@chrono-meta/fh-gate 1.4.27 → 1.4.29

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@chrono-meta/fh-gate",
-  "version": "1.4.27",
+  "version": "1.4.29",
   "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/agent-composer/SKILL.md

@@ -85,7 +85,7 @@ Execution method · agent selection · recording location → infer and proceed
 | "Do X" — target and completion criteria are clear | Clear | Immediately enter Step 0-b |
 | "Improve X" — improvement dimension unclear | Direction unclear | Ask for clarification |
 | "Something feels off" — diagnosis target unclear | Execution unclear → infer | Auto-dispatch harness-doctor |
-| "Run the pipeline" — trigger is clear | Clear | Run harvest-loop based on recent work |
+| "Wrap up / harvest this session" — trigger is clear | Clear | Run harvest-loop based on recent work ("run the pipeline" itself is ceded to pipeline-conductor — route end-to-end sweeps there, not here) |
 | No request — session context exists | Execution unclear → infer | Auto-compose based on CATALOG/MEMORY, then confirm |
 
 > **Detail**: See `SKILL_detail.md §Clarification` — 2-question format, 4-question structured confirmation, meta-prompting intervention — read when direction ambiguity requires a clarification block.

package/plugins/fh-meta/skills/phantom-quench/SKILL_detail.md

@@ -106,12 +106,42 @@ literal span — never on a model's agreement.
 
 | Citation form | Resolved URL |
 |---|---|
-| `arXiv:NNNN.NNNNN` / `arXiv:NNNN.NNNNNvK` | `https://arxiv.org/abs/NNNN.NNNNN` |
+| `arXiv:NNNN.NNNNN` / `arXiv:NNNN.NNNNNvK` | `https://arxiv.org/abs/NNNN.NNNNN` (abstract — the **canonical identifier** surface). For a section-/body-cited claim, prefer full text **first**, falling back in order `https://arxiv.org/html/NNNN.NNNNN` → `https://arxiv.org/pdf/NNNN.NNNNN` → `/abs/` (HTML is a *partial* backfill — see *Surface selection* below) |
 | bare DOI `10.xxxx/...` | `https://doi.org/10.xxxx/...` |
 | `http(s)://...` | use as-is (HTTPS-upgrade handled by WebFetch) |
 | Named source, no URL ("paper X shows Y") | **one** `WebSearch` to locate the canonical URL → then WebFetch it. Do **not** verify against the search snippet alone — the snippet is not the source. |
 | version token `pkg x.y.z` | the package registry/release page (npm/PyPI/GitHub releases) for that exact version |
 
+**Surface selection — the abstract is the wrong surface for a §-cited mechanism.** The `/abs/` page
+carries only the abstract; a claim that cites a section or attributes a *named body-level mechanism* to
+the paper (e.g. "§4.2 stale-but-confident", "the paper's Change Manifest") returns NONE / MENTIONS on
+`/abs/` even when fully supported in the body — an **artificial Unsupported** (measured 2026-06-17
+in-the-wild dogfood: a `/abs/`-only fetch falsely flagged a body-supported claim; it resolved only after
+escalating to full text). Two mechanical, judgment-free rules cover this:
+
+1. **Mechanical first-surface trigger** — if the claim text carries a section/figure marker (`§N`,
+   "section N", "Table N", "Figure N") → fetch full text **first**. Otherwise fetch `/abs/` first. (Do
+   *not* pre-decide "is this a body-level mechanism?" — that is a salience-dependent judgment; the
+   `§`-marker is the only mechanical signal, and the NONE-escalation below catches the rest.)
+2. **Abstract-NONE escalates** (the self-correcting backstop) — *any* `/abs/` fetch that returns
+   NONE/MENTIONS escalates once to full text before an Unsupported verdict (see Decision rules). A true
+   headline claim already returned ASSERTS on `/abs/`, so it never escalates; only a NONE does, and the
+   escalation is cheap insurance. This subsumes rule 1's misses — a body claim that *didn't* carry a `§`
+   marker still gets full text via its abstract-NONE.
+
+**HTML is a partial backfill — fall back, never Phantom.** arXiv HTML exists only for LaTeX-source
+papers (roughly post-2023) and some conversions fail, so `/html/{id}` can 404 for a perfectly real
+paper. A 404 on `/html/` (or `/pdf/`) is **surface-absence, not identifier-absence**: fall back
+`/html/` → `/pdf/` → `/abs/`. Only a 404 on **`/abs/`** (the canonical identifier surface) is Phantom.
+
+**Anchor the full-text fetch.** Full text is large; pass the cited section number / mechanism term *into*
+the WebFetch prompt (`"…the span in §4.2 that states…"`) so the span search is anchored, not whole-paper
+— this makes the `§`-citation the retrieval anchor, not merely a routing flag, and guards against a long
+paper's relevant span falling outside the fetch model's attention (a full-text false-NONE).
+
+Reserve `/abs/` for claims the abstract itself states (headline numbers, the paper's stated contribution
+— e.g. a top-line stat like "39–77% factual support" is abstract-level and `/abs/` is correct).
+
 **WebFetch prompt template** (ask for a *span*, not a *verdict* — this keeps the anchor non-model):
 
 ```
@@ -126,6 +156,12 @@ Polarity is load-bearing: a page saying "X does NOT hold" or "earlier work claim
 contains a span lexically matching the claim. Asking for the label keeps the *retriever* mechanical while
 surfacing the stance the *judge* needs — a NEGATES/MENTIONS span is **not** grounding.
 
+**Coinage at fetch time:** when the claim's key term is a known FH coinage (a name FH invented, not the
+paper's — e.g. "Change Manifest", "Validation Gate"), pass the *mechanism description* — not the coinage
+string — into the `<exact claim>` slot, so the retriever searches for the **relation**, not a label that
+isn't on the page. The Coinage-vs-source-term decision rule (below) then only adjudicates the returned
+span; it never has to overturn a NONE the prompt itself caused.
+
 **Span-evidence format** (a Grounded external verdict is invalid without this):
 
 ```
@@ -152,7 +188,16 @@ Grounded: N / Unsupported: N / Unreachable: N / Phantom: N
 - Span labelled **NEGATES or MENTIONS** → **Unsupported 🟠** (a negating or merely-mentioning span is not
   grounding — the false-Grounded trap).
 - Fetch succeeded, content readable, span = NONE or off-claim → **Unsupported 🟠** (cited-but-not-verified).
-- Identifier does not resolve at all (404 on a specific arXiv id, dead DOI, fabricated) → **Phantom ❌**.
+  **Surface-escalation first — abstract-NONE is the mechanical trigger (no judgment):** if the only
+  surface fetched was the abstract (`/abs/`), re-fetch full text (fall back `/html/{id}` → `/pdf/{id}`)
+  before classifying — *any* abstract-NONE escalates once, you do **not** first decide whether the claim
+  "looks body-level" (that judgment is the salience leak this rule removes). A body-supported claim must
+  not be failed on an abstract-only NONE. This is the second-surface escalation, **extended from
+  Unreachable to Unsupported-on-abstract** (2026-06-17 dogfood finding #2). Only after a full-text surface
+  *also* returns NONE/MENTIONS is the verdict Unsupported. (Single-pass: escalate once, then classify — no loop.)
+- Identifier does not resolve on the **canonical** surface (`/abs/` 404, dead DOI, fabricated id) →
+  **Phantom ❌**. A 404 on `/html/` or `/pdf/` while `/abs/` resolves is **surface-absence, not
+  identifier-absence** — fall back per *Surface selection*, never Phantom.
 - Identifier plausibly real but fetch blocked in this environment (paywall/403/timeout/cross-host) →
   **Unreachable ⏳** — provisional; note the limit, route to a second surface or the human gate; never auto-Phantom.
 - `WebFetch`/`WebSearch` **absent/disabled in the environment** (not one source — the capability) → Step 2-E
@@ -160,6 +205,40 @@ Grounded: N / Unsupported: N / Unreachable: N / Phantom: N
   (do not mark every claim Unreachable + CONDITIONAL_PASS — that falsely implies grounding was attempted).
 - **Never** upgrade NONE to Grounded because a second model "thinks it's probably right" (agreement bias).
 
+**Coinage-vs-source-term (the external analogue of Step 2's format normalization — but *judged*, not
+mechanical).** FH sometimes attributes a *coined* name to an external source ("the paper's Change
+Manifest", "its Validation Gate") when the paper uses different words for the same mechanism. A name
+mismatch alone is not automatically Unsupported — but the path to rescuing it is **deliberately narrow**,
+because this is one rationalization away from the agreement-bias trap above. The failure direction must
+stay conservative (over-flag, never false-Grounded):
+
+- **Possessive/attributive phrasing is presumed a terminology claim.** "the paper's X", "its X",
+  "X, per [paper]" → literal absence of the coinage on the source = **Unsupported 🟠**, *unless the FH
+  artifact separately states the underlying relation in non-coinage words* (so the mechanism is
+  independently legible without the label). A runner may **not** reclassify a bare "the paper's X" as a
+  mechanism claim on its own reading — the mechanism claim must already be spelled out **in the artifact**
+  before the span-judged path opens. (Closes the laundering move: relabeling a terminology claim as a
+  mechanism claim to save it.)
+- **When the mechanism IS spelled out, ground on the relation, not the string — mechanical de-label
+  test:** blank out *both* labels (FH's coinage and the source's term); the span must still literally
+  state the operative relation the claim asserts — the same inputs→outputs / the same conditional / the
+  same action. If, with both names blanked, the span no longer states the claim, the equivalence was
+  reader-supplied → **Unsupported 🟠**. (2026-06-17 dogfood: SkillOpt's "validation score" /
+  "rejected-edit buffer" survives the de-label test for FH's "Validation Gate" claim — the span states
+  the accept-only-on-improvement relation in its own words.)
+- **No self-granted Grounded on a contested match.** If the de-label test is contested rather than clean,
+  the runner may **not** write Grounded — the *only* path to Grounded for a disputed term-match is
+  **escalate to `/steel-quench` Wave 1** and let the isolated adversary surface the span or reject it.
+  Default otherwise = Unsupported. (Removes the runner's discretion to self-select the lenient branch.)
+- **Recommended conservative resolution (publish-facing).** For a publish-facing citation the safer fix
+  is to require the artifact to **quote the paper's own term** ("what SkillOpt calls a 'validation
+  score'") instead of asserting FH's coinage as the paper's — removing the judgment entirely. Reserve the
+  judged de-label path for low-stakes internal claims; match effort to stakes.
+
+Unlike Step 2's value-format normalization (mechanical: `300s` ≡ `300 seconds`), coinage-vs-term is
+**judged** — Grounded only via a clean de-label span or a passed steel-quench escalation, never a
+reader's semantic equivalence assertion.
+
 ---
 
 ## §Step3-Detail

package/plugins/fh-meta/skills/self-marketing-lint/SKILL.md

@@ -13,117 +13,18 @@ deprecated_date: 2026-06-02
 successor: harness-doctor
 ---
 
-# self-marketing-lint — FH Self-Marketing Language Detection + Replacement Suggestions
+# self-marketing-lint — DEPRECATED
 
-> Detection criteria are defined inline in this skill (see "Inline Baseline" below).
-> The skill is self-contained and does not depend on external personal-memory documents.
+This skill was absorbed into **harness-doctor** on 2026-06-02. Its self-marketing /
+cushion-word / version-brag detection patterns now live in **harness-doctor Step 3-L
+(`--lint` mode)**, which carries the canonical baseline.
 
-## Triggers
-
-- `/self-marketing-lint`
-- "Check FH files", "remove marketing language", "description diet"
-- "Catch self-marketing language", "run the lint"
-- When surfaced as a HIGH item in harvest-loop (P10 series)
-
-## Detection Patterns (Inline Baseline)
-
-The following criteria are self-contained inside this skill. No external personal-memory file is required.
-
-### Self-Marketing Anti-Patterns
-
-1. **Self-declarations of quality**: "industry-leading", "world-class", "production-ready", "the world's first" without external evidence
-2. **Round-counters and version brags**: "after 26 iterations", "v0.5 maturity", "22nd naming round", "prototype 9th iteration" — irrelevant to function
-3. **Cushion language**: "perfectly", "easily", "seamlessly", "instantly", "without burden" — content-free intensifiers
-4. **Owner self-reference loops**: "validated by our team", "proven internally" — circular evidence with no external check
-5. **Marketing word stacking**: 3+ adjectives in one sentence
-6. **Spec-only artifacts**: skills/agents missing a Done When section — declaration without observable output
-
-| Pattern type | Examples (detection targets) | Replacement direction |
-|---|---|---|
-| Version labels | "v0.3", "latest version", "(added 2026-05-24)" | Remove (git history manages versioning) |
-| Emphasis words | "fully automated", "instantly", "automatically", "full coverage" | Replace with functional description |
-| Iteration counts | "22nd naming", "26th round", "prototype 9th iteration" | Remove or move to external document |
-| Self-declarations | "This skill is the world's first", "Core FH skill" | Delete |
-| Cushion language | "easily", "conveniently", "without burden" | Replace with behavior-based description |
-| Spec-only (no outputs) | Skills missing Done When | Require adding an outputs section |
-
-## Execution Steps
-
-### Step 1. Scan Target Files
-
-```bash
-# full list of skill files
-find {FH_ROOT}/plugins -name "SKILL.md" | sort
-
-# full list of agent files
-find {FH_ROOT}/.claude/agents -name "*.md" | sort
-
-# docs/ public documents
-find {FH_ROOT}/docs -name "*.md" | sort
-```
-
-Scope: specified files only if target is given / full scan if unspecified.
-
-### Step 2. Confirm Inline Baseline
-
-Re-read the "Detection Patterns (Inline Baseline)" section above. No external file load is required — the skill is self-contained.
-
-### Step 3. Pattern Detection
-
-Grep detection patterns in each file's description field + body.
-
-```bash
-# version / iteration count patterns
-grep -n "v[0-9]\+\.[0-9]\|[0-9]\+th\|[0-9]\+th iteration\|prototype [0-9]\+" {file}
-
-# emphasis word patterns
-grep -n "fully\|instantly\|full coverage\|automatically\|conveniently\|easily\|without burden" {file}
-
-# self-declaration patterns
-grep -n "core skill\|world's first\|best\|most\|top" {file}
-```
-
-### Step 4. Output Replacement Suggestions
-
-Organize and output detection results per file.
-
-```
-File: {path}
-  L{line}: "{original text}"
-  → Suggested: "{replacement text}" (reason: {pattern type})
-  → Or: recommend removal (reason: {pattern type})
-```
-
-**No automatic edits**: Output suggestions only. Actual edits proceed after user confirmation.
-
-### Step 5. Summary Report
+**Use instead:**
 
 ```
-## self-marketing-lint results
-
-Target: {N} files
-Detected: {N} items
-
-| File | Detection count | Most common pattern |
-|---|---|---|
-...
-
-Recommendation: prioritize files with highest detection counts for review
+/harness-doctor --lint
 ```
 
-## Done When
-
-```
-1. Full scan of target files complete
-2. Inline baseline criteria applied (no external file dependency)
-3. Detection results organized and output per file
-4. No automatic edits — user confirmation gate maintained
-```
-
-## Connections
-
-| Situation | Connection |
-|---|---|
-| harvest-loop HIGH P10 series triggered | auto-suggest `/self-marketing-lint` |
-| During cold audit | `/steel-quench` + self-marketing-lint in parallel |
-| Before external distribution | `fh-meta:hub-persona-auditor` followed by self-marketing-lint |
+Triggers that previously reached this skill (`check FH files`, `remove marketing
+language`, `description diet`) now route to harness-doctor Step 3-L. This stub remains
+only so old references resolve.