S/R classification → fix target code/prompt → commit.
+
+---
+
+## §Report-Format — Report Template
+
+```markdown
+---
+name: [date] sim-conductor — Area [X]
+type: simulation-report
+date: YYYY-MM-DD
+areas: [A|B|C|D|E|all]
+target_profile: [artifact_type | audience | risk_level]
+m_count: N
+s_count: N
+r_count: N
+---
+
+## Target Profile
+artifact_type: [type] · audience: [internal|external|mixed] · risk_level: [low|medium|high]
+
+## M-tier ([N] items)
+| # | Issue | Location | Prescription |
+...
+
+## S-tier ([N] items)
+| # | Issue | Location | Priority |
+...
+
+## R-tier ([N] items)
+...
+
+## Emerging asset candidates
+...
+
+## Persona composition used
+[list of personas used + which were task-derived vs built-in]
+```
+
+---
+
+## §PR-Bash — PR Creation Scripts
+
+```bash
+cd "$HARNESS_ROOT"
+BRANCH="fix/sim-$(date +%Y%m%d)-m-tier"
+git checkout -b "$BRANCH"
+# [process M-tier items]
+git add -p
+git commit -m "fix(sim-conductor): resolve M-tier findings from simulation YYYY-MM-DD"
+git push -u origin "$BRANCH"
+# PR creation requires explicit user request per CLAUDE.md PR principle
+```
+
+---
+
+## §PathB-Detail — External Environment Fallback
+
+When `~/sim/` is absent:
+- Step 0: `SIM_CLONE="$HARNESS_ROOT"` (skip git clone)
+- Logical isolation only: inject "Treat the following as if you have no prior development context" into each Agent prompt
+- **Area B only recommended**: Area A/C technically executable but physical isolation not guaranteed (direct path references). Logical isolation via prompt directive is the only available control.
+- Area D/E: fully functional (target-file-based, no sim clone needed)

package/plugins/fh-meta/skills/skill-splitter/SKILL.md

@@ -0,0 +1,126 @@
+---
+name: skill-splitter
+description: Splits an over-loaded SKILL.md into always-loaded (SKILL.md) + on-demand (SKILL_detail.md) layers using a governance-semantic criterion — not length, but when the content is needed. Connects the two files with imperative pointers. Based on paper §9.5 Protocol-Priority Split pattern. Diagnoses, classifies, splits, and verifies in one pass.
+user-invocable: true
+allowed-tools: ["Read", "Write", "Edit", "Bash", "Grep", "Glob"]
+model: sonnet
+---
+
+# skill-splitter — SKILL.md Governance-Semantic Split
+
+> A SKILL.md that does everything in one file is not simple — it is unscoped.
+> The goal is two files each smaller than the original, not one file and its appendix.
+
+## Trigger Phrases
+
+| Phrase | Situation |
+|---|---|
+| "this skill is getting too long", "trim the skill", "split this SKILL.md" | Direct split request |
+| "context-doctor flagged this skill", "SKILL.md is bloated" | Post-diagnosis split |
+| "I can't see the key parts", "too much detail in the skill file" | Readability problem |
+| "separate the bash from the logic", "move the templates out" | Structural refactor request |
+| `/skill-splitter` | Explicit invocation |
+
+---
+
+## Core Principle — Governance-Semantic Criterion
+
+**The split criterion is NOT length. It is: when does this content need to be in memory?**
+
+| Layer | When needed | Examples |
+|---|---|---|
+| **SKILL.md (always-loaded)** | Every invocation — session boundaries, trigger recognition, step overview, decision tables | Triggers, principles, step names + criteria, key decision tables, Done When |
+| **SKILL_detail.md (on-demand)** | Only when executing a specific step | Bash scripts, format templates, edge cases, step-by-step execution detail |
+
+**Behavioral rules stay in SKILL.md** — if a rule governs *what counts* (e.g. "these patterns = closed"), it is behavioral logic and must be always-loaded regardless of length.
+
+**One test**: *"If a consumer agent had only SKILL.md, could they recognize the trigger, understand the full step sequence, and make the key decisions?"* → Yes = correct split. No = something behavioral is missing from SKILL.md.
+
+---
+
+## Step Overview
+
+```
+Step 1 — Diagnose
+    Read target SKILL.md → classify every section as Always / On-demand / Ambiguous
+    → Produce classification table
+
+Step 2 — Draft SKILL.md (trimmed)
+    Keep: triggers · principles · step names + one-line criteria · decision tables · Done When
+    Remove: bash scripts · format templates · multi-step execution detail · edge case catalogs
+    Add: imperative pointer for each removed section → SKILL_detail.md §SectionName
+
+Step 3 — Draft SKILL_detail.md
+    One ## §SectionName header per pointer in SKILL.md
+    Move removed content under its section
+    Front-matter: name, description, load: on-demand
+
+Step 4 — Verify
+    source-grounding-audit: every §pointer in SKILL.md resolves to ## §SectionName in SKILL_detail.md
+    sim-conductor Area D-skill: consumer agent with SKILL.md only → must reach grade F
+    → Any pointer mismatch or grade P/B → fix before commit
+```
+
+> **Detail**: See `SKILL_detail.md §Verification-Checklist` — pre-commit checklist table (8 checks) — read when running Step 4 verification.
+
+> **Detail**: See `SKILL_detail.md §Split-Execution` — step-by-step trimming procedure, SKILL_detail.md front-matter format, orphan §section check — read when executing Steps 2–3.
+
+> **Detail**: See `SKILL_detail.md §Classification` — ambiguous content decision algorithm, behavioral-vs-implementation test, 12 annotated examples — read when unsure which layer a section belongs to.
+
+---
+
+## Imperative Pointer Format (required)
+
+Pointers must be **imperative** (not advisory). The difference:
+
+| Form | Risk |
+|---|---|
+| Advisory: `"see SKILL_detail.md for details"` | Consumer agent may skip |
+| **Imperative**: `"> **Detail**: See \`SKILL_detail.md §SectionName\` — [what's there] — read when [specific condition]."` | Consumer agent loads on trigger |
+
+Every removed section must have exactly one imperative pointer at the point of removal in SKILL.md.
+
+> **Detail**: See `SKILL_detail.md §Pointers` — pointer format variants, multi-pointer blocks, pointer placement rules — read when writing pointers.
+
+---
+
+## Target Selection
+
+Run on a SKILL.md when **any one** of:
+- context-doctor flags the file as oversized
+- SKILL.md exceeds ~300 lines
+- sim-conductor Area D-skill grades P or B (cold-start fails)
+- Consumer agent reports "could not proceed — too much detail before step 1"
+
+**Not a target**: SKILL.md that has no bash scripts, format templates, or multi-step execution detail. Splitting a file with only behavioral content adds structure without governance value.
+
+---
+
+## Connected Skills
+
+| Situation | Skill |
+|---|---|
+| Diagnose which SKILL.md files are candidates | `/context-doctor` or `/harness-doctor` |
+| Verify §pointer grounding after split | `/source-grounding-audit` |
+| Verify cold-start still works after split | `/sim-conductor D skill <name>` |
+| Check new SKILL_detail.md for phantom claims | `/source-grounding-audit` |
+| Adversarial review of the split result | `/steel-quench` |
+
+---
+
+## Done When
+
+```
+Step 1 classification table produced
++ SKILL.md trimmed: triggers · principles · step overview · decision tables · Done When retained
++ SKILL.md has imperative pointer for every removed section (> **Detail**: See SKILL_detail.md §X)
++ SKILL_detail.md created: ## §SectionName header for every pointer in SKILL.md
++ source-grounding-audit: 0 phantoms (all §pointers resolve)
+  → Fallback (skill unavailable): run §Verification-Checklist manually from SKILL_detail.md
++ sim-conductor Area D-skill: grade F (consumer completes core task from SKILL.md alone)
+  → Fallback (skill unavailable): manually confirm "trigger → step overview → key decision → Done When" all present in SKILL.md
+```
+
+**Not done**: SKILL_detail.md has a §section with no pointer from SKILL.md (orphan section).
+**Not done**: Consumer agent grade P/B after split — a behavioral rule was moved to SKILL_detail.md when it should have stayed.
+**Not done**: Any behavioral rule present only in SKILL_detail.md.

package/plugins/fh-meta/skills/skill-splitter/SKILL_detail.md

@@ -0,0 +1,185 @@
+---
+name: skill-splitter-detail
+description: Detail reference for skill-splitter — content classification algorithm, pointer format variants, verification checklist. Load when executing a specific step.
+load: on-demand
+---
+
+# skill-splitter — Detail Reference
+
+> Load when executing a specific step. SKILL.md contains the core principle, step overview, pointer format requirement, and Done When.
+
+---
+
+## §Classification — Content Classification Algorithm
+
+### Decision algorithm (per section)
+
+```
+For each section in target SKILL.md:
+
+1. Is this content needed to RECOGNIZE the trigger or understand WHAT the skill does?
+   → YES → Always-loaded (keep in SKILL.md)
+
+2. Is this content a BEHAVIORAL RULE (governs what counts as X, what is allowed, what is forbidden)?
+   → YES → Always-loaded (keep in SKILL.md, regardless of length)
+   → "what counts as closed", "auto-allowed vs prohibited", "never touch" rules are behavioral
+
+3. Is this content needed only when EXECUTING a specific step?
+   → YES → On-demand (move to SKILL_detail.md)
+
+4. Is this a DECISION TABLE (compact lookup, not step-by-step procedure)?
+   → YES → Always-loaded (tables are high information density, keep)
+   → Exception: table > 20 rows → candidate for SKILL_detail.md
+
+5. Is this BASH CODE or a FORMAT TEMPLATE?
+   → YES → On-demand (move to SKILL_detail.md)
+   → Exception: one-liner illustration (≤1 line) used to identify a concept → may stay in SKILL.md
+```
+
+### 12 annotated examples
+
+| Content type | Layer | Reason |
+|---|---|---|
+| Trigger phrases table | Always | Needed to recognize invocation |
+| Step names + one-line criteria | Always | Overview — consumer needs to see the full sequence |
+| Collision type routing table (3 rows) | Always | Decision table, compact |
+| "natural-language close" patterns | Always | Behavioral rule — governs what counts as closed |
+| "Memory curator safety: only INDEX-ORPHAN auto-allowed" | Always | Behavioral rule — prohibition |
+| Multi-step synthesizer verdict matrix (5 rows) | Always | Decision table |
+| Bash script for STALE detection | On-demand | Execution detail, step-specific |
+| Agent() call format with prompt template | On-demand | Execution detail |
+| 20-row edge case catalog | On-demand | Reference, not needed every invocation |
+| Output format template (12 fields) | On-demand | Execution detail |
+| "Done When" block | Always | Completion criterion — always needed |
+| Connected skills table | Always | Navigation aid, compact |
+
+### Ambiguous content test
+
+When unsure, apply the **D-skill cold-start test mentally**:
+
+> *"If a consumer agent had only SKILL.md and typed the trigger phrase, would they need this content in the first 2 steps?"*
+> - YES → Always-loaded
+> - NO → On-demand
+
+Behavioral rules always pass this test (even if rarely triggered, the consumer needs to know the constraint exists).
+
+---
+
+## §Pointers — Pointer Format Variants
+
+### Standard pointer (single section reference)
+
+```markdown
+> **Detail**: See `SKILL_detail.md §SectionName` — [one-line description of what's there] — read when [specific condition that triggers need].
+```
+
+### Multi-item pointer (several related sections)
+
+```markdown
+> **Detail**: See `SKILL_detail.md §SectionName-A` (bash scripts) · `§SectionName-B` (format templates) — read when executing this step.
+```
+
+### Pointer placement rules
+
+1. Place immediately **after** the summary/overview of the removed content in SKILL.md
+2. If an entire step's execution detail is removed, place pointer **at the end of the step's one-line summary**
+3. One pointer per removed block — do not stack multiple pointers for the same removal
+4. The `— read when [condition]` clause is mandatory — it is the activation trigger for imperative loading
+
+### What makes a pointer imperative vs advisory
+
+Advisory (risky — consumer may skip):
+```
+"See SKILL_detail.md for bash scripts."
+"Refer to SKILL_detail.md §Step6-Detail for details."
+```
+
+Imperative (required form):
+```
+"> **Detail**: See `SKILL_detail.md §Step6-Detail` — bash for STALE detection, memory scan, skill usage leaderboard — read when executing Step 6."
+```
+
+The imperative form includes:
+- Blockquote + bold `**Detail**:` prefix (visual weight)
+- Backtick-quoted path
+- What's inside (one line)
+- When to read (activation condition)
+
+---
+
+## §Split-Execution — Step-by-Step Procedure
+
+### Step 1: Produce classification table
+
+Read target SKILL.md. For every H2/H3 section, output:
+
+```
+| Section | Lines | Classification | Reason |
+|---|---|---|---|
+| ## Trigger Phrases | 8 | Always | Invocation recognition |
+| ## Step 3 — bash scripts | 45 | On-demand | Execution detail, step-specific |
+| "auto-allowed" rule | 2 | Always | Behavioral rule |
+...
+```
+
+Count: Always N lines / On-demand N lines / Expected SKILL.md reduction: N%
+
+### Step 2: Trim SKILL.md
+
+1. For each On-demand section: delete content, replace with imperative pointer
+2. Pointer §SectionName must match the header you will create in SKILL_detail.md exactly
+3. Check: does SKILL.md still flow logically? Step overview must remain complete (names + criteria, no gaps)
+
+### Step 3: Create SKILL_detail.md
+
+Front-matter:
+```yaml
+---
+name: {skill-name}-detail
+description: Detail reference for {skill-name} — [what's here]. Load when executing a specific step.
+load: on-demand
+---
+```
+
+Opening line:
+```
+> Load when executing a specific step. SKILL.md contains [list what stays there].
+```
+
+For each pointer in SKILL.md: create matching `## §SectionName` header, paste removed content under it.
+
+Orphan check: every `## §SectionName` in SKILL_detail.md must have a corresponding pointer in SKILL.md. If not → either add pointer or merge with adjacent section.
+
+### Step 4: Verification sequence
+
+```bash
+# 1. Pointer extraction from SKILL.md
+grep -oE "SKILL_detail\.md §[A-Za-z0-9_-]+" plugins/{plugin}/skills/{name}/SKILL.md
+
+# 2. Section headers in SKILL_detail.md
+grep "^## §" plugins/{plugin}/skills/{name}/SKILL_detail.md
+
+# 3. Diff — every pointer must have a matching section
+# Any pointer with no matching section = Phantom → fix before commit
+```
+
+Then run:
+- `/source-grounding-audit` — artifact: SKILL.md, declared source: SKILL_detail.md
+- `/sim-conductor D skill {name}` — provide SKILL.md only, attempt core task from trigger phrase
+
+---
+
+## §Verification-Checklist
+
+Use before committing a completed split:
+
+| Check | Pass condition |
+|---|---|
+| Trigger phrases ≥ 3 | SKILL.md §Trigger Phrases has 3+ entries |
+| Done When defined | SKILL.md has Done When block with ≥1 measurable condition |
+| All §pointers resolve | source-grounding-audit: 0 phantoms |
+| Cold-start grade F | sim-conductor Area D-skill: consumer reaches core completion |
+| No behavioral rule in SKILL_detail only | Any rule governing "what counts as X" present in SKILL.md |
+| No orphan §sections | Every ## §SectionName in SKILL_detail.md has a pointer from SKILL.md |
+| SKILL.md ≤ 50% of original | Line count check |
+| SKILL.md flows without gaps | Reading SKILL.md alone gives complete step sequence understanding |