@chrono-meta/fh-gate 1.4.49 → 1.4.50

package/CATALOG.md

@@ -8,6 +8,12 @@ AI reads this file first when searching past work. Open individual files for det
 
 <!-- Add entries in reverse date order (newest at top) -->
 
+### 2026-07-07 | forge-harness | #sister-asset, #cross-audit, #revfactory, #harness-100, #agent-composer, #benchmarking, #linkedin, #source-verification, #diffusion-llm
+**File:** tracks/_audit/session_2026_07_07_revfactory-harness.md
+Sister-asset cross-audit of `revfactory/harness` + `revfactory/harness-100` (AX TF lead-recommended benchmarking target) vs FH — their axis = one-shot team-architecture generation + a 200-harness ready library (breadth/quick-start); FH's axis = dynamic composition (`agent-composer`) + governance (4-axis gate, irreversibility floors, continuity), which their pipeline lacks entirely. Plus LinkedIn source-verification for the operator's 2 queued insight links: diffusion-LLM paradigm-shift paper confirmed accurate (ICML 2026 Outstanding Paper Award, JustGRPO, arXiv:2601.15165 via official ICML blog); "Claude Code loop-engineering" post confirmed to be about `k021/claude-code-skills` — the *same* sister-asset Gemini already analyzed, not new content.
+- Decision: no functional import beyond a C-tier team-pattern naming label for `agent-composer` output; FH's governance moat holds (revfactory doesn't compete on that axis). Both LinkedIn links closed.
+- Open: adopt 6-pattern naming in agent-composer (operator HITL); harness-100-style pre-built library stays gated behind the existing 3+-recurrence trigger.
+
 ### 2026-06-27 | forge-harness | #sister-asset, #cross-audit, #hermes-agent, #nous-research, #self-improving-agent, #skills, #memory, #messaging-gateway
 **File:** tracks/_contrib/session_2026_06_27_hermes-agent-nous-self-improving-cross-audit.md (committed via _contrib consent lane — authored in an ephemeral cloud session where tracks/_audit/ is gitignored/non-durable)
 Sister-asset cross-audit of **Hermes Agent (Nous Research)** vs FH, triggered by a LinkedIn post (esperer) distilling Hermes' official *Tips & Best Practices* (post = faithful doc summary, not original methodology; same summary circulates on Threads). ~90% of Hermes' best-practice surface is already present in FH (persistent memory · auto-skill-from-repetition · skill self-improvement · context economy · delegation · model selection — all grounded to `plugins/*/skills/`), and on the **self-improvement + governance** axis FH is *ahead*: Hermes *advises* "review auto-generated skills," FH *mechanically enforces* it (pre-commit 4-axis gate + steel/phantom-quench + HITL). Key honest finding — most apparent "gaps" dissolve: cron/daemon is a **deliberate FH boundary** (`self_evolution_routine.md` §8 "recommendation surface, not a daemon"), external-memory-providers **already audited** (companion-store pluggable, 2026-06-11). Only genuine absence = **messaging gateway** (Telegram/Slack daily-driver), which is a *delivery channel*, not methodology.
@@ -177,7 +183,7 @@ CC built-ins utilization imports (operator-approved; video claims verified 9/13
 ### 2026-06-10 | forge-harness | #ingest-gate, #contradiction-scan, #crossref-lint, #llm-wiki, #karpathy
 **File:** .claude/rules/sync_push_protocols.md (+ harness-doctor SKILL.md, probes.md)
 Karpathy LLM-Wiki sister-audit imports (operator-approved; convergence case n=5, citable primary source): I1 — contradiction scan as Sync step 3 (ingest gate, judged + verify-bidirectional pair): new knowledge grepped against existing claims before indexing, conflicts flagged in both files, old-claim removal is HITL. I2 — harness-doctor L4 knowledge cross-ref lint: no CATALOG entry = S-tier index orphan, no inbound ref = R-tier orphan page. Probes G-SYNC-01/G-LINT-01 added (30 total).
-- Decision: scale escape (W1) deliberately NOT built — watch-item with trigger (CATALOG hundreds of entries / repeated search misses); operator-preferred first remedy = skill-splitter-style CATALOG split-mapping, RAG hybrid only after that.
+- Decision: scale escape (W1) deliberately NOT built — watch-item with trigger (CATALOG hundreds of entries / repeated search misses); operator-preferred first remedy = salience-splitter-style CATALOG split-mapping, RAG hybrid only after that.
 - Open: npm republish (harness-doctor SKILL.md shipped) — folded into the open 1.4.8 handoff.
 
 ### 2026-06-10 | forge-harness | #golden-probes, #offline-eval, #doc-code-coupling, #anthropic-4layer
@@ -282,7 +288,7 @@ Post-merge micro R-tier cleanup: corrected two prompt-regression probe expectati
 
 ### 2026-06-03 | forge-harness | #goal-quench, #skill-evolution, #mode-ladder, #sidecar-routing
 **File:** plugins/fh-meta/skills/goal-quench/SKILL.md
-Evolved goal-quench into a fluid core→pro→max mode ladder. Core: token-budget-gate + pipeline-conductor --quick; pro: +context-doctor +agent-composer; max: +plugin-recommender +cross-ecosystem-synergy-detection. Phase-1 budget verdict auto-recommends mode. Ran full-harness dogfood sweep (33 skills): fixed phantom refs, dead blocks, stale agent forks (4 deleted), trigger collision, and 3 skill-splitter splits.
+Evolved goal-quench into a fluid core→pro→max mode ladder. Core: token-budget-gate + pipeline-conductor --quick; pro: +context-doctor +agent-composer; max: +plugin-recommender +cross-ecosystem-synergy-detection. Phase-1 budget verdict auto-recommends mode. Ran full-harness dogfood sweep (33 skills): fixed phantom refs, dead blocks, stale agent forks (4 deleted), trigger collision, and 3 salience-splitter splits.
 - Decision: RED tier reframed as max-mode decomposition on-ramp, not hard block
 
 ### 2026-06-02 | _audit | sister-asset, token-efficiency, compression, headroom

package/CHEATSHEET.md

@@ -511,7 +511,7 @@ Claude agents feature
 |---|---|---|
 | `install-wizard` | First-install onboarding (zshrc, sentinels, the FH self-gate) | "first-time setup", "run the install wizard" |
 | `hub-cc-pr-reviewer` | Reads a PR diff → 8-matrix baseline-consistency check → review comment + merge call | "review this PR", "check this diff" |
-| `skill-splitter` | Splits an over-large SKILL.md that "does everything" into scoped files | "this skill is bloated", "SKILL.md too large", "split this skill" |
+| `salience-splitter` | Splits an over-large SKILL.md that "does everything" into scoped files | "this skill is bloated", "SKILL.md too large", "split this skill" |
 
 ### Agents (sub-agents, dispatched — not slash commands)
 

package/CLAUDE.md

@@ -142,7 +142,7 @@ Compose session-card candidates **into door ③ (field) and the 🔧 door (FH-de
 
 **Identity marker**: every greeting response (Step ②) opens with 🐿️ then an identity-revealing welcome line **on the same line** (a space after 🐿️; exact count not significant — the renderer collapses it — the invariant is *same-line*, not 🐿️ alone) — new / exploratory = "Welcome to FH." · returning = "Welcome back to FH." · operator (FH-dev state) = "The FH operator — good to see you." It is embedded in all skeletons above (do not strip it when composing doors); the exploratory branch template (`fh_detail_protocols.md` Step 2) uses the "Welcome to FH." line.
 
-**Guards**: explicit task-entry utterance → skip onboarding · once per session · code/debug requests → start working directly · project routing is a suggestion, mention at most once
+**Guards**: explicit task-entry utterance → skip onboarding **menu** (the door skeleton / greeting) — but this **never skips the Mode D companion-store freshness load** (pull + INDEX read + card-vs-commit reconcile); that is a data-load, not the menu, and it fires even when the first message is a task (measured miss 2026-07-05: task-first entry skipped the companion-store pull → stale memory → wrong recommendations; now hook-backed via `scripts/fh_session_load.sh`, see `modes_and_value.md §Session-start freshness`) · once per session · code/debug requests → start working directly · project routing is a suggestion, mention at most once
 **Metadata-is-not-intent guard**: the trigger is the user's **typed message only**. Session metadata — branch name (auto-derived from the first message, e.g. `claude/korean-greeting-*`), repo name, file paths — is **never** a task spec and never suppresses or redirects the greeting trigger. A bare greeting fires onboarding even when the branch name looks like a feature request; if the only "task" signal lives in metadata and not in what the user typed, treat the message as a greeting and run the greeting branch + door skeleton above.
 
 ## New Skill Creation Pre-Commit Gate
@@ -285,6 +285,12 @@ unknown) and surface **one line** — then proceed, never block:
 inviolable; a pin is not a cap — tier-floor resolution §Floor governance) · field-project operation
 sessions (no FH asset modification) never see this notice — the Sonnet default stays friction-free.
 
+> **Related — capability-escalation consent**: whether a session actually *escalates* to a stronger
+> model or a cross-family sidecar (not just this advisory notice) is governed separately by
+> `knowledge/shared/harness-core/capability_escalation_consent.md` — the negotiated-consent protocol
+> (UAP `sidecar_consent`/`floorup_consent`) that decides ask-once vs. no-surprise floor-up/sidecar use.
+> This notice is the passive advisory; that doc is the active escalation gate.
+
 ## Field-Harness Load-Bearing Change Gate (cross-family, pre-merge)
 
 The 4-axis gate above fires on **FH asset** changes. But the correlated blind spot it guards —
@@ -537,7 +543,7 @@ Proposal format: `"I see [X]. Want me to run /[skill] to [one-line description]?
 | "add this MCP server", "mount this MCP", "mcp.json에 추가", "connect this tool server" (external-MCP mount intent — **proactive**, fire *before* first tool call; mount intent only — a failing/erroring mounted server is `/mcp-circuit-breaker`'s row above) | `templates/.claude/rules/mcp_tool_gating.md` (name-keyed ask/allow table — never trust server annotations or names; fill §3 at mount time) |
 | "token budget", "how expensive", "estimate tokens", "will this cost a lot" | `/token-budget-gate` |
 | "did my rule change break anything", "regression check", "test harness changes" | `/prompt-regression` |
-| "SKILL.md too large", "split this skill", "skill is bloated", "skill file too long" | `/skill-splitter` |
+| "SKILL.md too large", "split this skill", "skill is bloated", "skill file too long" | `/salience-splitter` |
 | "review for the team", "CTO review", "decision-maker", "share with leadership", "approval deck" | `/apex-review` |
 | "run full pipeline", "verify everything", "end-to-end sweep", "chain all verifications" | `/pipeline-conductor` |
 | "help me write a prompt", "build a prompt", "improve this prompt", "prompt template" | `/meta-prompt-builder` |

package/README.md

@@ -214,7 +214,7 @@ two more signatures keep it running: `harvest-loop` (each session's lessons beco
 | `token-budget-gate` *(fh-commons)* | Pre-task token cost estimate | "How expensive is this?" |
 | `mcp-circuit-breaker` *(fh-commons)* | MCP tool failure pattern detection | "MCP keeps failing" |
 | `quench-challenger` *(fh-commons)* | Adversarial pressure-test agent | "Challenge this with a devil" |
-| *(+ additional assets)* | marketplace-gate · contention-layer · edit-manifest · fact-checker · goal-quench · hub-persona-auditor · install-doctor · memory-hygiene · persona-innovator · prompt-regression · public-surface-audit · skill-splitter | |
+| *(+ additional assets)* | marketplace-gate · contention-layer · edit-manifest · fact-checker · goal-quench · hub-persona-auditor · install-doctor · memory-hygiene · persona-innovator · prompt-regression · public-surface-audit · salience-splitter | |
 
 | Active count | Diagnosis |
 |:---:|---|
@@ -233,7 +233,7 @@ two more signatures keep it running: `harvest-loop` (each session's lessons beco
 | Gate / Guard | `token-budget-gate` · `asset-placement-gate` · `marketplace-gate` |
 | Discovery | `plugin-recommender` · `cross-ecosystem-synergy-detection` · `frontier-digest` · `verify-bidirectional` |
 | Content / Simulation | `sim-conductor` · `apex-review` · `meta-prompt-builder` · `deep-clarify` |
-| Setup | `install-wizard` · `hub-cc-pr-reviewer` · `skill-splitter` |
+| Setup | `install-wizard` · `hub-cc-pr-reviewer` · `salience-splitter` |
 
 > **Full phrasebook** — every skill + agent with its one-line definition and the plain-language phrase
 > that triggers it: [`CHEATSHEET.md` §12](CHEATSHEET.md#12-skills--agents--what-each-does-and-what-to-say).

package/knowledge/shared/harness-core/capability_escalation_consent.md

@@ -0,0 +1,125 @@
+# Capability-Escalation Consent Protocol
+
+> **Principle**: any escalation that raises **cost or trust surface** — recruiting a **cross-family
+> sidecar** (external model families see the work) or a **model-tier floor-up** (Sonnet → Opus, higher
+> $/token) — is **consent-gated and negotiated up front**, never sprung silently. A user who was never
+> asked, then finds a paid/external escalation happened, feels **blindsided (뒤통수)**. The harness
+> must run **intelligently on Claude Code alone at the Sonnet floor** for anyone who declined, using
+> **sub-agents** in place of cross-family sidecars — as a *first-class mode*, not a degraded one.
+
+This is a **규약 (convention)**, not a per-call decision: the same answer holds for the whole session
+once settled, and is **remembered across sessions** (UAP) so it is asked at most once.
+
+---
+
+## Two escalation axes (same consent shape)
+
+| Axis | Escalation | Why it needs consent | Floor when declined |
+|---|---|---|---|
+| **(a) Cross-family sidecar** | recruit codex / agy / gemini / local-4090 as a verifier | work leaves the Claude boundary (external family sees it) + external billing | **Tier-3 CC-only sub-agent** (isolation-decorrelation, honest same-family note) |
+| **(b) Model-tier floor-up** | Sonnet → Opus for a depth-heavy turn | higher $/token (Opus ≈ 3–5× Sonnet); the operator's real cost lever | **stay at Sonnet** (the established minimum-recommended floor) |
+
+**Sonnet = minimum-recommended model is already established** — so the declined-floor is safe: the
+harness is *designed* to run well at Sonnet, not crippled by it (`[[feedback_harness_aerodynamics_perceived_perf]]`).
+
+---
+
+## When consent is settled — two entry points
+
+### 1. Onboarding negotiation (install-wizard) — the no-surprise path
+
+install-wizard **explicitly negotiates both axes** at setup, as named items:
+- *"Allow cross-family sidecars (codex/agy/local) for adversarial verification of load-bearing changes? They add external-family decorrelation but external billing applies."* → records `sidecar_consent`.
+- *"Allow automatic Sonnet→Opus floor-up on depth-heavy turns? Opus is ~3–5× the cost; declining keeps you at the Sonnet floor and asks per-occasion instead."* → records `floorup_consent`.
+
+Settling this at onboarding is the **whole point**: the escalation is then *expected*, never a surprise
+on the bill or the egress log.
+
+### 2. Runtime ask-once (for anyone who skipped onboarding setup) — the graceful path
+
+A user who skipped explicit setup is **not** auto-escalated. Instead, the **first time** an escalation
+is actually needed:
+- **Ask once**, at the moment of need, framed with the cost/trust reason:
+  *"This turn needs Opus depth (higher cost) — proceed at Opus, or stay at Sonnet?"* /
+  *"This load-bearing change is best verified cross-family (external billing) — recruit a sidecar, or verify with CC sub-agents only?"*
+- **Accept** → record consent (UAP), proceed, no re-ask.
+- **Decline** → record decline (UAP), **mark the escalation "recommended only"** going forward (surface
+  it as a one-line recommendation when relevant, **never a re-nag**), and **proceed at the floor**
+  (Sonnet / Tier-3 sub-agent).
+
+The ask fires **once per axis**; the answer is remembered. A declined axis becomes a standing floor, not
+a per-turn question.
+
+---
+
+## The declined mode is first-class, not "degraded"
+
+When an axis is declined (or no sidecar is reachable), the fallback is **the user's chosen normal
+operating mode**, and must be framed that way:
+
+- **Cross-family declined → Tier-3 CC-only sub-agent verification.** Multiple **isolated** Claude
+  sub-agents adversarially verify (isolation-decorrelation), with an **honest same-family note**
+  ("no cross-family diversity — verification is isolation-decorrelated only"). This is intelligent
+  operation, **not** "reduced value / degraded" language. (`[[feedback_judge_robustness_mechanical_anchor]]`
+  still holds: the governor keeps the terminal verdict + a mechanical anchor.)
+- **Floor-up declined → stay at Sonnet**, run the turn with good harness structure. No apology framing.
+
+**Reframe rule**: the Sidecar Resolution Protocol's Tier-3 line and auto-decorrelation's degrade ladder
+must not read "no diversity; reduced value" for a *declined* user — that pathologizes their choice.
+Distinguish **declined** (chosen floor — first-class) from **unavailable-but-wanted** (genuine
+degrade-with-note). Only the latter carries the degrade framing.
+
+### Reconciliation with the corp fail-closed invariant
+
+The pmh corp degrade-invariant (`local_pmh_context.md` auto-decorrelation Step 6) says a load-bearing
+change with **no reachable cross-family panel → NOT-CONVERGED / ask operator**. That is the
+**unavailable-but-expected** case (consent given, panel down). The **declined** case is different: the
+user opted out, so Tier-3 sub-agent verification **is** the standing bar (with the honest same-family
+note), and load-bearing changes proceed under it — not blocked. Membership: `floorup_consent`/
+`sidecar_consent == declined` in the UAP routes to the first-class floor; absence of a *wanted* panel
+routes to fail-closed.
+
+---
+
+## UAP persistence (behavioral pref — never domain content)
+
+`operational_adaptation.md` UAP gains two fields:
+- `sidecar_consent: accepted | declined | unset`
+- `floorup_consent: accepted | declined | unset`
+
+**READ** (session start / at point of need): `declined` → route to floor, surface as recommendation
+only, no re-nag. `accepted` → escalation available (still per-need, no auto-run). `unset` → ask-once
+on first need (runtime path above).
+**WRITE**: on onboarding settle, or on the first runtime accept/decline.
+
+Ephemeral/cloud sessions (UAP wiped) → operate from the **Sonnet floor + CC-only** default (the safe
+floor), do not fabricate consent.
+
+---
+
+## Cost lens (why this IS the cost-governance mechanism)
+
+The floor-up consent prompt is **the operator's cost lever made explicit** (measured: Sonnet-economized
+≈ 34× subscription; Opus-full 40–60×, `cost_report_2026-07-07`). Default Sonnet floor + escalate-Opus-
+**only-where-depth-pays**-with-consent is precisely "amortize Opus on the turns that earn it." Pair with
+a **free internal-model execution lane** where the environment has one (`[[feedback_multimodel_3lane_architecture]]`)
+to offload exec cost off the paid API entirely. The protocol turns hard-won economizing into a mechanical default.
+
+---
+
+## Done When
+
+- **Both axes negotiated at onboarding** (install-wizard names sidecar + floor-up as consent items). *[mandatory-pass — grep install-wizard for both items]*
+- **Runtime ask-once wired** for `unset` consent at first need; decline → recommend-only + floor, no re-nag. *[judged — pair: Sonnet target-tier blind sim of a skipped-onboarding session hitting first floor-up]*
+- **UAP fields present + READ applied** (`sidecar_consent`, `floorup_consent`). *[mandatory-pass — grep operational_adaptation.md]*
+- **Declined-mode reframed first-class** (no "degraded/reduced-value" language on a declined user's Tier-3 path). *[judged — pair: challenger reads auto-decorrelation + Sidecar Resolution for pathologizing language]*
+- **Corp fail-closed reconciled** (declined ≠ unavailable-but-wanted). *[judged — pair: the target-tier sim above exercises both branches]*
+
+## Guards
+
+- **Human override inviolable** — a consent is a default, never a cap; the operator can always force a
+  tier/sidecar for a given turn (`[[feedback_verify_before_downgrade]]` floor governance).
+- **Never auto-switch the session model** — the floor-up ASK proposes; the human acts (`/model`). The
+  protocol never flips the model itself.
+- **Sonnet floor is a floor, not a ceiling** — declined-floor-up still escalates *within* Sonnet's
+  harness depth (sub-agents, good structure); it does not cap capability, only cost/trust surface.

package/knowledge/shared/harness-core/fh_detail_protocols.md

@@ -37,12 +37,31 @@ ls ../ | grep -iE '(forge-harness|meta-harness|-harness|-hub)'
 ls .claude/registry/LOCAL_SKILL_REGISTRY.md 2>/dev/null
 ```
 - File exists and modified within 7 days → load into session
-- Missing or older than 7 days → regenerate:
+- Missing or older than 7 days → regenerate.
+
+**No hardcoded root — derive the install location (users install FH anywhere).** The projects root is
+the *parent of the FH repo*, discovered at runtime, never a literal `~/projects` / `~/PycharmProjects`
+(a hardcoded root silently returns 0 on any machine whose layout differs — the 2026-07-05 dead-path
+`fail-open` bug: `find ~/projects` on a `~/PycharmProjects` machine → 0 catches → the registry is
+overwritten empty and cross-project summon goes dark):
 ```bash
-find ~/projects -path "*/.claude/skills/*/SKILL.md" \
-  -not -path "*/forge-harness/*" 2>/dev/null
+HUB="${CLAUDE_PROJECT_DIR:-$(pwd)}"          # FH 레포 위치 (설치 위치 무관, 감지)
+ROOT="$(cd "$HUB/.." 2>/dev/null && pwd)"     # 형제 프로젝트가 사는 부모 = 프로젝트 루트
+# 두 레이아웃 모두 포착: .claude/skills/*/SKILL.md AND 루트-레벨 */SKILL.md (예: gstack).
+# vendored(.venv·site-packages·node_modules·.git) 제외 — 없으면 playwright/streamlit 스킬까지 삼킴.
+FOUND="$(find "$ROOT" -name SKILL.md \
+  -not -path "*/.venv/*"  -not -path "*/site-packages/*" \
+  -not -path "*/node_modules/*"  -not -path "*/.git/*" \
+  -not -path "$HUB/*" 2>/dev/null)"    # exclude FH's own subtree by DERIVED path, not a name-literal (works when FH is cloned under any dir name)
 ```
-Group by project → update `.claude/registry/LOCAL_SKILL_REGISTRY.md`. Propose cross-project skills when request maps to registry. Scan once per session.
+Then **fail-closed** (irreversible-ish: a silent empty overwrite blinds the bus): if `$FOUND` is empty
+**and** the existing registry has >0 entries, do **not** overwrite — flag `⚠️ scan returned 0 (root=$ROOT);
+kept existing registry` and skip the rewrite. Only rewrite when the scan is non-empty (or the registry
+was absent). Group by project (parent dir name). Record per skill: name · path · description · trigger
+phrases · `requires_cwd` · `direct-executable` · `origin(FH|project|external)`+trust. **Non-FH skills are
+propose-only (ask-tier), never auto-run** — a cross-project skill body is an injection surface. Propose
+cross-project skills when a request maps to the registry. Scan once per session. (Detection belongs at
+install too — `/install-wizard` records HUB/ROOT so the runtime never guesses; see install-wizard.)
 
 ### Step 2 — Active Proposal
 

package/knowledge/shared/harness-core/measurement-integrity-checklist.md

@@ -1,7 +1,7 @@
 # Measurement-Integrity Checklist — cross-model measurement pre-flight
 
 > A cross-model measurement is only trustworthy if its **instrument** is verified first.
-> Measurement integrity is a *precondition*, not a result. Three observed failure modes, each with a
+> Measurement integrity is a *precondition*, not a result. Four observed failure modes, each with a
 > concrete countermeasure. Consult this before any FH measurement that compares models (sims, sidecar
 > comparisons, capability-equalizer runs, the-bible model panels, A6-class experiments).
 
@@ -16,6 +16,7 @@ becomes a gate other skills invoke, revisit the weight.
 | 1 | **Silent model fallback** — passing a model *slug* silently resolved to a weaker model (e.g. an `agy` slug fell back to Flash) instead of the intended one. The run *looks* like the named model but isn't. | **Pin the display name, not the slug** (e.g. `"Gemini 3.1 Pro (High)"`, not a bare slug). Confirm the resolved identity, don't assume the slug binds. |
 | 2 | **Non-deterministic borderline verdicts** — contested/borderline cases flip across runs (observed: haiku 4/4 flip; flagship models flip too — flipping is **not** a tier signal). A single draw is noise, not a measurement. | **reps ≥ 3 on any borderline/contested verdict.** A single run on a contested case is inadmissible. Report the flip pattern (STABLE vs FLIP), not just the modal verdict. |
 | 3 | **Generic self-identity probe** — a probe any model passes ("are you working? → OK") proves nothing about *which* model answered. | **Use a discriminating probe** — one that two different models answer *differently*. A generic-pass probe is invalid. The probe is a **pattern, not a fixed string**: a probe that discriminates Opus 4.8 from Sonnet 4.6 today may both-pass a future model generation, so **re-validate the probe each model generation** (same staleness class `memory-hygiene` exists to catch). |
+| 4 | **Serving-path / quantization variance** — the *same* display-name model served over two different backends (different quantization/infra) is a **different instrument** and yields materially different measurements. Observed: one GLM-5.2 model family gave effect-size delta **+0.21** when served via an internal NVFP4-quantized deployment vs **+0.08** via an OpenRouter relay — same model name, ~2.6× different effect (n=864, reps≥3). A correctly-pinned display name (item #1) is **necessary but not sufficient**. | **Pin *and record* the serving path** — backend host + quantization, not just the display name. Two runs are comparable only if the serving path matches; a name match across different infra is an implicit apples-to-oranges. When you cannot hold it fixed, **report the serving path as a measured variable**, not a constant. |
 
 ## Why these are entangled (and why they matter beyond their own scope)
 
@@ -28,10 +29,17 @@ single-draw artifact. Item #3 (discriminating probe) **embodies** the judge-robu
 mechanical-anchor principle — don't trust self-reported identity, prove it discriminatingly
 ([[feedback_judge_robustness_mechanical_anchor]]).
 
+Item #4 (serving-path variance) **sharpens** item #1 into a two-part identity: #1 catches the *wrong
+model* (a slug that fell back); #4 catches the *right model on the wrong instrument* (a correct name
+served over a different quantization/backend). The verified identity a measurement records is therefore
+**name + serving path**, not name alone — a family-decorrelation claim (cross-family sidecar) is only
+sound once the serving path of each family is itself pinned, else "different family" silently smuggles
+"different infra" ([[reference_measurement_serving_path_variance]]).
+
 ## Done When
 
-- The checklist enumerates all three failure modes, each with its countermeasure.
-  *Check class: mandatory-pass (binary — three items present, each with a countermeasure).*
+- The checklist enumerates all four failure modes, each with its countermeasure.
+  *Check class: mandatory-pass (binary — four items present, each with a countermeasure).*
 - The probe item specifies a **discriminating** test and rejects generic probes.
   *Check class: judged, pair: a probe that two different models both pass must FAIL this check; a
   discriminating one must distinguish them.*
@@ -57,3 +65,15 @@ mechanical log.
 ambiguity). Sister findings: [[feedback_correlated_blindspot_union_over_majority]] (reps≥3 prerequisite),
 [[feedback_judge_robustness_mechanical_anchor]] (discriminating-probe = mechanical anchor),
 [[reference_agy_model_catalog]] (display-name pin — agy slug fallback documented there).
+
+**#4 added** (2026-07-05): serving-path variance surfaced in a cross-family verdict-invariance run
+(n=864, borderline fixtures × 2 conditions × K=6 paraphrase × reps≥3). An identical GLM-5.2 model name
+served over an internal NVFP4-quantized deployment vs an OpenRouter relay gave +0.21 vs +0.08
+effect-size delta — quantifying that "same model name ⇒ same measurement" is false. Provenance +
+generalizable finding: [[reference_measurement_serving_path_variance]].
+
+**External corroboration** (2026-07): the local-LLM community independently reports the same hazard —
+practitioners conflate "running model X" with running a *pruned/quantized derivative* of X (aggressive
+low-bit quantization + expert pruning measurably degrade long-context quality while the model *name* is
+unchanged). This is a general measurement pitfall, not FH-specific: a leaderboard or replication that
+pins only the display name silently compares different instruments across serving paths.

package/knowledge/shared/harness-core/multi_model_sidecar_strategy.md

@@ -167,6 +167,35 @@ buff and degrades that model's realized intelligence** — not just Claude's. Th
   is autocomplete/QA only, which is exactly its demoted role. (Derived 2026-07-03, operator + cross-vendor
   Gemini concurrence; extends the governor=native-CC point to every vendor.)
 
+### Batch-judging corollary — the native harness is for interactive/agentic work, not batch scoring
+
+The vendor-native harness gives a model its highest capability for **interactive, agentic** tasks
+(repo-grounded audit, multi-step design, tool-use) — but that *same* agentic loop is a **liability for
+high-volume batch judging**: a deterministic verdict emitted over N fixtures, where there is nothing for a
+tool-use loop to do. Measured 2026-07-04 (H1 verdict-invariance run): the native `codex exec` spins a full
+agentic session per judge (hooks + reasoning ≈ an order of magnitude more tokens than a bare completion),
+and native `agy -p` (once its headless permission-wait is cleared) returns *agentic prose* — a "Summary of
+Work" — rather than a parseable last-line verdict. Both **complete**, but at a cost/parse profile wrong for
+batch.
+
+So the dispatch splits by *shape of the task*, not just by family:
+
+- **Batch cross-family judging** (steel-quench Step 0.6 verdict-invariance, auto-decorrelation over many
+  items, any fixed-fixture flip count) → **clean completion APIs** (OpenRouter, model pinned by
+  *display-name* + `served`-field silent-route check) **+ free local** (a 4090 ollama endpoint). Clean,
+  cheap, parseable, per-call pinnable. *Caveat*: a local thinking model needs a large enough output budget
+  or it truncates inside `<think>` and emits an empty verdict — a config axis, not a capacity limit.
+- **Interactive / agentic verification** (repo-grounded catching, the divergence audits where cross-family
+  disagreement *localizes* a bug) → the **native CLIs** (`codex`, `agy`/Antigravity), where the harness
+  earns its overhead.
+
+This is **not** a contradiction of the harness-depth thesis — it *is* it. The harness lifts capability
+exactly where judgment + tools + iteration matter; for a one-shot self-contained verdict the loop has no
+work, so its depth becomes pure cost. Pick the naked API for batch scoring, the native harness for agentic
+audit. (Derived 2026-07-04, operator + H1 measurement; the batch-side dual of the vendor-native thesis
+above. The native-harness Gemini path via `agy` is headless-usable again once tool-permission auto-proceed
+is set — see [[reference_agy_model_catalog]] for the pin/permission mechanics.)
+
 **Maintenance-Cost Rule** — a compatibility layer is cheap as a *thin entrypoint*, expensive when it
 *duplicates canonical knowledge*. The test:
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@chrono-meta/fh-gate",
-  "version": "1.4.49",
+  "version": "1.4.50",
   "description": "FH runtime adapters — run FH governance, skills, and agents via Claude or Codex with machine-parseable gates.",
   "license": "MIT",
   "keywords": [

package/plugins/fh-meta/skills/auto-decorrelation/SKILL.md

@@ -100,11 +100,20 @@ Local 4090 = **canary tier** (evidence-of, never terminal verdict).
 
 ## Step 6 — Degrade ladder (the intelligent scale-down)
 
+**Consent branch first — declined ≠ degraded** (`[[capability_escalation_consent]]`): if the UAP has
+`sidecar_consent: declined`, do **not** probe/recruit — route straight to **Tier-3 CC-only sub-agent
+verification** (multiple isolated Claude sub-agents, isolation-decorrelation) as a **first-class chosen
+mode**, with an honest *same-family* note but **no "reduced value / degraded" framing** — the user chose
+this floor. `unset` → ask-once at first load-bearing need (accept → proceed; decline → record + this
+branch). Only proceed to the discovery ladder below when consent is `accepted`.
+
 1. frontier cross-family CLI present → recruit it (decorrelated, at-floor) — best.
 2. only local 4090 present → canary pre-screen + in-session opus governor (canary, not full decorrelation).
 3. nothing present → in-session same-family + **honest below-floor/same-family note** (residual named).
 
-Env non-determinism (CLI presence varies) → **silent degrade, never hard-fail**.
+Env non-determinism (CLI presence varies) → **silent degrade, never hard-fail**. Distinguish this
+**unavailable-but-wanted** case (consent given, panel down → degrade-with-note; for a *load-bearing corp*
+surface, fail-closed per `local_pmh_context.md`) from the **declined** case above (chosen floor, first-class).
 
 ## Step 7 — Output
 

package/plugins/fh-meta/skills/context-doctor/SKILL.md

@@ -117,7 +117,7 @@ Run the audit bash (§Step-Bash) and apply thresholds:
 | CLAUDE.md | Exceeds 300 lines | Section-by-section compression / move completed sections to archive |
 | MEMORY.md | Exceeds 180 lines | Check entry count + move `✅ CLOSED` items to archive section |
 | memory/*.md single file | Exceeds 30K (300 lines) | Suggest splitting accumulated history into separate files |
-| SKILL.md (any) | > 300 lines AND no SKILL_detail.md | Propose `/skill-splitter` — governance-semantic split (not compression); compression removes content, splitting routes it on-demand |
+| SKILL.md (any) | > 300 lines AND no SKILL_detail.md | Propose `/salience-splitter` — governance-semantic split (not compression); compression removes content, splitting routes it on-demand |
 
 **Frequency**: When explicitly called with `/context-doctor` or auto-invoked at session start when MEMORY.md is detected at 180+ lines.
 
@@ -259,7 +259,7 @@ context-doctor (token/context) · harness-doctor (structure) · sim-conductor (s
 |---|---|
 | Want to also check structure after resolving token waste | `/harness-doctor` |
 | Want to validate prescription results from external user perspective | `/sim-conductor Area A` |
-| SKILL.md diagnosed as over-loaded (> 300 lines, no SKILL_detail.md) | `/skill-splitter` — governance-semantic split |
+| SKILL.md diagnosed as over-loaded (> 300 lines, no SKILL_detail.md) | `/salience-splitter` — governance-semantic split |
 | All three skills mentioned simultaneously | Three-Doctor Loop circuit activated — diagnosis→prescription→re-diagnosis cycle
 
 ## Done When
@@ -278,4 +278,4 @@ context-doctor (token/context) · harness-doctor (structure) · sim-conductor (s
 **→ Three-Doctor Loop chain (auto-propose after diagnosis):**
 - Prescription modifies SKILL.md / rules / CLAUDE.md → **propose `/harness-doctor`** re-check after fix (structural integrity)
 - Prescription addresses user-facing context (onboarding, README, install guides) → **propose `/sim-conductor Area A`** (external user impact validation)
-- SKILL.md detected as over-loaded → **auto-propose `/skill-splitter`**: `"I see [skill-name] SKILL.md is [N] lines with no SKILL_detail.md. Want me to run /skill-splitter to do a governance-semantic split?"`
+- SKILL.md detected as over-loaded → **auto-propose `/salience-splitter`**: `"I see [skill-name] SKILL.md is [N] lines with no SKILL_detail.md. Want me to run /salience-splitter to do a governance-semantic split?"`

package/plugins/fh-meta/skills/context-doctor/SKILL_detail.md

@@ -73,11 +73,11 @@ wc -l memory/MEMORY.md 2>/dev/null
 # memory/*.md files exceeding 30K
 find memory -name "*.md" -size +30k 2>/dev/null | xargs wc -l | sort -rn | head -10
 
-# SKILL.md files > 300 lines with no SKILL_detail.md (skill-splitter candidates)
+# SKILL.md files > 300 lines with no SKILL_detail.md (salience-splitter candidates)
 find plugins -name "SKILL.md" 2>/dev/null | while read f; do
   lines=$(wc -l < "$f")
   detail=$(dirname "$f")/SKILL_detail.md
-  [ "$lines" -gt 300 ] && [ ! -f "$detail" ] && echo "[skill-splitter candidate] $f ($lines lines)"
+  [ "$lines" -gt 300 ] && [ ! -f "$detail" ] && echo "[salience-splitter candidate] $f ($lines lines)"
 done
 ```
 

package/plugins/fh-meta/skills/goal-quench/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: goal-quench
 description: >-
-  Wraps /goal with a tiered safety + orchestration ladder. core (default): a token budget gate (pre-run estimate), mid-run budget thresholds, and an automatic post-run quality verification via pipeline-conductor — closing /goal's two gaps (Haiku evaluates completion, pipeline-conductor evaluates correctness). pro: adds context-doctor token reduction and agent-composer goal decomposition. max: adds plugin-recommender capability-gap fill and cross-ecosystem-synergy-detection pre-validation. The Phase-1 budget verdict auto-recommends the mode. Triggered by "goal with quality gate", "safe goal", "goal-quench", "orchestrate this goal", or before running /goal on high-stakes tasks.
+  Wraps /goal with a tiered safety + orchestration ladder. core (default): a token budget gate (pre-run estimate), mid-run budget thresholds, and an automatic post-run quality verification via pipeline-conductor — closing /goal's two gaps (Haiku evaluates completion, pipeline-conductor evaluates correctness). pro: adds context-doctor token reduction and agent-composer goal decomposition. max: adds internal-first capability-gap fill (LOCAL_SKILL_REGISTRY skill-bus scan → plugin-recommender external search → auto_project_mapping map/cluster) and cross-ecosystem-synergy-detection pre-validation. The Phase-1 budget verdict auto-recommends the mode. Triggered by "goal with quality gate", "safe goal", "goal-quench", "orchestrate this goal", or before running /goal on high-stakes tasks.
 user-invocable: true
 allowed-tools: ["Read", "Write", "Bash", "Grep"]
 model: sonnet
@@ -39,7 +39,7 @@ goal-quench is a ladder, not a fixed shape. The default (**core**) is the narrow
 |---|---|---|---|
 | **core** (default) | budget gate + mid-run thresholds + post-run quality gate | token-budget-gate, pipeline-conductor --quick | budget GREEN / YELLOW |
 | **pro** | token-reduction pre-pass + goal decomposition into Waves | + context-doctor, agent-composer | budget ORANGE |
-| **max** | capability-gap fill + synergy pre-validation before the run | + plugin-recommender, cross-ecosystem-synergy-detection | budget RED |
+| **max** | **internal-first** capability-gap fill (skill-bus scan → external → map/cluster) + synergy pre-validation | + LOCAL_SKILL_REGISTRY scan, plugin-recommender, cross-ecosystem-synergy-detection, auto_project_mapping | budget RED |
 
 Each mode is a **superset** of the one before it — pro does everything core does, plus more. Nothing in core is removed by escalating.
 
@@ -105,6 +105,7 @@ The clean-streak graduation above governs **unattended-no-review** operation (ru
 - "goal-quench", before any long /goal session
 - `/goal-quench --pro`, `/goal-quench --max`, "orchestrate this goal", "decompose this goal", "optimize then run this goal"
 - "this goal is too big for one run", "find a tool for this goal if FH lacks one" (→ max mode)
+- "assemble capabilities for this project", "develop project X with everything FH has", "what can FH + tools do for this project", "use all of FH's and others' sharp parts on this" (→ max mode, **project-dev capability-assembly** entry — internal skill-bus scan first, then external gap-fill, then map+cluster; token-efficient, non-expert-friendly)
 - Automatically proposed when user mentions `/goal` on tasks estimated > 15K tokens
 - Mode is auto-recommended by the Phase-1 budget verdict (GREEN/YELLOW → core, ORANGE → pro, RED → max)
 
@@ -190,14 +191,16 @@ goal-quench does **not** re-implement agent-composer's gates — its destructive
 
 > **Detail**: See `SKILL_detail.md §Queue-Format` — queue file format, plan-ready output text, sidecar `.active` fields — read when writing the queue (Step B) or the sidecar fields (Step D).
 
-### Step C — plugin-recommender + synergy pre-validation · max only
+### Step C — capability fill: internal-first, then external · max only
 
-Triggered only when agent-composer Step 0.2 reports a capability **GAP** (`fit_score < 0.5` on a required-weight sub-task):
-1. `plugin-recommender` searches FH + Codex + Claude Code marketplaces for a fitting skill/agent.
-2. For each candidate, `cross-ecosystem-synergy-detection` pre-validates fit + overlap **before anything is installed**.
-3. User decides: install / skip / general-purpose fallback (agent-composer's degraded-composition rule applies — `⚠️ degraded: [role]`).
+Triggered only when agent-composer Step 0.2 reports a capability **GAP** (`fit_score < 0.5` on a required-weight sub-task). Fill order is **internal before external** — honors no-reinvention (use FH's + sibling projects' sharp parts before installing anything), and cheaper **on an internal hit** (a local registry read ≪ an external marketplace search; on an internal *miss* the scan is added overhead before the same external path):
 
-max mode never installs anything silently — discovery and synergy-check are surfaced for approval first.
+1. **Internal scan first.** Run the `LOCAL_SKILL_REGISTRY` scan / Cross-Project Skill Bus (the runtime-root-derived scan under `.claude/registry/`) to check whether an existing **FH meta-skill or a sibling-project skill** already fills the gap. **Trust-gated dispatch** (the fields the registry carries are *consumed*, not just surfaced): an **FH meta-skill** hit → propose directly (Agent + Context Card, no install, token-cheap); a **non-FH sibling** hit → if `trust≠trusted` OR `writes≠read-only` → **ask-tier HITL before dispatch** (never propose-directly — sibling code is an injection surface), else propose. **Degrade direction**: if `.claude/registry/` is missing/malformed, treat as an internal-miss and **fall through to external (step 2), warn** — never silently skip capability fill (that drops the GAP sub-task unnoticed; capability-fill is a reversible surface → advisory-degrade per the Surface-Class Degrade Invariant).
+2. **External only if internal has nothing.** `plugin-recommender` searches FH + Codex + Claude Code marketplaces + external repos for a fitting skill/agent; `cross-ecosystem-synergy-detection` pre-validates fit + overlap **before anything is installed**.
+3. User decides: use-internal / install-external / skip / general-purpose fallback (agent-composer's degraded-composition rule applies — `⚠️ degraded: [role]`).
+4. **Approved external capability → map + cluster (deferred to session close).** An adopted external skill or a newly-relevant sibling project is registered via `auto_project_mapping` — but that write **edits CLAUDE.md's track-mapping table and creates `tracks/{project}/` (an FH-asset change that trips the 4-axis auto-gate and shifts the onboarding is-mapped signal — [[feedback_tracks_dir_is_mapped_signal]])**, so it is **deferred to session close**, not run mid-`/goal`-loop: it passes the 4-axis gate as its own change instead of interrupting the run. The **next** run then discovers the capability internally at step 1 — the loop compounds.
+
+max mode never installs or maps anything silently — internal hit, external discovery, synergy-check, and mapping are all surfaced for approval first.
 
 ### Step D — scope-driven sidecar configuration · pro + max
 
@@ -308,8 +311,10 @@ After each goal-quench run, append a calibration entry to `tracks/_meta/goal_que
 - `token-budget-gate` — Phase 1 cost estimation (all modes)
 - `context-doctor` — Phase 1.5 Step A token-reduction pre-pass (pro + max)
 - `agent-composer` — Phase 1.5 Step B goal decomposition into Waves (pro + max)
-- `plugin-recommender` — Phase 1.5 Step C capability-gap fill (max only, GAP-triggered)
+- `LOCAL_SKILL_REGISTRY` scan / Cross-Project Skill Bus — Phase 1.5 Step C internal-first scan (max only, GAP-triggered, before external)
+- `plugin-recommender` — Phase 1.5 Step C external capability-gap fill (max only, GAP-triggered, only if no internal hit)
 - `cross-ecosystem-synergy-detection` — Phase 1.5 Step C pre-validation of discovered candidates (max only)
+- `auto_project_mapping` — Phase 1.5 Step C map/cluster of an adopted external capability (max only, approval-gated, deferred to session close)
 - `pipeline-conductor` — Phase 3 quality gate (`--quick` for core/pro, `--full` for max; called by Stop hook)
 - `field-harvest` — capture calibration data as reusable pattern after 10 runs
 
@@ -321,7 +326,9 @@ After each goal-quench run, append a calibration entry to `tracks/_meta/goal_que
 Phase 1: token-budget-gate verdict output + mode resolved (core default, or pro/max via budget verdict / explicit flag)
 + .claude/goal-quench.active written (with mode: field) + thresholds injected
 + If pro/max: Phase 1.5 ran — context-doctor pre-pass + agent-composer plan;
-  max additionally: GAP-triggered plugin-recommender + cross-ecosystem-synergy-detection pre-validation
+  max additionally: GAP-triggered internal LOCAL_SKILL_REGISTRY scan FIRST (trust-gated dispatch);
+  plugin-recommender + cross-ecosystem-synergy pre-validation ONLY if no internal hit;
+  auto_project_mapping map/cluster deferred to session close only if an external capability was adopted (each surfaced for approval)
 + Phase 3 (on next response after /goal): .pending file detected + pipeline-conductor run
   (--quick for core/pro, --full for max)
 + Verification verdict output (CLEAN/PENDING/BLOCKED/ESCALATE)

package/plugins/fh-meta/skills/harness-doctor/SKILL.md

@@ -76,7 +76,7 @@ confirms the gap*, never a verdict the report emits on its own; surface each gap
 | CLAUDE.md 100~200 lines | S-tier warning |
 | CLAUDE.md 200+ lines | M-tier — separation or reduction needed |
 | 15+ `##` sections in CLAUDE.md | S-tier warning |
-| SKILL.md > 300 lines AND no `SKILL_detail.md` | S-tier — propose `/skill-splitter` (governance-semantic split, not compression) |
+| SKILL.md > 300 lines AND no `SKILL_detail.md` | S-tier — propose `/salience-splitter` (governance-semantic split, not compression) |
 | Rules files unreferenced in CLAUDE.md | R-tier |
 | weekly_audit 14~30 days elapsed | S-tier |
 | weekly_audit 30+ days elapsed | M-tier |
@@ -318,7 +318,7 @@ Verdict: PASS (M-tier 0, "Structure healthy") | CONDITIONAL_PASS (S/R remain, no
 **Three-Doctor Loop chain** (auto-propose after prescription report):
 - M-tier + context/token waste → propose `/context-doctor`
 - M-tier + user-facing behavior changes → propose `/sim-conductor Area A`
-- SKILL.md S-tier (> 300 lines, no SKILL_detail.md) → auto-propose `/skill-splitter`: `"I see [skill] SKILL.md is [N] lines with no SKILL_detail.md. Want me to run /skill-splitter?"`
+- SKILL.md S-tier (> 300 lines, no SKILL_detail.md) → auto-propose `/salience-splitter`: `"I see [skill] SKILL.md is [N] lines with no SKILL_detail.md. Want me to run /salience-splitter?"`
 - Both → propose full Three-Doctor Loop: context-doctor → sim-conductor Area A
 
 ---

package/plugins/fh-meta/skills/harvest-loop/SKILL.md

@@ -137,7 +137,7 @@ FAIL routing: First FAIL → 1 re-synthesis allowed. FAIL after re-synthesis →
 
 ### Step 6 — Curator Lifecycle Review
 
-**6-1 SKILL.md Lifecycle**: 30+ day unused → [STALE] candidate. `pinned: true` → never touch. ≥70% overlap → merge candidate suggestion. **> 300 lines AND no `SKILL_detail.md`** → propose `/skill-splitter` (governance-semantic split — not compression; the grew-through-harvest pattern is a natural split trigger).
+**6-1 SKILL.md Lifecycle**: 30+ day unused → [STALE] candidate. `pinned: true` → never touch. ≥70% overlap → merge candidate suggestion. **> 300 lines AND no `SKILL_detail.md`** → propose `/salience-splitter` (governance-semantic split — not compression; the grew-through-harvest pattern is a natural split trigger).
 
 **6-1-a Archive-candidate auto-tag**: When 0 invocations in 30 days detected (cross-check `tracks/_meta/skill_usage.md`), auto-append `#archive-candidate` tag to that skill's CATALOG.md entry. No file deletion — tag only. User reviews tagged entries at next session start.
 

package/plugins/fh-meta/skills/install-wizard/SKILL.md

@@ -139,7 +139,9 @@ Y (add integration plan items to Step 1) / N (add-only, keep existing rules) / S
 
 **[Prerequisite] install-doctor conflict diagnosis (only in environments without install history)**: if `~/.cc_sentinels/{project-name}_wizard_done` doesn't exist (first install), call `/install-doctor --plugin fh-meta` first. CONFLICT/WARNING items → add ❗ markers to Step 2 proposal list. Items the doctor already diagnosed (`FH plugin install` · `zshrc hook` · `.claudeignore`) → map results directly, skip re-diagnosis; all other items → check directly.
 
-Auto-check each item as PASS / MISS / FAIL. Check items: `.claudeignore` · `local_fh_context.md` · `zshrc hook` · `weekly_audit` freshness · `sentinel` setup · FH plugin install · `.git/info/exclude` · MCP plugin · `deep-insight` plugin (optional) · `fh_env_context.jsonc` · `phantom-gate` (Python + AI-output projects only) · domain pattern pack (optional, none ship by default) · local-LLM offload (optional — surface ONLY if the Step 0 bash emitted the literal line `Local LLM runtime: detected`; absent that line, this item does not exist).
+Auto-check each item as PASS / MISS / FAIL. Check items: `.claudeignore` · `local_fh_context.md` · `zshrc hook` · `weekly_audit` freshness · `sentinel` setup · FH plugin install · `.git/info/exclude` · MCP plugin · `deep-insight` plugin (optional) · `fh_env_context.jsonc` · `phantom-gate` (Python + AI-output projects only) · domain pattern pack (optional, none ship by default) · local-LLM offload (optional — surface ONLY if the Step 0 bash emitted the literal line `Local LLM runtime: detected`; absent that line, this item does not exist) · **`env-delta SessionStart hook`** (Mode D — `scripts/fh_env_delta_scan.sh` registered in SessionStart; MISS if absent when the hub has sibling repos) · **`capability-escalation consent`** (per `[[capability_escalation_consent]]` — negotiate BOTH axes at onboarding so no later escalation is a surprise: *"Allow cross-family sidecars (external billing) for load-bearing verification?"* → `sidecar_consent`, and *"Allow Sonnet→Opus floor-up on depth-heavy turns (~3–5× cost)? Decline = stay at the Sonnet floor, asked per-occasion instead."* → `floorup_consent`; record both to the UAP. MISS if the UAP lacks either field — a skipped user is asked-once at first need, never sprung).
+
+**env-delta detection (Mode D — the mechanical floor for claim ② auto-trigger)**: FH's "undeployed-asset discovery + auto-mapping" (CLAUDE.md claim ②) previously fired **only on explicit invocation** — a new sibling repo pulled, or a task-first session in an unmapped project, was **not** self-detected (the onboarding menu is suppressed on task-first entry by the metadata-is-not-intent / task-first guards, which are load-bearing and must stay). The `scripts/fh_env_delta_scan.sh` SessionStart hook (sibling of `fh_session_load.sh`) closes this **mechanically**: it scans the projects root for git repos that are neither mapped (`tracks/{name}/`) nor wizard-done nor skip-sentineled, and emits a **one-line PROPOSAL** into turn-0 context, firing regardless of task-first entry. It **proposes only** — mapping/install stays HITL; a skipped repo is recorded via a `{name}_mapping_skipped` sentinel so it never re-nags. This is the mechanical anchor over the prose/salience layer (three-family audit 2026-07-06 rated ② PARTIAL/THEATER precisely because the auto-trigger lived only in prose).
 
 **Local-LLM offload (conditional, recommend-only)**: when Step 0 detected a local LLM runtime (Ollama / LM Studio), surface one optional item — route to `/plugin-recommender` for local-model offload tooling. FH recommends, never rebuilds (no-reinvention). Two complementary offload shapes the user picks per workload: **input-side context routing** (a small local model returns line ranges, so the cloud model receives only the dense slices instead of whole files) and **output-side generation delegation** (the local model generates and self-reviews code while the frontier model decomposes and validates). The benefit is tier-dependent — largest in headless/scripted pipelines and on weaker cloud tiers; an interactive session already triages via targeted reads. Local models suit **bounded, well-specified** work (triage, codebase explanation, instructed maintenance), not long-horizon autonomous tasks where small models loop or hallucinate — so the frontier model keeps decomposition and validation. Skip silently when no local runtime is present.
 

package/plugins/fh-meta/skills/public-surface-audit/SKILL.md

@@ -103,38 +103,16 @@ literal private token.
 
 ## Step 3. Scan
 
-For each pattern in Step 1, grep the tracked set, then drop allowlisted hits.
-
-```bash
-cd "$REPO_PATH" || exit 1
-# Build the tracked-file list once.
-git ls-files > /tmp/_psa_tracked.txt
-
-# Load your real patterns from the gitignored source (one "severity<TAB>regex" per line).
-PATTERN_SRC="${PSA_PATTERNS:-.claude/rules/.public-surface-patterns}"
-# Absent file ≠ CLEAN. An absent file is unconfigured (silent-failure risk); an EMPTY file is an
-# explicit "no tokens to protect" → CLEAN. Distinguish the two.
-[ -e "$PATTERN_SRC" ] || { echo "⚪ NOT CONFIGURED: no pattern source at $PATTERN_SRC. Create it (empty = explicit CLEAN) before trusting any verdict. Not scanning."; exit 2; }
-
-# One grep pass per pattern row; the regex comes from the file, never hardcoded here.
-while IFS=$'\t' read -r severity regex; do
-  [ -z "$regex" ] && continue
-  grep -nIE "$regex" $(cat /tmp/_psa_tracked.txt) 2>/dev/null | sed "s/^/[$severity] /"
-done < "$PATTERN_SRC"
-```
-
-For each pattern, run `grep -nIE "<regex>" $(git ls-files)`:
-- `-n` → line numbers (required for `file:line` output)
-- `-I` → skip binary files
-- `-E` → extended regex (alternation in the pattern table)
-
-Then remove any hit whose `file` + matched `token` is on the Step 2 allowlist. Do this for **every**
+For each pattern in Step 1, grep the tracked set, then drop allowlisted hits. Do this for **every**
 pattern row before producing the report — do not stop at the first HIT.
 
-**Binary / generated carve-out**: `-I` already skips binaries. Additionally note (do not auto-suppress)
-hits inside generated artifacts (e.g. `paper/*.html` exported from a private source) — these are real
-leaks on the public surface and must be reported, but the fix is "regenerate from a sanitized source",
-not "edit the HTML by hand". Flag them with a `(generated artifact)` note.
+**Binary / generated carve-out**: skip binaries. Additionally note (do not auto-suppress) hits inside
+generated artifacts (e.g. `paper/*.html` exported from a private source) — these are real leaks on the
+public surface and must be reported, but the fix is "regenerate from a sanitized source," not "edit the
+HTML by hand." Flag them with a `(generated artifact)` note.
+
+> **Detail**: See `SKILL_detail.md §Step3-Scan-Script` — the grep-per-pattern bash implementation,
+> flag reference (`-n`/`-I`/`-E`) — read when executing this step.
 
 ---
 
@@ -151,23 +129,13 @@ reports):
   `changeme`, `REDACTED`, `xxxx`, AWS-doc keys like `AKIAIOSFODNN7EXAMPLE`). A high-entropy *example* is
   not a secret.
 
-```bash
-# FP-hygiene tests the MATCHED TOKEN only — never the whole line. A line-level `grep -v` would
-# suppress a real leak that merely *mentions* an example (e.g. `user=<realname> # see EXAMPLE.md`),
-# violating PSA's "allowlist tight" rule. So extract the matched span per hit and drop it only when
-# the span is *entirely* a placeholder/example (anchored ^…$).
-PLACEHOLDER='^(<[a-z0-9_-]+>|\{project\}|EXAMPLE|dummy|changeme|REDACTED|xxxx)$'
-grep -nIE "$regex" $(cat /tmp/_psa_tracked.txt) 2>/dev/null | while IFS= read -r hit; do
-  tok=$(printf '%s' "$hit" | grep -oiE "$regex" | head -1)
-  printf '%s' "$tok" | grep -qiE "$PLACEHOLDER" && continue   # token IS a placeholder → drop
-  printf '%s\n' "$hit"
-done
-```
-
 This differs from the Step 2 allowlist: Step 2 suppresses by **file::token legitimacy**, Step 3b by
 **token value-shape**. Both run — Step 2 then Step 3b. Keep it tight (PSA's "allowlist tight" rule): if a
 token only *contains* an example substring but is otherwise a real private value, it still reports.
 
+> **Detail**: See `SKILL_detail.md §Step3b-FP-Hygiene-Script` — the token-shape-only extraction bash
+> implementation (never line-level) — read when executing this step.
+
 ---
 
 ## Step 3c. Ignore-Verification — the gitignore-mistake safeguard (root-cause, runs before Step 4)
@@ -178,39 +146,14 @@ tracked leak. For each expected-private path, `git check-ignore -v` reports whet
 by which layer** (committed `.gitignore` · local `.git/info/exclude` · global `core.excludesFile`). A
 path that resolves to no ignore rule is a MISS — surface it before it is ever `git add`ed.
 
-```bash
-# Expected-private set = conventional FH local-only files, EXTENDED with any `# private-path: <path>`
-# lines the operator added to the gitignored pattern source (self-extends per repo — not a frozen
-# operator snapshot). Built one-path-per-line + while-read so it is portable across bash AND zsh
-# (zsh does not word-split an unquoted variable, so `for f in $VAR` would break). A non-existent file
-# is skipped; an all-absent set emits n/a, never a silent pass.
-present=$({ printf '%s\n' CLAUDE.local.md .claude/rules/.public-surface-patterns \
-             .claude/rules/local_fh_context.md tracks/_meta/user_adaptation_profile.md
-           grep -E '^# private-path:' .claude/rules/.public-surface-patterns 2>/dev/null \
-             | sed -E 's/^# private-path:[[:space:]]*//'; } \
-         | awk 'NF' | sort -u | while IFS= read -r f; do [ -e "$f" ] && printf '%s\n' "$f"; done)
-[ -z "$present" ] && echo "n/a (no expected-private files present in this repo — add '# private-path:' lines to the pattern source if any exist)"
-printf '%s\n' "$present" | while IFS= read -r f; do
-  [ -z "$f" ] && continue
-  # Tracked status is tested FIRST: a file can match an ignore rule yet still be force-added
-  # (`git add -f`) — the exact ignored-but-committed mechanism behind the PR #109 leak. Tracked wins,
-  # so an ignored-but-committed file reports TRACKED (not a false-clean OK).
-  if git ls-files --error-unmatch "$f" >/dev/null 2>&1; then
-    echo "TRACKED $f (already committed — Step 3 scans its contents; un-track if it must be private: git rm --cached)"
-  elif rule=$(git check-ignore -v "$f" 2>/dev/null); then
-    echo "OK      $f → ignored by [$rule]"
-  else
-    echo "MISS    $f (exists, NOT ignored, NOT tracked — one 'git add .' from a leak; add an ignore rule)"
-  fi
-done
-```
-
 - **OK** — ignored; the report names which layer covers it (operator-private files should resolve to
   `.git/info/exclude` or `.gitignore`, never end up tracked).
 - **MISS** — exists but no ignore rule covers it → counts as **🟠 MED** in the Step 4 verdict (a latent
   leak, not yet a breach).
 - **TRACKED** — already committed: Step 3 scans its *contents*; this also flags it for un-tracking if it
-  was meant to be private.
+  was meant to be private. Tracked status is tested **before** ignore status — a file can match an
+  ignore rule yet still be force-added (`git add -f`), the exact ignored-but-committed mechanism behind
+  the PR #109 leak, so TRACKED must win over a false-clean OK.
 
 Why this is the safeguard for **gitignore mistakes** (a wrong assumption about what is ignored):
 `.gitignore` is committed/shared, `.git/info/exclude` is local/personal, and a global `core.excludesFile`
@@ -218,6 +161,10 @@ ignores across all repos — `git check-ignore -v` is the one command that says
 applies, so an "I thought it was ignored" error surfaces here instead of in a public PR (the PR #109
 class of leak). Diagnostic-only: this step never writes — it reports, the operator adds the ignore rule.
 
+> **Detail**: See `SKILL_detail.md §Step3c-Ignore-Verification-Script` — the expected-private-path
+> enumeration + `git check-ignore`/`git ls-files --error-unmatch` bash implementation, bash/zsh
+> portability note — read when executing this step.
+
 ---
 
 ## Step 4. Report

package/plugins/fh-meta/skills/public-surface-audit/SKILL_detail.md

@@ -0,0 +1,150 @@
+---
+name: public-surface-audit-detail
+description: On-demand execution detail for public-surface-audit — scan scripts, report/JSON templates, provenance.
+load: on-demand
+---
+
+## §Step3-Scan-Script
+
+```bash
+cd "$REPO_PATH" || exit 1
+# Build the tracked-file list once.
+git ls-files > /tmp/_psa_tracked.txt
+
+# Load your real patterns from the gitignored source (one "severity<TAB>regex" per line).
+PATTERN_SRC="${PSA_PATTERNS:-.claude/rules/.public-surface-patterns}"
+# Absent file ≠ CLEAN. An absent file is unconfigured (silent-failure risk); an EMPTY file is an
+# explicit "no tokens to protect" → CLEAN. Distinguish the two.
+[ -e "$PATTERN_SRC" ] || { echo "⚪ NOT CONFIGURED: no pattern source at $PATTERN_SRC. Create it (empty = explicit CLEAN) before trusting any verdict. Not scanning."; exit 2; }
+
+# One grep pass per pattern row; the regex comes from the file, never hardcoded here.
+while IFS=$'\t' read -r severity regex; do
+  [ -z "$regex" ] && continue
+  grep -nIE "$regex" $(cat /tmp/_psa_tracked.txt) 2>/dev/null | sed "s/^/[$severity] /"
+done < "$PATTERN_SRC"
+```
+
+For each pattern, run `grep -nIE "<regex>" $(git ls-files)`:
+- `-n` → line numbers (required for `file:line` output)
+- `-I` → skip binary files
+- `-E` → extended regex (alternation in the pattern table)
+
+Then remove any hit whose `file` + matched `token` is on the Step 2 allowlist. Do this for **every**
+pattern row before producing the report — do not stop at the first HIT.
+
+**Binary / generated carve-out**: `-I` already skips binaries. Additionally note (do not auto-suppress)
+hits inside generated artifacts (e.g. `paper/*.html` exported from a private source) — these are real
+leaks on the public surface and must be reported, but the fix is "regenerate from a sanitized source",
+not "edit the HTML by hand". Flag them with a `(generated artifact)` note.
+
+---
+
+## §Step3b-FP-Hygiene-Script
+
+```bash
+# FP-hygiene tests the MATCHED TOKEN only — never the whole line. A line-level `grep -v` would
+# suppress a real leak that merely *mentions* an example (e.g. `user=<realname> # see EXAMPLE.md`),
+# violating PSA's "allowlist tight" rule. So extract the matched span per hit and drop it only when
+# the span is *entirely* a placeholder/example (anchored ^…$).
+PLACEHOLDER='^(<[a-z0-9_-]+>|\{project\}|EXAMPLE|dummy|changeme|REDACTED|xxxx)$'
+grep -nIE "$regex" $(cat /tmp/_psa_tracked.txt) 2>/dev/null | while IFS= read -r hit; do
+  tok=$(printf '%s' "$hit" | grep -oiE "$regex" | head -1)
+  printf '%s' "$tok" | grep -qiE "$PLACEHOLDER" && continue   # token IS a placeholder → drop
+  printf '%s\n' "$hit"
+done
+```
+
+This differs from the Step 2 allowlist: Step 2 suppresses by **file::token legitimacy**, Step 3b by
+**token value-shape**. Both run — Step 2 then Step 3b. Keep it tight (PSA's "allowlist tight" rule): if a
+token only *contains* an example substring but is otherwise a real private value, it still reports.
+
+---
+
+## §Step3c-Ignore-Verification-Script
+
+```bash
+# Expected-private set = conventional FH local-only files, EXTENDED with any `# private-path: <path>`
+# lines the operator added to the gitignored pattern source (self-extends per repo — not a frozen
+# operator snapshot). Built one-path-per-line + while-read so it is portable across bash AND zsh
+# (zsh does not word-split an unquoted variable, so `for f in $VAR` would break). A non-existent file
+# is skipped; an all-absent set emits n/a, never a silent pass.
+present=$({ printf '%s\n' CLAUDE.local.md .claude/rules/.public-surface-patterns \
+             .claude/rules/local_fh_context.md tracks/_meta/user_adaptation_profile.md
+           grep -E '^# private-path:' .claude/rules/.public-surface-patterns 2>/dev/null \
+             | sed -E 's/^# private-path:[[:space:]]*//'; } \
+         | awk 'NF' | sort -u | while IFS= read -r f; do [ -e "$f" ] && printf '%s\n' "$f"; done)
+[ -z "$present" ] && echo "n/a (no expected-private files present in this repo — add '# private-path:' lines to the pattern source if any exist)"
+printf '%s\n' "$present" | while IFS= read -r f; do
+  [ -z "$f" ] && continue
+  # Tracked status is tested FIRST: a file can match an ignore rule yet still be force-added
+  # (`git add -f`) — the exact ignored-but-committed mechanism behind the PR #109 leak. Tracked wins,
+  # so an ignored-but-committed file reports TRACKED (not a false-clean OK).
+  if git ls-files --error-unmatch "$f" >/dev/null 2>&1; then
+    echo "TRACKED $f (already committed — Step 3 scans its contents; un-track if it must be private: git rm --cached)"
+  elif rule=$(git check-ignore -v "$f" 2>/dev/null); then
+    echo "OK      $f → ignored by [$rule]"
+  else
+    echo "MISS    $f (exists, NOT ignored, NOT tracked — one 'git add .' from a leak; add an ignore rule)"
+  fi
+done
+```
+
+Why this is the safeguard for **gitignore mistakes** (a wrong assumption about what is ignored):
+`.gitignore` is committed/shared, `.git/info/exclude` is local/personal, and a global `core.excludesFile`
+ignores across all repos — `git check-ignore -v` is the one command that says *which* rule (if any)
+applies, so an "I thought it was ignored" error surfaces here instead of in a public PR (the PR #109
+class of leak). Diagnostic-only: this step never writes — it reports, the operator adds the ignore rule.
+
+---
+
+## §Report-Template
+
+```
+public-surface-audit — Operator-Private Token Scan
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+  Target: {REPO_PATH}   |   Tracked files scanned: {N}
+
+  🔴 HIGH  ({count})
+    {file}:{line}  →  {matched token}   [class: username | company asset]
+  🟠 MED   ({count})
+    {file}:{line}  →  {matched token}   [class: absolute home path | ignore-MISS (Step 3c)]
+  🟡 LOW   ({count})
+    {file}:{line}  →  {matched token}   [class: companion-store | private wiring]
+
+  Allowlist-suppressed: {count} hit(s) (legitimate references — not leaks)
+
+  Verdict:
+    ⚪ NOT CONFIGURED — pattern source absent (nothing scanned — NOT a clean result; set up first)
+    🟢 CLEAN        — pattern source present (incl. empty), 0 HIGH + 0 MED + 0 LOW (after allowlist)
+    🟡 REVIEW       — 0 HIGH + 0 MED, LOW-only (drift, not a breach)
+    🔴 LEAK         — 1+ HIGH or 1+ MED (block publish / fix before commit)
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+---
+
+## §JSON-Schema
+
+```json
+{
+  "target": "{REPO_PATH}",
+  "tracked_files": 0,
+  "findings": [
+    {"file": "path", "line": 42, "token": "<matched>", "severity": "HIGH", "class": "username"}
+  ],
+  "counts": {"HIGH": 0, "MED": 0, "LOW": 0, "suppressed": 0},
+  "verdict": "CLEAN"
+}
+```
+
+---
+
+## §Sister-Asset-Provenance
+
+Step 3b (FP hygiene) and Step 5 (`--json`) were imported from **garrytan/gstack** `gstack-redact`
+(`lib/redact-engine.ts`) during a hands-on sister-asset cross-audit (2026-06-06; see
+`tracks/_audit/session_2026_06_06_gstack_sister_handson.md`). They are adapted to PSA's operator-IP
+ontology — `gstack-redact`'s generic secret/PII classes (AWS / PEM / JWT / hostname) stay out of PSA's
+scope (orthogonal coverage: PSA = operator-IP leak, redact = generic secret). The reverse direction
+(PSA's operator private-codename + bare-username classes, which `gstack-redact` structurally cannot
+detect) is a candidate contribution back to gstack.

package/plugins/fh-meta/skills/{skill-splitter → salience-splitter}/SKILL.md

@@ -1,15 +1,25 @@
 ---
-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.
+name: salience-splitter
+description: Splits an over-loaded always-loaded context asset — a SKILL.md, CLAUDE.md, or memory index — into a lean always-loaded layer + an on-demand layer, using a governance-semantic criterion (not length, but when the content is needed), connected by imperative pointers. Based on paper §9.5 Protocol-Priority Split pattern. Diagnoses, classifies, splits, and verifies in one pass. Renamed from skill-splitter (old name still routes here).
 user-invocable: true
 allowed-tools: ["Read", "Write", "Edit", "Bash", "Grep", "Glob"]
 model: sonnet
 ---
 
-# skill-splitter — SKILL.md Governance-Semantic Split
+# salience-splitter — Governance-Semantic Context 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.
+> **Renamed from `skill-splitter` (2026-07-05).** Old-name references still route here. The rename reflects
+> the generalized scope: the split criterion applies to **any always-loaded context asset** — a SKILL.md,
+> a CLAUDE.md, or a memory index — not only skills. The label now names the substance (salience-tiering of
+> always-loaded context), not one instance of it. Dogfood: applied to a memory index 2026-07-05.
+
+> An always-loaded asset that holds everything in one layer is not simple — it is unscoped.
+> The goal is a lean always-loaded layer + an on-demand layer, not one file and its appendix.
+
+**Scope (all three share one criterion — salience: what must be in attention every load):**
+- **SKILL.md** → always-loaded SKILL.md + on-demand `SKILL_detail.md`
+- **CLAUDE.md** → lean rules + on-demand detail docs (imperative `> **Detail**: See …` pointers)
+- **memory index** → hot `MEMORY.md` + on-demand `MEMORY_archive.md` (cold/closed/superseded entries)
 
 ## Trigger Phrases
 
@@ -19,7 +29,7 @@ model: sonnet
 | "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 |
+| `/salience-splitter` | Explicit invocation |
 
 ---
 
@@ -63,7 +73,7 @@ Step 4 — Verify
 
 > **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 §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.
 
@@ -76,7 +86,7 @@ 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 |
+| **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.
 
@@ -113,7 +123,7 @@ Run on a SKILL.md when **any one** of:
 ```
 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.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
 + phantom-quench: 0 phantoms (all §pointers resolve)
   → Fallback (skill unavailable): run §Verification-Checklist manually from SKILL_detail.md

package/plugins/fh-meta/skills/{skill-splitter → salience-splitter}/SKILL_detail.md

@@ -1,10 +1,10 @@
 ---
-name: skill-splitter-detail
-description: Detail reference for skill-splitter — content classification algorithm, pointer format variants, verification checklist. Load when executing a specific step.
+name: salience-splitter-detail
+description: Detail reference for salience-splitter — content classification algorithm, pointer format variants, verification checklist. Load when executing a specific step.
 load: on-demand
 ---
 
-# skill-splitter — Detail Reference
+# salience-splitter — Detail Reference
 
 > Load when executing a specific step. SKILL.md contains the core principle, step overview, pointer format requirement, and Done When.
 
@@ -70,13 +70,13 @@ Behavioral rules always pass this test (even if rarely triggered, the consumer n
 ### 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].
+> **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.
+> **Detail**: See `SKILL_detail.md §<SectionName-A>` (bash scripts) · `§<SectionName-B>` (format templates) — read when executing this step.
 ```
 
 ### Pointer placement rules
@@ -96,7 +96,7 @@ Advisory (risky — consumer may skip):
 
 Imperative (required form):
 ```
-"> **Detail**: See `SKILL_detail.md §Step6-Detail` — bash for STALE detection, memory scan, skill usage leaderboard — read when executing Step 6."
+"> **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: