paperlint 2.0.0

package/lib/skill-trigger-cases.mjs

@@ -0,0 +1,446 @@
+/**
+ * skill-trigger-cases.mjs — the per-skill trigger cases, in ONE place.
+ *
+ * WHY A SHARED MODULE AND NOT 21 FILES OF PROMPTS. Coverage is decided by
+ * placement (vigiles, 2026-08-11), so every skill needs a test file inside its own
+ * directory. That rule is right — a test that merely names a surface was crediting
+ * surfaces nothing touched — but satisfying it by copying prompts 21 times would
+ * recreate the drift the rule exists to prevent. So the colocated files are thin
+ * and the DATA lives here, reviewed as one table where collisions are visible.
+ *
+ * ── HOW TO READ A CASE ──────────────────────────────────────────────────────────
+ *
+ *   skill     the skill under test
+ *   why       the territory it owns, in one line — the thing a sibling must NOT take
+ *   prompts   should-fire. FOUR each, deliberately mixed (below)
+ *   collides  siblings whose own should-fire prompts become this case's
+ *             irrelevantPrompts. That is how "the WRONG skill must not fire" is
+ *             asserted, and it is symmetric: each side of a collision is measured
+ *             from both directions.
+ *
+ * 🔴 THE PROMPTS ARE MIXED ON PURPOSE, AND THE MIX DECIDES THE NUMBER.
+ * `framing-vs-vocabulary.eval.mjs` measured the same skills two ways: prompts that
+ * NAME the action scored 78%, prompts that describe a SITUATION scored 25%. A set
+ * written only in the description's own verbs would report a flattering rate and
+ * measure how well the prompts were copied from the description. So each case
+ * carries both kinds — a couple that name the action, a couple that describe the
+ * situation the way the author actually types, in the language he actually types.
+ * Expect the aggregate to sit BELOW a same-vocabulary set. That is the point.
+ *
+ * Russian is over-represented on purpose too: `pipeline-language.eval.mjs`
+ * measured English 78.1% vs Russian 60.4% over 32 matched pairs (paired sign test
+ * p = 0.0074), and every description here is English while the author writes
+ * Russian. A set that quietly dropped the Russian prompts would hide the gap.
+ *
+ * The eight cases below marked HARVESTED were written for
+ * `paper-pipeline/pipeline-firing.eval.mjs` and moved here unchanged, so the two
+ * runners cannot drift apart. The thirteen marked NEW were added 2026-08-11 to
+ * finish the corpus; they have never been run, so their rates are UNKNOWN rather
+ * than assumed good.
+ */
+
+const PACKAGE_CASES = [
+  // ── HARVESTED (written for pipeline-firing.eval.mjs, measured) ───────────────
+  {
+    skill: "tighten-paper",
+    why: "structural bloat — cut/fold/merge, NOT sentence craft",
+    prompts: [
+      "статья раздулась, середина провисает — что резать?",
+      "this draft is 14 pages against a 9 page limit, what goes",
+      "после трёх раундов ревью там одна вода и хеджи, нужен план сокращения",
+      "sections 4 and 5 say the same thing twice, and nobody would skim any of it",
+    ],
+    collides: ["grade-paper-writing", "argument-arc"],
+  },
+  {
+    skill: "grade-paper-writing",
+    why: "prose craft — the sentence is the unit, NOT the section",
+    prompts: [
+      "оцени как написано — читается как стена жаргона",
+      "is the writing any good or does it read like shit",
+      "abstract звучит криво хотя по смыслу всё на месте, дай оценку прозе",
+      "grade the craft: title, abstract, sentence clarity, hedge stacking",
+    ],
+    collides: ["tighten-paper", "argument-arc"],
+  },
+  {
+    skill: "argument-arc",
+    why: "argument architecture — does one conclusion become inevitable",
+    prompts: [
+      "ревьюер второй раз пишет что мы кидаем в него идеи без связи",
+      "does the paper actually carry a reader to one conclusion or just list stuff",
+      "мы вводим пять именованных штук и три числа — по-моему это перебор",
+      "before the big rewrite I want one sentence per section, bottom up",
+    ],
+    collides: ["tighten-paper", "grade-paper-writing"],
+  },
+  {
+    skill: "paper-adversarial-review",
+    why: "ONE hostile reviewer, fast",
+    prompts: [
+      "red-team эту статью, чем будет бить reviewer 2",
+      "would reviewer 2 buy this claim about the hook finding",
+      "найди слабые места до сабмита — один злой но честный рецензент",
+      "what is our desk reject risk and where do we overclaim",
+    ],
+    collides: ["pc-panel-review"],
+  },
+  {
+    skill: "pc-panel-review",
+    why: "the WHOLE committee + an accept probability, not one reviewer",
+    prompts: [
+      "какая вероятность принятия у этой статьи, если честно",
+      "simulate the whole program committee, not one reviewer",
+      "что решат на PC discussion — accept или reject",
+      "нужно несколько независимых ревьюеров с разными линзами плюс мета-ревью от чейра",
+    ],
+    collides: ["paper-adversarial-review"],
+  },
+  {
+    skill: "cold-read-diff",
+    why: "fires AFTER a prose edit — the reader with no context",
+    prompts: [
+      "я переписал третий абзац intro — проверь что предложения вообще что-то значат",
+      "just edited the threats section, would a reader with no context get it",
+      "поправил формулировки в 4.2, прогони свежим читателем до того как я закрою правку",
+      "эти предложения короткие, правдивые и всё равно непонятно что они утверждают",
+    ],
+    collides: ["grade-paper-writing", "tighten-paper"],
+  },
+  {
+    skill: "verify-citations",
+    why: "are the cites REAL — a pre-submit metadata gate",
+    prompts: [
+      "проверь что все цитаты настоящие перед сабмитом",
+      "did we hallucinate any of these refs",
+      "сверь метаданные по каждому cite — год, венью, авторы, doi",
+      "one bibtex entry looks invented to me, check the whole bibliography",
+    ],
+    collides: ["map-prior-work"],
+  },
+  {
+    skill: "map-prior-work",
+    why: "who already did this — BEFORE drafting, reshapes the contribution",
+    prompts: [
+      "кто уже это сделал до нас — хочу знать до того как начну писать",
+      "sweep the landscape: everyone working on this, prior versus concurrent",
+      "нужен скелет related work и вердикт что мы ещё можем клеймить своим",
+      "find every competing group in this space and date them against our submission",
+    ],
+    collides: ["verify-citations"],
+  },
+
+  // ── NEW 2026-08-11 (never run — rates UNKNOWN, not assumed good) ─────────────
+  {
+    skill: "analyze-sibling-paper",
+    why: "ONE named rival, deep-read — is our delta real or cosmetic",
+    prompts: [
+      "вышла статья почти про то же самое, надо понять насколько мы пересекаемся",
+      "рецензент пишет «это же уже сделано в X» — разбери X целиком",
+      "read this competitor end to end and tell me the strongest scoop they could claim",
+      "они препринт выложили раньше нас на месяц — это prior work или concurrent",
+    ],
+    collides: ["map-prior-work", "verify-citations"],
+  },
+  {
+    skill: "build-benchmark",
+    why: "design + RUN the study, and ship the artifact reviewers execute",
+    prompts: [
+      "надо спроектировать сам замер, а не описывать его — как считать честно",
+      "нужны парные прогоны, доверительные интервалы и поправка на множественные сравнения",
+      "build the artifact a reviewer can run that recomputes every number in the paper",
+      "у нас n маленькое и разброс большой — это результат или шум",
+    ],
+    collides: ["research-ideate", "draft-paper"],
+  },
+  {
+    skill: "camera-ready",
+    why: "AFTER accept — de-anonymize, archive, disclose",
+    prompts: [
+      "пришёл accept, что теперь делать до дедлайна финальной версии",
+      "надо снять анонимизацию и выложить артефакт уже под своим именем",
+      "we got in — swap the view-only artifact for a real repo and get a DOI",
+      "перед публичным релизом надо закрыть responsible disclosure по найденной дыре",
+    ],
+    collides: ["submit-paper", "extend-paper"],
+  },
+  {
+    skill: "draft-paper",
+    why: "numbers exist → WRITE it: sections, claim-sizing, threats",
+    prompts: [
+      "числа есть, пора писать текст",
+      "turn these findings into an actual submission",
+      "нужен abstract и intro, и явный абзац «чего мы НЕ утверждаем»",
+      "как сформулировать вклад чтобы он был точным, а не раздутым",
+    ],
+    collides: ["argument-arc", "build-benchmark"],
+  },
+  {
+    skill: "extend-paper",
+    why: "accepted workshop paper → a SECOND, stronger publication",
+    prompts: [
+      "воркшопную статью приняли — как сделать из неё полноценную вторую",
+      "сколько нового материала нужно чтобы это не считалось дублем",
+      "we have one accepted short paper and I want a body of work, not a one-hit",
+      "куда подавать расширенную версию и не нарушу ли я правило про dual submission",
+    ],
+    collides: ["camera-ready", "find-venue"],
+  },
+  {
+    skill: "find-venue",
+    why: "WHERE to send it — ranked by indexing and accept-probability",
+    prompts: [
+      "куда это вообще подавать",
+      "нужен список площадок с дедлайнами и понять какие индексируются",
+      "which workshop gives real peer review and lands in the ACM DL",
+      "успеваем ли мы куда-нибудь до конца осени и с какой вероятностью пройдём",
+    ],
+    collides: ["plan-paper-timeline", "study-accepted-papers"],
+  },
+  {
+    skill: "harden-paper",
+    why: "the pre-submit GATE — a vector of gates, worst first",
+    prompts: [
+      "статья готова к сабмиту или нет",
+      "прогони финальную проверку перед загрузкой, всё ли закрыто",
+      "what is still missing before I upload this thing",
+      "хочу вердикт по всем осям сразу, а не одно ревью",
+    ],
+    collides: ["pc-panel-review", "paper-adversarial-review"],
+  },
+  {
+    skill: "plan-paper-timeline",
+    why: "CFP dates → a scheduled, buffered plan on the calendar",
+    prompts: [
+      "дедлайн 15 сентября — разложи по датам что когда делать",
+      "поставь события в календарь: сабмит, дедлайн, камера-реди",
+      "build the schedule backwards from the deadline with buffer for the review passes",
+      "две статьи подряд получаются, разведи их чтобы не слиплись перед подачей",
+    ],
+    collides: ["find-venue"],
+  },
+  {
+    skill: "render-paper",
+    why: "compile the LaTeX and produce readable page images",
+    prompts: [
+      "собери pdf и покажи страницы",
+      "хочу посмотреть как это выглядит вёрсткой с телефона",
+      "compile the tex and screenshot the pages so I can read them",
+      "сборка падает на шрифтах, а мне нужен свежий pdf",
+    ],
+    collides: ["submit-paper"],
+  },
+  {
+    skill: "research-ideate",
+    why: "is this idea WORTH doing — go/no-go before any effort",
+    prompts: [
+      "есть идея — замерить, как часто агенты молча игнорируют правила из CLAUDE.md. стоит ли вообще за неё браться",
+      "это тянет на публикацию или это блог-пост",
+      "is there a fast MVP finding here that does not need a scary result",
+      "думаю переключиться на бенчмарк промпт-инъекций в MCP-серверах — не уведёт ли это нас из нашей линии в сторону",
+    ],
+    collides: ["map-prior-work", "sweep-design-space"],
+  },
+  {
+    skill: "study-accepted-papers",
+    why: "mine the VENUE's accepted corpus for what makes papers strong there",
+    prompts: [
+      "посмотри что принимали на этой площадке и чем те статьи сильны",
+      "хочу дотянуть с accept до strong accept — что для этого нужно именно там",
+      "read a dozen accepted papers from this venue and diff our draft against them",
+      "какие приёмы у них заходят — артефакты, threat model, запоминающееся название",
+    ],
+    collides: ["find-venue", "harden-paper"],
+  },
+  {
+    skill: "submit-paper",
+    why: "the actual upload — artifact hosting, HotCRP, double-blind hygiene",
+    prompts: [
+      "пора загружать, проведи по шагам",
+      "как захостить артефакт анонимно чтобы не спалить авторство",
+      "walk me through the hotcrp form and the double blind checks",
+      "в pdf не осталось ли моего имени в метаданных",
+    ],
+    collides: ["camera-ready", "harden-paper"],
+  },
+  {
+    skill: "sweep-design-space",
+    why: "GENERATE mechanisms — when the last three ideas all rhymed",
+    prompts: [
+      "твои варианты стухли, предложи что-то из другой оперы",
+      "мы третий раз крутим одни и те же два решения",
+      "sweep the design space properly — type systems, capabilities, other professions",
+      "нужен механизм, а не ещё один замер того же самого",
+    ],
+    collides: ["research-ideate"],
+  },
+
+  // ── NEW 2026-08-11 (second batch — the three the coverage sweep found) ───────
+  // These finish the paper lane. Same rule as the batch above: never run, so
+  // their rates are UNKNOWN, not assumed good.
+  {
+    skill: "paper-status",
+    why: "REPORT state — measured page count, git, which gates ran. Reports; does not grade or fix",
+    prompts: [
+      "что там по статье, что осталось",
+      "готова ли статья к сабмиту или ещё нет",
+      "what's the status of the paper right now",
+      "я вернулся после недели, напомни где мы остановились",
+    ],
+    // harden-paper is the sharpest collision in the corpus: both answer a
+    // question shaped like "is it ready?", but one MEASURES and the other JUDGES.
+    // If the selector cannot separate these two, a status question triggers a
+    // 36-minute gate run.
+    collides: ["harden-paper", "paper-pipeline"],
+  },
+  {
+    skill: "osf-artifact-upload",
+    why: "PUSH a file to the OSF project over the API — the mechanics, not the decision to submit",
+    prompts: [
+      "залей новый artifact.zip на osf",
+      "я починил README в артефакте, нужно обновить версию на osf",
+      "replace the artifact file on the OSF project via the api",
+      "рецензентам нужна свежая версия артефакта по той же view-only ссылке",
+    ],
+    collides: ["submit-paper"],
+  },
+  {
+    skill: "paper-pipeline",
+    why: "DRIVE the whole lifecycle — the conductor, when the ask is the whole arc not one step",
+    prompts: [
+      "хочу довести статью от идеи до принятия, веди меня по всему процессу",
+      "с чего вообще начинать новую статью",
+      "run the whole paper pipeline for this idea, not just one stage",
+      "непонятно в каком порядке всё это делать и что от чего зависит",
+    ],
+    collides: ["paper-status", "plan-paper-timeline"],
+  },
+];
+
+// ═════════════════════════════════════════════════════════════════════════════════════════════
+// THE CONSUMER'S OWN CASES — A FIFTH CARRIER, NOT A FORK (2026-09-12)
+//
+// The table above covers THIS PACKAGE'S skills. A consumer installs the package alongside skills
+// of its own, and those collide with each other far harder than the paper skills do: the first
+// consumer's three non-paper cases share the very wording of the request («write a message…»),
+// which is exactly the confusion a trigger measurement exists to expose. Dropping them would not
+// make the measurement smaller, it would make it WRONG — `irrelevantFor()` builds each case's
+// false-positive set out of its colliders, so a missing collider silently turns a precision
+// measurement into a recall measurement.
+//
+// 🔴 AND THEY CANNOT SIMPLY BE CARRIED HERE: the first consumer's cases name private skills and
+// quote private prose. A package that ships one user's private vocabulary has one possible user —
+// the same argument that moved `papersRoot()`, `ledgerPath()`, `citeChecks` and `scriptsRoot()`
+// out of the code and into a declaration. This is the fifth of those carriers and reads the same
+// key from the same file.
+//
+//     // <consumer>/package.json
+//     "paperlint": { "triggerCases": ".claude/lib/skill-trigger-cases.local.mjs" }
+//
+// The module it names must export `CASES` in the shape documented at the top of this file.
+//
+// ⚠️ DECLARED-AND-MISSING THROWS; UNDECLARED IS SILENT. A consumer with no extra skills is the
+// normal case and must not be nagged. A consumer that declared a path and typoed it is a
+// keystroke away from measuring precision against half a corpus and never being told — the same
+// distinction `scriptsRoot()` makes between `undefined` and `null`.
+//
+// ⚠️ A CONSUMER MAY NOT REDEFINE A PACKAGE CASE. Two tables answering "what should fire on this
+// prompt" is two truths; the override would live in the consumer and the original would keep
+// looking authoritative here. Throws, naming the skill.
+import { readFileSync } from "node:fs";
+import { join, resolve } from "node:path";
+import { pathToFileURL } from "node:url";
+import {
+  CONFIG_KEY,
+  consumerRoot,
+  settingsOf,
+} from "../skills/paper-pipeline/scripts/consumer.mjs";
+
+const DEFAULT_LOCAL_CASES = ".claude/lib/skill-trigger-cases.local.mjs";
+
+async function consumerCases() {
+  const root = consumerRoot();
+  let declared;
+  try {
+    declared = settingsOf(
+      JSON.parse(readFileSync(join(root, "package.json"), "utf8")),
+    )?.triggerCases;
+  } catch {
+    declared = undefined;
+  }
+  const rel = declared === undefined ? DEFAULT_LOCAL_CASES : declared;
+  if (typeof rel !== "string" || rel.length === 0)
+    throw new TypeError(
+      `${CONFIG_KEY}: "triggerCases" must be a non-empty string, got ${JSON.stringify(rel)}`,
+    );
+  const abs = resolve(root, rel);
+  let mod;
+  try {
+    mod = await import(pathToFileURL(abs).href);
+  } catch (e) {
+    if (declared === undefined && e?.code === "ERR_MODULE_NOT_FOUND") return [];
+    throw new Error(
+      `${CONFIG_KEY}: could not load the consumer's trigger cases from ${abs}` +
+        (declared === undefined
+          ? ` (the default location; declare another in package.json under ` +
+            `"${CONFIG_KEY}": { "triggerCases": … } if the file lives elsewhere).`
+          : ` — it is declared in package.json as "triggerCases": ${JSON.stringify(declared)}.`) +
+        `
+A missing case set does not shrink the measurement, it CHANGES it: ` +
+        `\`irrelevantFor()\` builds the false-positive set out of a case's colliders, so an ` +
+        `absent collider turns a precision number into a recall number with no sign that it did.` +
+        `
+${String(e?.message ?? e)}`,
+    );
+  }
+  const extra = mod?.CASES;
+  if (!Array.isArray(extra))
+    throw new TypeError(
+      `${CONFIG_KEY}: ${abs} must export an array named CASES`,
+    );
+  const mine = new Set(PACKAGE_CASES.map((c) => c.skill));
+  for (const c of extra)
+    if (mine.has(c.skill))
+      throw new Error(
+        `${CONFIG_KEY}: ${abs} redefines the case for "${c.skill}", which this package already ` +
+          `owns. Two tables answering the same question is two truths; rename the skill or drop ` +
+          `the local case.`,
+      );
+  return extra;
+}
+
+/** This package's cases plus the consumer's own, as one table. */
+export const CASES = [...PACKAGE_CASES, ...(await consumerCases())];
+
+/** Case lookup by skill name — the colocated files' entry point. */
+export const BY_SKILL = new Map(CASES.map((c) => [c.skill, c]));
+
+/**
+ * The irrelevant set for a case: its colliders' should-fire prompts, sliced so no
+ * two cases receive an identical set. Near-duplicate sets pass the per-set
+ * diversity gate while making two cases' false-positive rates the same
+ * measurement twice.
+ */
+export function irrelevantFor(skill, want = 4) {
+  const c = BY_SKILL.get(skill);
+  if (!c) throw new Error(`unknown skill: ${skill}`);
+  // Take enough from EACH collider to reach `want` in total. Fixed 2026-08-11:
+  // this used to take a flat 2 per collider, which is 4 for a two-collider case
+  // and only 2 for a one-collider case — and the diversity gate needs at least
+  // `minPrompts` (4) in BOTH sets. Every one-collider skill therefore threw
+  // instead of measuring: map-prior-work, paper-adversarial-review,
+  // pc-panel-review, plan-paper-timeline, render-paper, sweep-design-space,
+  // verify-citations — seven of twenty-one, exactly the set with `collides.length
+  // === 1`. Two-collider cases are unaffected (ceil(4/2) === 2), so numbers
+  // measured before the fix stay comparable.
+  const per = Math.ceil(want / c.collides.length);
+  return c.collides
+    .flatMap((sib, i) => {
+      const s = BY_SKILL.get(sib);
+      if (!s) throw new Error(`${skill} collides with unknown skill "${sib}"`);
+      // Rotate by the collider's index so two cases sharing a collider do not
+      // receive byte-identical irrelevant sets.
+      return [...s.prompts.slice(i), ...s.prompts.slice(0, i)].slice(0, per);
+    })
+    .slice(0, want);
+}

package/lib/skill-trigger-cases.mutations.mjs

@@ -0,0 +1,65 @@
+/**
+ * skill-trigger-cases.mutations.mjs — the non-vacuity proof for the carrier harness.
+ *
+ * Run: `node lib/skill-trigger-cases.mutations.mjs`
+ *
+ * Every case here removes ONE property the harness claims to hold and requires the harness to
+ * name that property rather than merely go red. Written with the carrier, not after it: a harness
+ * whose success state is a single printed line cannot be observed working, and this package's own
+ * `scripts/run-mutations.mjs` refuses a harness no battery can kill for exactly that reason.
+ */
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+import { runMutations } from "./mutation-driver.mjs";
+import { consumerRoot } from "../skills/paper-pipeline/scripts/consumer.mjs";
+
+const HERE = dirname(fileURLToPath(import.meta.url));
+const SRC = join(HERE, "skill-trigger-cases.mjs");
+const HARNESS = join(HERE, "skill-trigger-cases.harness.mjs");
+
+const MUTATIONS = [
+  [
+    "a typoed declaration falls back to the default instead of throwing",
+    [
+      [
+        SRC,
+        'if (declared === undefined && e?.code === "ERR_MODULE_NOT_FOUND") return [];',
+        'if (e?.code === "ERR_MODULE_NOT_FOUND") return [];',
+      ],
+    ],
+    "a typoed declaration loaded quietly",
+    "the undeclared/declared distinction — a mistyped path becomes a silently halved corpus",
+  ],
+  [
+    "the consumer may shadow a case this package owns",
+    [[SRC, "if (mine.has(c.skill))", "if (false && mine.has(c.skill))"]],
+    "a shadowing case was accepted",
+    "the one-truth rule: two tables answer «what should fire here» and the package's keeps looking authoritative",
+  ],
+  [
+    "the consumer's cases are dropped entirely",
+    [
+      [
+        SRC,
+        "export const CASES = [...PACKAGE_CASES, ...(await consumerCases())];",
+        "export const CASES = [...PACKAGE_CASES];",
+      ],
+    ],
+    "the file at the default location was not merged",
+    "the merge itself — every consumer measures precision against the package's skills alone",
+  ],
+];
+
+process.exit(
+  runMutations({
+    root: consumerRoot(),
+    runner: "vigiles",
+    cases: MUTATIONS.map(([name, edits, expect, disables]) => ({
+      name,
+      disables,
+      edits,
+      harness: HARNESS,
+      expect,
+    })),
+  }),
+);