paperlint 3.2.0 → 4.1.0

package/skills/grade-paper-writing/SKILL.md.spec.ts

@@ -10,31 +10,10 @@ export default experimental_skill({
   name: "grade-paper-writing",
   description:
     'Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — nine dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake), naming the offending sentence and the fix for each. Runs the PERSONA cold-read stall pass (a committed non-academic persona subagent, per-section) whose stall inventory — not the rubric number — is the readability gate other skills consume (pc-panel-review, paper-adversarial-review, harden-paper). Defaults to a blind multi-grader panel; after fixes, mandates the claims-preservation diff (row `claims`). NOT a content/defect review (paper-adversarial-review / pc-panel-review), NOT a structural cut plan (tighten-paper — run that first on a bloated draft), NOT venue-bar content strength (study-accepted-papers). Compose with harden-paper (which calls it), draft-paper (generative counterpart), render-paper.',
-  tools: [
-    "Read",
-    "Write",
-    "Edit",
-    "Grep",
-    "Glob",
-    "Agent",
-    "Skill",
-    "Bash(node .claude/skills/paper-pipeline/scripts/announce.mjs:*)",
-    "Bash(node .claude/skills/paper-pipeline/scripts/ledger.mjs:*)",
-  ],
+  tools: ["Read", "Write", "Edit", "Grep", "Glob", "Agent", "Skill"],
   body: `
 # grade-paper-writing — grade how the paper READS, then fix it sentence by sentence
 
-## Run me
-
-🔴 FIRST, before any other step:
-
-\`\`\`
-node .claude/skills/paper-pipeline/scripts/announce.mjs grade-paper-writing <paper-dir>
-\`\`\`
-
-An advisory pass cannot be observed failing — silence is both its error state and its normal
-state — so starting is an event, and events get written down.
-
 The review skills hunt scientific defects; \`study-accepted-papers\` checks whether the contribution clears
 the venue's bar. Neither tells you the paper is a slog to read. This skill does exactly that: it grades
 the **writing craft** against how the best-written papers actually read, and turns each low score into a
@@ -228,8 +207,7 @@ is what happened.
    side effect.
 
 7. 🔴 **Put \`WORST-SECTION: <section>\` on the FIRST LINE of the report.** Machine-readable, one
-   section, lowercase, no decoration. \`pipeline-check.mjs\` reads the last four reports and raises
-   \`repeat-finding\` when the same section is worst twice running. This exists because the abstract
+   section, lowercase, no decoration, so the next round can compare it with the last report. This exists because the abstract
    was named worst **five consecutive times**, was faithfully recorded in the scorecard all five
    times, and was never once the next task. A review loop that only appends findings converges on a
    paper whose defects are all thoroughly documented and none of them fixed. **Naming the worst
@@ -290,29 +268,6 @@ findings:
 ---
 \`\`\`
 
-## Record the verdict
-
-🔴 LAST step, once the deliverable exists:
-
-\`\`\`
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record grade-paper-writing <paper-dir> FINDING <count> <report-path>
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record grade-paper-writing <paper-dir> ABSTAINED <reason> "<one line>"
-\`\`\`
-
-**FINDING** — \`<count>\` is the number of entries in the PERSONA stall inventory, and \`<report-path>\`
-is that inventory. Add \`--blocking\` when a dimension scored 1–2, or stall density is over its cap.
-**ABSTAINED** — \`no-witness\`: the inventory came back empty. \`blocked\`: no persona reader was run,
-so there is no inventory — which is not an empty inventory and must never be filed as one.
-
-🔴 **There is no PASS**, and the distinction the two abstentions draw is the one that was being
-lost: "a reader stalled nowhere" and "nobody read it" produced the same word.
-
-🔴 **Record the stall count, never the rubric number.** The inventory is what \`pc-panel-review\`,
-\`paper-adversarial-review\` and \`harden-paper\` consume; a 44/60 in the ledger would let a paper whose
-readability gate is failing look clean to every skill downstream.
-
-\`prose-lint.mjs\` files separately as \`grade-paper-writing/prose-lint\`. One check, one row.
-
 ## Rules
 - Grade **writing, not science.** A correct paper can still score a 2 here; say so. Don't drift into
   content/defect review — that's \`paper-adversarial-review\` / \`pc-panel-review\`.

package/skills/grade-paper-writing/prose-lint.mjs

@@ -692,9 +692,8 @@ for (const f of args.filter((a) => !a.startsWith("--"))) {
     // `hedge-density`, `discourse-subject`, `undefined-coinage`.
     if (lines.length) {
       flagged += lines.length;
-      // The header STATES the count. run-mechanical.mjs prefers a stated "— N finding(s)" over
-      // counting output lines, and without it recorded header + findings + footer: one flagged
-      // caption became three findings in the ledger.
+      // The header STATES the count, so a caller need not count output lines — header and footer
+      // are not findings.
       console.error(
         `✍️  prose-lint — ${f.split("/").pop()} — ${lines.length} finding(s):`,
       );

package/skills/grade-paper-writing/prose-lint.test.ts

@@ -1,5 +1,5 @@
 /**
- * prose-lint.mjs as a process — the way the skill and `run-mechanical.mjs` run it. Every mode, on
+ * prose-lint.mjs as a process — the way the skill runs it. Every mode, on
  * fixtures written here, with the WHOLE output compared: the report is the product, so a changed
  * number or a dropped line is a changed result.
  */

package/skills/harden-paper/SKILL.md

@@ -4,7 +4,7 @@ description: Use when asking "is this paper actually ready to submit?" / "what's
 allowed-tools: [Read, Write, Edit, Grep, Glob, Bash, WebSearch, WebFetch, Agent, Skill]
 ---
 
-<!-- vigiles:sha256:7fb5e8709eb4c363 compiled from skills/harden-paper/SKILL.md.spec.ts -->
+<!-- vigiles:sha256:5283c1dbb4693721 compiled from skills/harden-paper/SKILL.md.spec.ts -->
 
 # harden-paper — the multi-axis pre-submit gate
 
@@ -15,17 +15,6 @@ allowed-tools: [Read, Write, Edit, Grep, Glob, Bash, WebSearch, WebFetch, Agent,
 > doesn't systematically hit — the submission-readiness gate. Run it after the draft is written and
 > before `submit-paper`. It does not re-implement the sub-skills; it calls them and checks the rest.
 
-## Run me
-
-🔴 FIRST, before any other step:
-
-```
-node .claude/skills/paper-pipeline/scripts/announce.mjs harden-paper <paper-dir>
-```
-
-An advisory pass cannot be observed failing — silence is both its error state and its normal
-state — so starting is an event, and events get written down.
-
 **Principle:** a paper dies from any *one* unclosed axis. Reviewers only *read* — so every claim,
 number, and omission must survive on the page. Go axis by axis; for each, **find → fix → re-verify**,
 don't just list. Below, each axis says what to run / check and the specific traps that recur.
@@ -285,33 +274,6 @@ framing, the one quotable stat, the reusable artifact, any dry-humour line and w
 list of items only the author can close (host artifact, click submit, disclose-before-public, post the
 preprint). If every gate passes and every axis is PASS/FIXED → hand off to `submit-paper`.
 
-## Record the verdict
-
-🔴 LAST step, once the deliverable exists:
-
-```
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record harden-paper <paper-dir> FINDING <count> <report-path>
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record harden-paper <paper-dir> ABSTAINED <reason> "<one line>"
-```
-
-**FINDING** — `<count>` is the number of failing gates plus still-OPEN axes and `<report-path>` the
-worst-gate-first vector. Add `--blocking` when any gate in the vector fails, as opposed to an axis
-merely being open.
-**ABSTAINED** — `no-witness`: every gate answered and no axis is open, which is the state that hands
-off to `submit-paper`. `blocked`: a gate in the vector has never been run, so the vector has a hole
-in it and there is nothing to hand off.
-
-🔴 **There is no PASS, and this skill must not reinvent one.** Its whole design is that the verdict
-is a VECTOR reported worst-gate-first, and a stored scalar acquittal was the single value that could
-flatten it. `no-witness` here means every gate answered and none of them found anything — it is not
-"ready", and the readiness call is a sentence a human writes in the report.
-
-🔴 **Never record the accept-probability as the count.** A single scalar hiding a multi-dimensional
-failure is the exact thing this skill refuses to report in prose, and a ledger row is prose that
-survives longer.
-
-The artifact-coverage script files separately as `harden-paper/artifact-coverage`. One check, one row.
-
 ## Composes with
 `tighten-paper` (axis 0, run FIRST — structure before sentences), `paper-adversarial-review` +
 `pc-panel-review` + `verify-citations` + `grade-paper-writing` (called by this), `render-paper` (if LaTeX

package/skills/harden-paper/SKILL.md.spec.ts

@@ -32,17 +32,6 @@ export default experimental_skill({
 > doesn't systematically hit — the submission-readiness gate. Run it after the draft is written and
 > before \`submit-paper\`. It does not re-implement the sub-skills; it calls them and checks the rest.
 
-## Run me
-
-🔴 FIRST, before any other step:
-
-\`\`\`
-node .claude/skills/paper-pipeline/scripts/announce.mjs harden-paper <paper-dir>
-\`\`\`
-
-An advisory pass cannot be observed failing — silence is both its error state and its normal
-state — so starting is an event, and events get written down.
-
 **Principle:** a paper dies from any *one* unclosed axis. Reviewers only *read* — so every claim,
 number, and omission must survive on the page. Go axis by axis; for each, **find → fix → re-verify**,
 don't just list. Below, each axis says what to run / check and the specific traps that recur.
@@ -302,33 +291,6 @@ framing, the one quotable stat, the reusable artifact, any dry-humour line and w
 list of items only the author can close (host artifact, click submit, disclose-before-public, post the
 preprint). If every gate passes and every axis is PASS/FIXED → hand off to \`submit-paper\`.
 
-## Record the verdict
-
-🔴 LAST step, once the deliverable exists:
-
-\`\`\`
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record harden-paper <paper-dir> FINDING <count> <report-path>
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record harden-paper <paper-dir> ABSTAINED <reason> "<one line>"
-\`\`\`
-
-**FINDING** — \`<count>\` is the number of failing gates plus still-OPEN axes and \`<report-path>\` the
-worst-gate-first vector. Add \`--blocking\` when any gate in the vector fails, as opposed to an axis
-merely being open.
-**ABSTAINED** — \`no-witness\`: every gate answered and no axis is open, which is the state that hands
-off to \`submit-paper\`. \`blocked\`: a gate in the vector has never been run, so the vector has a hole
-in it and there is nothing to hand off.
-
-🔴 **There is no PASS, and this skill must not reinvent one.** Its whole design is that the verdict
-is a VECTOR reported worst-gate-first, and a stored scalar acquittal was the single value that could
-flatten it. \`no-witness\` here means every gate answered and none of them found anything — it is not
-"ready", and the readiness call is a sentence a human writes in the report.
-
-🔴 **Never record the accept-probability as the count.** A single scalar hiding a multi-dimensional
-failure is the exact thing this skill refuses to report in prose, and a ledger row is prose that
-survives longer.
-
-The artifact-coverage script files separately as \`harden-paper/artifact-coverage\`. One check, one row.
-
 ## Composes with
 \`tighten-paper\` (axis 0, run FIRST — structure before sentences), \`paper-adversarial-review\` +
 \`pc-panel-review\` + \`verify-citations\` + \`grade-paper-writing\` (called by this), \`render-paper\` (if LaTeX

package/skills/map-prior-work/SKILL.md

@@ -5,21 +5,10 @@ context: fork
 allowed-tools: [WebSearch, WebFetch, Read, Write, Grep, Glob, Bash, Agent]
 ---
 
-<!-- vigiles:sha256:353e5a21b731a548 compiled from skills/map-prior-work/SKILL.md.spec.ts -->
+<!-- vigiles:sha256:55971644308c470f compiled from skills/map-prior-work/SKILL.md.spec.ts -->
 
 # map-prior-work — find out who already did it, while you can still change course
 
-## Run me
-
-🔴 FIRST, before any other step:
-
-```
-node .claude/skills/paper-pipeline/scripts/announce.mjs map-prior-work <paper-dir>
-```
-
-An advisory pass cannot be observed failing — silence is both its error state and its normal
-state — so starting is an event, and events get written down.
-
 The expensive failure this prevents, in one sentence: **you discover the paper that already did your
 contribution after the draft is finished, when the only move left is rewriting the claim under deadline.**
 
@@ -173,29 +162,6 @@ you looked.
 
 ---
 
-## Step 8 — record the verdict
-
-🔴 LAST step, once the deliverable exists:
-
-```
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record map-prior-work <paper-dir> FINDING <count> <report-path>
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record map-prior-work <paper-dir> ABSTAINED <reason> "<one line>"
-```
-
-**FINDING** — `<count>` is the number of rivals triaged 🔴 or 🟠 (they threaten the claim),
-`<report-path>` is the sweep file. Add `--blocking` when a rival did the contribution first and the
-claim has to change.
-**ABSTAINED** — `no-witness`: swept, and nothing found threatens what you claim. `blocked`: the
-paper has no stated contribution yet, so there is nothing to sweep against.
-
-That blocking finding is the outcome this skill was written from, discovered at T−10 days by
-accident. Recorded here it is a dated fact about the field rather than a memory of a bad night.
-
-🔴 **There is no PASS**, and this skill is where the reason is easiest to see. "Nothing threatens
-the claim" is a statement about everything that exists, made by a sweep that read a bounded list —
-so it can never be witnessed, only ever failed to be refuted. `no-witness` says that in one word,
-and the sweep's own coverage-gaps section says how far the reading actually went.
-
 ## Rules
 
 - **Mark evidence level on every claim:** `[A]` full text · `[B]` abstract/metadata · `[C]` your

package/skills/map-prior-work/SKILL.md.spec.ts

@@ -24,17 +24,6 @@ export default experimental_skill({
   body: `
 # map-prior-work — find out who already did it, while you can still change course
 
-## Run me
-
-🔴 FIRST, before any other step:
-
-\`\`\`
-node .claude/skills/paper-pipeline/scripts/announce.mjs map-prior-work <paper-dir>
-\`\`\`
-
-An advisory pass cannot be observed failing — silence is both its error state and its normal
-state — so starting is an event, and events get written down.
-
 The expensive failure this prevents, in one sentence: **you discover the paper that already did your
 contribution after the draft is finished, when the only move left is rewriting the claim under deadline.**
 
@@ -188,29 +177,6 @@ you looked.
 
 ---
 
-## Step 8 — record the verdict
-
-🔴 LAST step, once the deliverable exists:
-
-\`\`\`
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record map-prior-work <paper-dir> FINDING <count> <report-path>
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record map-prior-work <paper-dir> ABSTAINED <reason> "<one line>"
-\`\`\`
-
-**FINDING** — \`<count>\` is the number of rivals triaged 🔴 or 🟠 (they threaten the claim),
-\`<report-path>\` is the sweep file. Add \`--blocking\` when a rival did the contribution first and the
-claim has to change.
-**ABSTAINED** — \`no-witness\`: swept, and nothing found threatens what you claim. \`blocked\`: the
-paper has no stated contribution yet, so there is nothing to sweep against.
-
-That blocking finding is the outcome this skill was written from, discovered at T−10 days by
-accident. Recorded here it is a dated fact about the field rather than a memory of a bad night.
-
-🔴 **There is no PASS**, and this skill is where the reason is easiest to see. "Nothing threatens
-the claim" is a statement about everything that exists, made by a sweep that read a bounded list —
-so it can never be witnessed, only ever failed to be refuted. \`no-witness\` says that in one word,
-and the sweep's own coverage-gaps section says how far the reading actually went.
-
 ## Rules
 
 - **Mark evidence level on every claim:** \`[A]\` full text · \`[B]\` abstract/metadata · \`[C]\` your

package/skills/osf-artifact-upload/osf-artifact-upload.harness.ts

@@ -1,10 +1,9 @@
 /**
  * osf-artifact-upload — the free, deterministic tier. No model, no network, no OSF.
  *
- * WHY THIS ONE IS NOT A `checkSkill()` ONE-LINER. It carries no
- * `announce.mjs <self>` line, so by the pipeline's own membership contract it is
- * not a stage — it is a utility `submit-paper` composes with, and the shared
- * per-skill checks would (correctly) reject it.
+ * WHY THIS ONE IS NOT A `checkSkill()` ONE-LINER. It is not a stage — it is a
+ * utility `submit-paper` composes with — and what matters about it is the rules
+ * it states about its own commands, which the shared per-skill checks do not read.
  *
  * WHAT IT CHECKS: the two rules this skill states about its OWN commands, in its
  * own §gotchas — that every `curl` carries `-g` (bracketed OSF params glob and

package/skills/paper-adversarial-review/SKILL.md

@@ -2,10 +2,10 @@
 name: paper-adversarial-review
 description: Use when asking "red-team this paper" / "what would Reviewer 2 attack?" / "find the weaknesses before I submit" — ONE fast hostile-but-fair review of a paper draft. Surfaces overclaims, methodology holes, missing baselines/citations, novelty concerns, threats to validity, and desk-reject risks, with concrete fixes, led by a multi-axis 1–5 scorecard whose card composes into a pc-panel-review panel. NOT the multi-reviewer accept/reject decision (pc-panel-review), a writing grade (grade-paper-writing — this skill consumes its stall inventory rather than re-grading prose), or the full pre-submit gate (harden-paper). Especially for security and agentic-coding measurement papers.
 context: fork
-allowed-tools: [Read, Write, Grep, Glob, Agent, Skill, Bash(node .claude/skills/paper-pipeline/scripts/announce.mjs:*), Bash(node .claude/skills/paper-pipeline/scripts/ledger.mjs:*)]
+allowed-tools: [Read, Write, Grep, Glob, Agent, Skill]
 ---
 
-<!-- vigiles:sha256:2da081a03ae7fee3 compiled from skills/paper-adversarial-review/SKILL.md.spec.ts -->
+<!-- vigiles:sha256:68dcbcc1b7ab5348 compiled from skills/paper-adversarial-review/SKILL.md.spec.ts -->
 
 # paper-adversarial-review — be the reviewer who wants to reject it
 
@@ -15,17 +15,6 @@ allowed-tools: [Read, Write, Grep, Glob, Agent, Skill, Bash(node .claude/skills/
 > CFP-fit score (formerly `venue-review-sim`). Reach for an atom (this, or venue-fit mode) for a quick
 > spot-check; run the panel for the decision.
 
-## Run me
-
-🔴 FIRST, before any other step:
-
-```
-node .claude/skills/paper-pipeline/scripts/announce.mjs paper-adversarial-review <paper-dir>
-```
-
-An advisory pass cannot be observed failing — silence is both its error state and its normal
-state — so starting is an event, and events get written down.
-
 Purpose: catch what a real peer reviewer will hit, while it's still fixable. Peer review is
 **read, not run** — reviewers judge the *text, figures, and reported numbers*, rarely execute code.
 So the paper must be self-contained and airtight on its face. This skill simulates the skeptical
@@ -91,28 +80,6 @@ venue + paper type (short/full/position/benchmark). Ask for the structured outpu
 - **Strongest reframing** if the contribution is undersold or misframed.
 - **What to cut** (for length / TMI) and **what to add** (for soundness).
 
-## Record the verdict
-
-🔴 LAST step, once the deliverable exists:
-
-```
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record paper-adversarial-review <paper-dir> FINDING <count> <report-path>
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record paper-adversarial-review <paper-dir> ABSTAINED <reason> "<one line>"
-```
-
-**FINDING** — `<count>` is the number of ranked weaknesses, `<report-path>` the review with its
-scorecard. Add `--blocking` when the leaning is reject, or there is a live desk-reject risk.
-**ABSTAINED** — `no-witness`: the hostile read found nothing. `blocked`: there was no rendered
-draft to attack.
-
-🔴 **There is no PASS, and this skill is the reason the word was dangerous.** A hostile read that
-found nothing either met a very good paper or was not hostile, and the old vocabulary wrote the
-same word for both. `no-witness` says the true thing — this reader has no witness for the paper
-being sound — and leaves the second possibility standing where it belongs, in view.
-
-Record the weakness count, not the Overall 1–5. The average is exactly what this skill tells its own
-reviewer to distrust — the lowest one or two axes are the story.
-
 ## Rules
 - Be adversarial but *fair* and *specific* — "weak" is useless; "the 65% claim in the abstract isn't
   supported because the pooled bill is −1%, reword to X" is useful.

package/skills/paper-adversarial-review/SKILL.md.spec.ts

@@ -11,16 +11,7 @@ export default experimental_skill({
   description:
     'Use when asking "red-team this paper" / "what would Reviewer 2 attack?" / "find the weaknesses before I submit" — ONE fast hostile-but-fair review of a paper draft. Surfaces overclaims, methodology holes, missing baselines/citations, novelty concerns, threats to validity, and desk-reject risks, with concrete fixes, led by a multi-axis 1–5 scorecard whose card composes into a pc-panel-review panel. NOT the multi-reviewer accept/reject decision (pc-panel-review), a writing grade (grade-paper-writing — this skill consumes its stall inventory rather than re-grading prose), or the full pre-submit gate (harden-paper). Especially for security and agentic-coding measurement papers.',
   context: "fork",
-  tools: [
-    "Read",
-    "Write",
-    "Grep",
-    "Glob",
-    "Agent",
-    "Skill",
-    "Bash(node .claude/skills/paper-pipeline/scripts/announce.mjs:*)",
-    "Bash(node .claude/skills/paper-pipeline/scripts/ledger.mjs:*)",
-  ],
+  tools: ["Read", "Write", "Grep", "Glob", "Agent", "Skill"],
   body: `
 # paper-adversarial-review — be the reviewer who wants to reject it
 
@@ -30,17 +21,6 @@ export default experimental_skill({
 > CFP-fit score (formerly \`venue-review-sim\`). Reach for an atom (this, or venue-fit mode) for a quick
 > spot-check; run the panel for the decision.
 
-## Run me
-
-🔴 FIRST, before any other step:
-
-\`\`\`
-node .claude/skills/paper-pipeline/scripts/announce.mjs paper-adversarial-review <paper-dir>
-\`\`\`
-
-An advisory pass cannot be observed failing — silence is both its error state and its normal
-state — so starting is an event, and events get written down.
-
 Purpose: catch what a real peer reviewer will hit, while it's still fixable. Peer review is
 **read, not run** — reviewers judge the *text, figures, and reported numbers*, rarely execute code.
 So the paper must be self-contained and airtight on its face. This skill simulates the skeptical
@@ -106,28 +86,6 @@ venue + paper type (short/full/position/benchmark). Ask for the structured outpu
 - **Strongest reframing** if the contribution is undersold or misframed.
 - **What to cut** (for length / TMI) and **what to add** (for soundness).
 
-## Record the verdict
-
-🔴 LAST step, once the deliverable exists:
-
-\`\`\`
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record paper-adversarial-review <paper-dir> FINDING <count> <report-path>
-node .claude/skills/paper-pipeline/scripts/ledger.mjs record paper-adversarial-review <paper-dir> ABSTAINED <reason> "<one line>"
-\`\`\`
-
-**FINDING** — \`<count>\` is the number of ranked weaknesses, \`<report-path>\` the review with its
-scorecard. Add \`--blocking\` when the leaning is reject, or there is a live desk-reject risk.
-**ABSTAINED** — \`no-witness\`: the hostile read found nothing. \`blocked\`: there was no rendered
-draft to attack.
-
-🔴 **There is no PASS, and this skill is the reason the word was dangerous.** A hostile read that
-found nothing either met a very good paper or was not hostile, and the old vocabulary wrote the
-same word for both. \`no-witness\` says the true thing — this reader has no witness for the paper
-being sound — and leaves the second possibility standing where it belongs, in view.
-
-Record the weakness count, not the Overall 1–5. The average is exactly what this skill tells its own
-reviewer to distrust — the lowest one or two axes are the story.
-
 ## Rules
 - Be adversarial but *fair* and *specific* — "weak" is useless; "the 65% claim in the abstract isn't
   supported because the pooled bill is −1%, reword to X" is useful.

package/skills/paper-pipeline/PIPELINE-MAP.md

@@ -200,7 +200,7 @@ format.
 ## 3. Venue constants belong in a card, not in code
 
 > **Done (#79, 2.1.0), in a different shape than planned below.** The constants live in **venue
-> presets**, JSONC files in `submit-paper/references/venues/`: `acm-sigconf.jsonc` is the ACM
+> presets**, JSONC files in the package's `presets/`: `acm-sigconf.jsonc` is the ACM
 > family (the `format` block — page size, columns, fonts, font sizes — plus the `tex` packages),
 > `agenticdev.jsonc` and `aisec.jsonc` extend it and add their `kinds` (page limits) and `rules`,
 > `realm.jsonc` stands alone. A paper picks one in `<paper>/paperlint.json`
@@ -332,7 +332,7 @@ point. Nothing to check.
 | page limit           | **4**           | `venues/realm.md` 8 · `build-submission.sh` 8 · `PIPELINE-STATUS` "8/8" · the build prints **9**                                      |
 | the TeX package list | **5**           | `ensure-toolchain.sh` (executable) · `SKILL.md` · `render-paper.harness.mjs` (pins it ✅) · `build-submission.sh` · `PIPELINE-STATUS` |
 
-The canonical source is the venue preset (`venues/<name>.jsonc`, resolved through its `extends`
+The canonical source is the venue preset (`presets/<name>.jsonc`, resolved through its `extends`
 chain): its `format` block for the numbers (page size, columns, fonts, font sizes, and `kinds` for
 page limits), its `tex` block (plus `tex-base.jsonc`) for packages — since 2026-09-24;
 `ensure-toolchain.sh` held them until then — and its `rules` block for the checks the venue

package/skills/paper-pipeline/SKILL.md

@@ -4,7 +4,7 @@ description: The orchestrator for writing a research paper end-to-end, from idea
 allowed-tools: [Read, Write, Edit, Grep, Glob, Bash, WebSearch, WebFetch, Agent, Skill]
 ---
 
-<!-- vigiles:sha256:291d2266e29874be compiled from skills/paper-pipeline/SKILL.md.spec.ts -->
+<!-- vigiles:sha256:c3eea3ac60e9c816 compiled from skills/paper-pipeline/SKILL.md.spec.ts -->
 
 # paper-pipeline — the conductor for the whole organism
 
@@ -30,7 +30,7 @@ of that — who is filing what, and when — stays in the author's own private n
   space, anonymity, OpenReview mechanics. Fetch once, not per paper. Venue-specific facts stay in
   `submit-paper/references/venues/<venue>.md`. (No prose equivalent yet for ACM or IEEE — write one
   when the next paper goes there. ACM's machine-checked format already exists: the
-  `paperlint:acm-sigconf` preset, `submit-paper/references/venues/acm-sigconf.jsonc`.)
+  `paperlint:acm-sigconf` preset, `presets/acm-sigconf.jsonc` in the package.)
 
 ## The readiness scorecard (which boxes are checked)
 Stages run in subagents and report into chat, which evaporates — so "did everything run?" becomes
@@ -44,30 +44,6 @@ their grade/result, what's open,* and a one-line **submit-ready verdict**. Rules
 - **"Is paper X ready / what's checked?"** = read this file (verdict + any `☐`/`◐`/`⚠` rows), don't
   reconstruct from chat.
 - Distinct from `SUBMIT-CHECKLIST.md` (venue-format compliance); both must be green before upload.
-- 🔴 **Enforcement leg (prose-isn't-policy), and it is a script, not a paragraph.**
-  `scripts/pipeline-check.mjs` reads this file on every paper edit and reports the failures that
-  actually happened on real papers: a CONTINUOUS check whose last run predates the current text, a GATE
-  that ran without its required input, **a `Requires` cell that omits an edge the pipeline declares**
-  (the canonical set is `scripts/pipeline-edges.mjs`, with the admitting sentence per edge), a study
-  that ran with no stated claim, the `access` row not green inside the moderation window, and a verdict
-  line that is just an accept probability. Run it directly
-  with `node .claude/skills/paper-pipeline/scripts/pipeline-check.mjs <paper-dir>`; it is wired into
-  `.claude/hooks/paper-status-gates.sh` and asserted in `.claude/hooks/hooks.harness.mjs`.
-  **The grading itself stays judgment** (that is the skills' job) — only the mechanically checkable
-  parts are compiled. *Added 2026-08-03, the day three hooks in this repo were found to have been dead
-  for weeks: an advisory that is never exercised is indistinguishable from one that works.*
-
-  🔴 **As of 2026-08-26, this file is read by TWO things.** Sixteen checks over it live as
-  `@eslint/markdown` rules in `eslint-rules/pipeline-status.mjs` (run:
-  `npx eslint <paper-dir>/PIPELINE-STATUS.md`; five of them are `error`), while
-  `scripts/pipeline-check.mjs` holds six whose input lies OUTSIDE the file — git, the clock, the
-  `reviews/` directory, `process.env`. The classification of all 22 lives in the
-  author's private notes (`<papers-root>/research/2026-08-26-klassifikatsiya-pipeline-check.md`).
-
-  ⚠️ The paragraph above lived for a day NOT HERE but directly in `SKILL.md` — i.e. in the
-  generated file. The very first `vigiles compile` (27.08) ate it, and it was not a human who
-  noticed but an agent whose compile had just erased someone else's edit. **An edit to `SKILL.md`
-  without `.spec.ts` is an edit with an expiration date: the next compile.**
 
 ## 🗺️ Where every check actually runs — and where it can be walked around
 
@@ -133,9 +109,7 @@ Prior art is named in the file header (`adewale/skill-eval-harness` does this to
 
 🔴 **Still manual, and no mechanism has been found for it:** the **cold read** — a reader with no
 context restating each sentence — is the only check that catches a sentence that is short, true,
-threshold-passing and meaningless. It needs a fresh subagent, so no hook can run it. What IS
-mechanised is noticing that it is *overdue*: `pipeline-check.mjs` reports `stale-cold-read` when the
-prose changed after the last one.
+threshold-passing and meaningless. It needs a fresh subagent, so no hook can run it.
 
 Install the git hook once per clone: `git config core.hooksPath .githooks`.
 

package/skills/paper-pipeline/SKILL.md.spec.ts

@@ -47,7 +47,7 @@ of that — who is filing what, and when — stays in the author's own private n
   space, anonymity, OpenReview mechanics. Fetch once, not per paper. Venue-specific facts stay in
   \`submit-paper/references/venues/<venue>.md\`. (No prose equivalent yet for ACM or IEEE — write one
   when the next paper goes there. ACM's machine-checked format already exists: the
-  \`paperlint:acm-sigconf\` preset, \`submit-paper/references/venues/acm-sigconf.jsonc\`.)
+  \`paperlint:acm-sigconf\` preset, \`presets/acm-sigconf.jsonc\` in the package.)
 
 ## The readiness scorecard (which boxes are checked)
 Stages run in subagents and report into chat, which evaporates — so "did everything run?" becomes
@@ -61,30 +61,6 @@ their grade/result, what's open,* and a one-line **submit-ready verdict**. Rules
 - **"Is paper X ready / what's checked?"** = read this file (verdict + any \`☐\`/\`◐\`/\`⚠\` rows), don't
   reconstruct from chat.
 - Distinct from \`SUBMIT-CHECKLIST.md\` (venue-format compliance); both must be green before upload.
-- 🔴 **Enforcement leg (prose-isn't-policy), and it is a script, not a paragraph.**
-  \`scripts/pipeline-check.mjs\` reads this file on every paper edit and reports the failures that
-  actually happened on real papers: a CONTINUOUS check whose last run predates the current text, a GATE
-  that ran without its required input, **a \`Requires\` cell that omits an edge the pipeline declares**
-  (the canonical set is \`scripts/pipeline-edges.mjs\`, with the admitting sentence per edge), a study
-  that ran with no stated claim, the \`access\` row not green inside the moderation window, and a verdict
-  line that is just an accept probability. Run it directly
-  with \`node .claude/skills/paper-pipeline/scripts/pipeline-check.mjs <paper-dir>\`; it is wired into
-  \`.claude/hooks/paper-status-gates.sh\` and asserted in \`.claude/hooks/hooks.harness.mjs\`.
-  **The grading itself stays judgment** (that is the skills' job) — only the mechanically checkable
-  parts are compiled. *Added 2026-08-03, the day three hooks in this repo were found to have been dead
-  for weeks: an advisory that is never exercised is indistinguishable from one that works.*
-
-  🔴 **As of 2026-08-26, this file is read by TWO things.** Sixteen checks over it live as
-  \`@eslint/markdown\` rules in \`eslint-rules/pipeline-status.mjs\` (run:
-  \`npx eslint <paper-dir>/PIPELINE-STATUS.md\`; five of them are \`error\`), while
-  \`scripts/pipeline-check.mjs\` holds six whose input lies OUTSIDE the file — git, the clock, the
-  \`reviews/\` directory, \`process.env\`. The classification of all 22 lives in the
-  author's private notes (\`<papers-root>/research/2026-08-26-klassifikatsiya-pipeline-check.md\`).
-
-  ⚠️ The paragraph above lived for a day NOT HERE but directly in \`SKILL.md\` — i.e. in the
-  generated file. The very first \`vigiles compile\` (27.08) ate it, and it was not a human who
-  noticed but an agent whose compile had just erased someone else's edit. **An edit to \`SKILL.md\`
-  without \`.spec.ts\` is an edit with an expiration date: the next compile.**
 
 ## 🗺️ Where every check actually runs — and where it can be walked around
 
@@ -150,9 +126,7 @@ Prior art is named in the file header (\`adewale/skill-eval-harness\` does this
 
 🔴 **Still manual, and no mechanism has been found for it:** the **cold read** — a reader with no
 context restating each sentence — is the only check that catches a sentence that is short, true,
-threshold-passing and meaningless. It needs a fresh subagent, so no hook can run it. What IS
-mechanised is noticing that it is *overdue*: \`pipeline-check.mjs\` reports \`stale-cold-read\` when the
-prose changed after the last one.
+threshold-passing and meaningless. It needs a fresh subagent, so no hook can run it.
 
 Install the git hook once per clone: \`git config core.hooksPath .githooks\`.
 

package/skills/paper-pipeline/grade-paper-writing-ablation.eval.mjs

@@ -226,7 +226,7 @@
  *
  *      ⚠️ 2026-08-09: A0 IS TWO TOKENS OFF THE TEXT MEASURED BELOW. The scorecard row ids were
  *      renamed from two-letter codes to words (`Hn` → `claims`) because the two-letter space had
- *      collided — see `scripts/pipeline-edges.mjs`. The target's description carried one such
+ *      collided. The target's description carried one such
  *      reference, so "…mandates the Hn claim-preservation diff." became "…mandates the
  *      claims-preservation diff (row `claims`)." A0 is still read from disk and every guard still
  *      passes; the numbers below were measured on the pre-rename bytes and are left as recorded.