dsh-vibe-math 2.5.0 → 2.7.0

package/tests/audit-persona-sensitivity.mjs

@@ -83,16 +83,16 @@ const probes = [
   {
     name: 'v2-lean-tools-line-removed',
     preset: 'vibe-math-v2', file: 'agent.cordis.yml', nth: 'all', expect: 2,
-    guarantee: 'the three Lean tools must be named in the persona the main agent receives',
-    from: '      - vibe_math_lean_run / vibe_math_lean_archive / vibe_math_lean_lib — Lean formal\n        verification (execute / archive / list the reuse library). The scheduler\'s child agents\n        use them too; they work in every mode.\n',
+    guarantee: 'the five Lean tools must be named in the persona the main agent receives',
+    from: '      - vibe_math_lean_run / vibe_math_lean_archive / vibe_math_lean_lib / vibe_math_lean_read /\n        vibe_math_lean_job — Lean formal verification (execute / archive / list + read the reuse\n        library / inspect or wait for a background compile job). leanAsync=true (default) queues the\n        compile and returns a jobId; nothing counts as proved until that job settles ok (so "有把握"\n        is not the same as "已验证"). The scheduler\'s child agents use them too; they work in every\n        mode.\n',
     to: '',
   },
   {
     name: 'v2-phantom-tool-mentioned',
     preset: 'vibe-math-v2', file: 'agent.cordis.yml', nth: 'all', expect: 2,
     guarantee: 'the persona must not advertise a tool that is not registered (the agent would call it and fail)',
-    from: '      - vibe_math_lean_run / vibe_math_lean_archive / vibe_math_lean_lib — Lean formal',
-    to: '      - vibe_math_lean_exec / vibe_math_lean_run / vibe_math_lean_archive / vibe_math_lean_lib — Lean formal',
+    from: '      - vibe_math_lean_run / vibe_math_lean_archive / vibe_math_lean_lib / vibe_math_lean_read /',
+    to: '      - vibe_math_lean_exec / vibe_math_lean_run / vibe_math_lean_archive / vibe_math_lean_lib / vibe_math_lean_read /',
   },
 
   // ── v3 ─────────────────────────────────────────────────────────────────────
@@ -148,27 +148,80 @@ const probes = [
     name: 'v4-usage-advertises-phantom-subcommand',
     preset: 'vibe-math-v4', file: 'vibe-math-v4.js', nth: 'first', expect: 1,
     guarantee: 'the unknown-subcommand usage string must not advertise a subcommand that no branch implements',
-    // The anchor must be the CURRENT literal: Round A changed both the usage string and the
+    // The anchor must be the CURRENT literal. Round A changed both the usage string and the
     // hint to advertise `start [problem]` / `set <k=v>...` (vibe-math-v4.js:2492). While the
     // anchor kept the pre-fix literal it matched 0 times, so this probe reported SETUP-FAIL
     // every run and the F-section agreement between usage and branches had NO sensitivity
     // proof at all (the hint probe at :155 was updated, this one was not).
-    from: "usage:'configure|start [problem]|resume|pause|abort|status|report|message <to|all> <content>|meeting|members|add|remove|set <k=v>...'",
-    to: "usage:'configure|start [problem]|resume|pause|abort|status|report|message <to|all> <content>|meeting|members|add|remove|scan|set <k=v>...'",
+    // Round H (final paper) inserted `paper [...]` into both the usage and the hint; the anchor
+    // follows the new literal again, so the probe keeps proving the usage<->branches agreement.
+    from: "usage:'configure|start [problem]|resume|pause|abort|status|report|message <to|all> <content>|meeting|paper [lang=zh|en] [format=md|tex|both] [editor=office|resident:<id>] [force]|members|add|remove|set <k=v>...'",
+    to: "usage:'configure|start [problem]|resume|pause|abort|status|report|message <to|all> <content>|meeting|paper [lang=zh|en] [format=md|tex|both] [editor=office|resident:<id>] [force]|members|add|remove|scan|set <k=v>...'",
   },
   {
     name: 'v4-hint-advertises-unimplemented-subcommand',
     preset: 'vibe-math-v4', file: 'vibe-math-v4.js', nth: 'first', expect: 1,
     guarantee: 'the typing hint must not advertise a subcommand the handler does not implement',
-    from: "message <to|all> <content>|meeting|members|add|remove|set]'",
-    to: "message <to|all> <content>|meeting|add|remove|set]'",
+    // Same reason as above: the hint now carries the `paper [...]` entry between `meeting` and
+    // `members`, so the anchor names the current literal and drops an IMPLEMENTED subcommand
+    // (`members`) from the mutated hint — the suite's implemented-subset-of-hint check must fail.
+    from: "message <to|all> <content>|meeting|paper [lang=zh|en] [format=md|tex|both] [editor=office|resident:<id>] [force]|members|add|remove|set]'",
+    to: "message <to|all> <content>|meeting|paper [lang=zh|en] [format=md|tex|both] [editor=office|resident:<id>] [force]|add|remove|set]'",
   },
   {
     name: 'v5-persona-slash-list-drops-add-remove',
     preset: 'vibe-math-v5', file: 'agent.cordis.yml', nth: 'all', expect: 2,
     guarantee: "the persona's /v5 list must match the command hint (a human reading the persona must see every subcommand)",
-    from: 'meeting|hire|fire|add|remove|set)',
-    to: 'meeting|hire|fire|set)',
+    // Round H appended `|paper` to the persona's /v5 list (and to the command hint), so the
+    // anchor carries the new tail; the mutation still drops the two subcommands whose absence
+    // the Round-G persona actually had (`add|remove`), which the hint<->persona equality check
+    // must catch.
+    from: 'meeting|hire|fire|add|remove|set|paper)',
+    to: 'meeting|hire|fire|set|paper)',
+  },
+  // ── final-paper surface (Round H) ──────────────────────────────────────────
+  // These two pin the surface the final-paper feature ADDED, so a later edit cannot quietly
+  // drop it: the v4 hint must keep advertising the subcommand its handler implements, and the
+  // v5 persona must keep naming the tools the plugin actually registers.
+  {
+    name: 'v4-hint-stops-advertising-paper',
+    preset: 'vibe-math-v4', file: 'vibe-math-v4.js', nth: 'first', expect: 1,
+    guarantee: 'the /v4 hint must keep advertising the final-paper subcommand the handler implements',
+    from: "|meeting|paper [lang=zh|en] [format=md|tex|both] [editor=office|resident:<id>] [force]|members|add|remove|set]'",
+    to: "|meeting|members|add|remove|set]'",
+  },
+  {
+    name: 'v5-paper-tool-renamed-in-plugin',
+    preset: 'vibe-math-v5', file: 'vibe-math-v5.js', nth: 'first', expect: 1,
+    guarantee: 'renaming the final-paper tool must be caught (the persona still advertises vibe_v5_paper)',
+    from: "  registerTool('vibe_v5_paper',",
+    to: "  registerTool('vibe_v5_paperX',",
+  },
+  // ── the persona↔hint list itself (the gap that hid `paper` in v2/v3/v4) ────────────────────
+  // The surface suite used to SKIP this comparison whenever the persona list contained `...`
+  // (v3's trailing ellipsis) or an argument placeholder with `...` (v4's `set <k=v>...`), and v2
+  // never enumerated at all. Each probe below drops ONE implemented subcommand from the persona
+  // list and requires the suite to go RED, so a future omission cannot hide behind an ellipsis.
+  {
+    name: 'v2-persona-slash-list-drops-paper',
+    preset: 'vibe-math-v2', file: 'agent.cordis.yml', nth: 'all', expect: 2,
+    guarantee: 'the v2 persona must name EVERY implemented /vibe subcommand explicitly (it never enumerated before)',
+    from: '|decisions|agents|paper [lang=zh|en] [format=both|md|tex] [force]).',
+    to: '|decisions|agents).',
+  },
+  {
+    name: 'v3-persona-slash-list-drops-paper',
+    preset: 'vibe-math-v3', file: 'agent.cordis.yml', nth: 'all', expect: 2,
+    guarantee: "the v3 persona's /vibe list must name every implemented subcommand (the old trailing '...' skipped the check)",
+    from: '|decisions|agents|paper [lang=zh|en] [format=both|md|tex] [force]).',
+    to: '|decisions|agents).',
+  },
+  {
+    name: 'v4-persona-slash-list-drops-paper',
+    preset: 'vibe-math-v4', file: 'agent.cordis.yml', nth: 'all', expect: 2,
+    guarantee: "a dropped v4 persona entry must be caught even though the list contains `set <k=v>...` (the old ellipsis escape)",
+    from: '|meeting|paper [lang=zh|en] [format=md|tex|both] [editor=office|resident:<id>] [force]|members',
+    to: '|meeting|members',
   },
 ]
 

package/tests/audit-persona-surface.test.mjs

@@ -132,29 +132,35 @@ const PRESETS = [
     dir: 'vibe-math-v2',
     js: 'vibe-math-v2.js',
     prefix: 'vibe_math_',
-    tools: 25,
+    tools: 28,
     // member-facing write-lock / scheduler-metadata tools; the v2 coordinator never
     // writes Markdown itself, so they stay out of its persona.
     undocumented: [],
-    lean: { tools: ['vibe_math_lean_run', 'vibe_math_lean_archive', 'vibe_math_lean_lib'], extra: [] },
+    lean: { tools: ['vibe_math_lean_run', 'vibe_math_lean_archive', 'vibe_math_lean_lib', 'vibe_math_lean_read', 'vibe_math_lean_job'], extra: [] },
+    // the shared math-computation tool (docs/math-computation.md) is prefix-less by
+    // design (INTERFACE-FREEZE.md §3), so it is pinned explicitly here.
+    math: { tools: ['math_computation'] },
   },
   {
     dir: 'vibe-math-v3',
     js: 'vibe-math-v3.js',
     prefix: 'vibe_math_',
-    tools: 33,
+    tools: 36,
     // `vibe_math_sync_meta` is no longer undocumented: the promotion-contract line of the
     // persona now names it as the place to report `lemmas[].价值/关键性` (v3 audit M1 —
     // without a prompt-side source for that field the promotion main line is unreachable).
     // The write-lock pair stays out: solver/method-keeper prompts carry it via kcWriteRules().
     undocumented: ['vibe_math_claim_write', 'vibe_math_release_write'],
-    lean: { tools: ['vibe_math_lean_run', 'vibe_math_lean_archive', 'vibe_math_lean_lib'], extra: [] },
+    lean: { tools: ['vibe_math_lean_run', 'vibe_math_lean_archive', 'vibe_math_lean_lib', 'vibe_math_lean_read', 'vibe_math_lean_job'], extra: [] },
+    // the shared math-computation tool (docs/math-computation.md) is prefix-less by
+    // design (INTERFACE-FREEZE.md §3), so it is pinned explicitly here.
+    math: { tools: ['math_computation'] },
   },
   {
     dir: 'vibe-math-v4',
     js: 'vibe-math-v4.js',
     prefix: 'vibe_v4_',
-    tools: 32,
+    tools: 35,
     // resident-facing tools (mail, library cards, task board, write lock). The
     // coordinator drives residents through vibe_v4_message / _meeting / _add_member.
     undocumented: [
@@ -164,15 +170,20 @@ const PRESETS = [
       'vibe_v4_task_done', 'vibe_v4_list_tasks', 'vibe_v4_report_context',
       'vibe_v4_claim_write', 'vibe_v4_release_write',
     ],
-    lean: { tools: ['vibe_v4_lean_run', 'vibe_v4_lean_archive', 'vibe_v4_lean_lib'], extra: ['vibe_v4_formal_report'] },
+    lean: { tools: ['vibe_v4_lean_run', 'vibe_v4_lean_archive', 'vibe_v4_lean_lib', 'vibe_v4_lean_read', 'vibe_v4_lean_job'], extra: ['vibe_v4_formal_report'] },
+    // the shared math-computation tool (docs/math-computation.md) is prefix-less by
+    // design (INTERFACE-FREEZE.md §3), so it is pinned explicitly here.
+    math: { tools: ['math_computation'] },
   },
   {
     dir: 'vibe-math-v5',
     js: 'vibe-math-v5.js',
     prefix: 'vibe_v5_',
-    tools: 35,
+    tools: 40,
     // member- and academician-facing tools; the office (main agent) holds only the
-    // institute-level controls plus the hiring authority.
+    // institute-level controls plus the hiring authority. The final-paper pair is an
+    // OFFICE control and IS named in the persona (vibe_v5_paper / vibe_v5_finalize_paper),
+    // so it must not appear in this snapshot.
     undocumented: [
       'vibe_v5_wait', 'vibe_v5_record_progress', 'vibe_v5_record_proposition',
       'vibe_v5_record_method', 'vibe_v5_record_subproblem', 'vibe_v5_read_library',
@@ -180,11 +191,14 @@ const PRESETS = [
       'vibe_v5_task_list', 'vibe_v5_task_get', 'vibe_v5_task_update',
       'vibe_v5_overview', 'vibe_v5_assign', 'vibe_v5_prioritize', 'vibe_v5_nudge',
     ],
-    lean: { tools: ['vibe_v5_lean_run', 'vibe_v5_lean_archive', 'vibe_v5_lean_lib'], extra: [] },
+    lean: { tools: ['vibe_v5_lean_run', 'vibe_v5_lean_archive', 'vibe_v5_lean_lib', 'vibe_v5_lean_read', 'vibe_v5_lean_job'], extra: [] },
+    // the shared math-computation tool (docs/math-computation.md) is prefix-less by
+    // design (INTERFACE-FREEZE.md §3), so it is pinned explicitly here.
+    math: { tools: ['math_computation'] },
   },
 ]
 
-const LEAN_PARAMS = ['formalVerify', 'leanCommand', 'leanArgs', 'leanTimeoutMs']
+const LEAN_PARAMS = ['formalVerify', 'leanCommand', 'leanArgs', 'leanTimeoutMs', 'leanAsync', 'leanJobsMaxParallel', 'leanInitiative', 'leanSearchPaths']
 
 for (const P of PRESETS) {
   const ymlPath = join(BASE, P.dir, 'agent.cordis.yml')
@@ -193,9 +207,19 @@ for (const P of PRESETS) {
   const src = readFileSync(jsPath, 'utf8')
 
   // ---- registered tools -------------------------------------------------
+  // Two registration shapes exist now:
+  //   1. a literal `registerTool('<name>', …)` (most tools, including the shared math tool in
+  //      v2/v3, which register it by its frozen prefix-less name);
+  //   2. the shared math tool in v4/v5, where the plugin only wires a host callback and the module
+  //      calls `host.register(MATH_TOOL_NAME, …)`. The module's name is FROZEN
+  //      (MATH_TOOL_NAME = 'math_computation', INTERFACE-FREEZE.md §3), so a source that wires
+  //      `registerMathComputation(` contributes exactly that one tool. Without this the shared tool
+  //      would be invisible to this guard on v4/v5 - a silent coverage hole.
   const registered = new Set()
   for (const m of src.matchAll(/registerTool\(\s*'([A-Za-z0-9_]+)'/g)) registered.add(m[1])
+  if (/registerMathComputation\s*\(/.test(src)) registered.add('math_computation')
   eq(registered.size, P.tools, `${P.dir}: registered tool count changed (update this snapshot deliberately)`)
+  for (const tool of (P.math ? P.math.tools : [])) ok(registered.has(tool), `${P.dir}: the shared math tool ${tool} is not registered`)
 
   // ---- A. both literal blocks exist and are non-empty -------------------
   const blocks = literalBlocks(yml)
@@ -224,7 +248,13 @@ for (const P of PRESETS) {
     // Uppercase is part of the token class on purpose: a drifted name such as
     // `vibe_v4_formal_reportX` must be captured whole and judged unregistered, instead of
     // being read as a mention of the registered `vibe_v4_formal_report`.
-    const re = new RegExp('\\b' + P.prefix.replace(/_$/, '') + '_[A-Za-z0-9_]+', 'g')
+    // Names that do NOT carry this preset's prefix (the shared `math_computation`) are added to the
+    // alternation from the registry, so "every registered tool must be mentioned" stays true for
+    // them too instead of forcing them into the `undocumented` snapshot.
+    const prefixRoot = P.prefix.replace(/_$/, '')
+    const prefixless = [...registered].filter((n) => !new RegExp('^' + prefixRoot + '_').test(n))
+    const escapeRe = (s) => s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')
+    const re = new RegExp('\\b' + prefixRoot + '_[A-Za-z0-9_]+' + (prefixless.length ? '|\\b' + prefixless.map(escapeRe).join('\\b|\\b') + '\\b' : ''), 'g')
     for (const m of body.matchAll(re)) {
       if (body[m.index + m[0].length] === '*') { wildcards.add(m[0]); continue }
       mentioned.add(m[0])
@@ -258,12 +288,19 @@ for (const P of PRESETS) {
   if (!hintOpen) eq(implemented.filter((c) => !hinted.includes(c)).sort(), [], `${P.dir}: implemented slash subcommand(s) are missing from the hint`)
   eq(used.sort(), implemented, `${P.dir}: the usage string and the implemented slash subcommands disagree`)
   const personaEnum = /slash command mirrors[^(]*\(([\s\S]*?)\)/.exec(blocks.prefix.join('\n') + '\n' + blocks.text.join('\n'))
-  if (personaEnum && personaEnum[1].includes('|')) {
+  // An ellipsis must NOT switch this check off (that is how a dropped `paper` entry survived here),
+  // and `...` inside an argument placeholder (`set <k=v>...`) is not an abbreviation at all. So the
+  // persona is compared against the BRANCHES (the ground truth), and every implemented subcommand
+  // must be named EXPLICITLY. Only a HINT that is itself abbreviated makes the reverse direction
+  // (persona ⊆ hint) unprovable, so only that direction is relaxed then.
+  if (!personaEnum) {
+    ok(false, `${P.dir}: the persona never enumerates its /vN subcommands (list them so a human can see every one)`)
+  } else {
     const personaCmds = cmdsFrom(personaEnum[1].replace(/^\s*\/[a-z0-9]+\s+/, ''))
-    if (personaCmds.length) {
-      eq(personaCmds.filter((c) => !hinted.includes(c)).sort(), [], `${P.dir}: the persona enumerates slash subcommand(s) the hint does not know`)
-      if (!/\.\.\./.test(personaEnum[1]) && !hintOpen) eq(personaCmds.slice().sort(), hinted.slice().sort(), `${P.dir}: the persona's /vN list and the command hint disagree`)
-    }
+    ok(personaCmds.length > 0, `${P.dir}: no slash subcommand could be parsed from the persona's /vN list`)
+    eq(personaCmds.filter((c) => !implemented.includes(c)).sort(), [], `${P.dir}: the persona enumerates slash subcommand(s) the handler does not implement`)
+    eq(implemented.filter((c) => !personaCmds.includes(c)).sort(), [], `${P.dir}: the persona's /vN list omits subcommand(s) the handler implements — list them explicitly, do not hide behind "..."`)
+    if (!hintOpen) eq(personaCmds.slice().sort(), implemented.slice().sort(), `${P.dir}: the persona's /vN list and the command hint disagree`)
   }
 
   // ---- E. the Lean feature's prompt surface ----------------------------

package/tests/audit-v5-integrity.mjs

@@ -90,6 +90,20 @@ for (const m of src.matchAll(/(?:const|let|var)\s+\{([^}]+)\}\s*=/g)) {
   for (const part of m[1].split(',')) { const n = part.split(':').pop().trim(); if (n) defined.add(n) }
 }
 for (const m of src.matchAll(/catch\s*\(\s*([A-Za-z_$][\w$]*)\s*\)/g)) defined.add(m[1])
+// ESM imports are DEFINITIONS too: the preset imports the shared `math_computation` module
+// (INTERFACE-FREEZE §5.1), so calling an imported function is not a ReferenceError. Without
+// this every imported symbol was reported as "called but never defined"; a call to a name that
+// is neither declared nor imported is still flagged.
+for (const m of src.matchAll(/import\s*(?:\*\s*as\s+([A-Za-z_$][\w$]*)|([A-Za-z_$][\w$]*)\s*,?\s*)?(?:\{([^}]*)\}\s*)?from\s*['"][^'"]*['"]/g)) {
+  if (m[1]) defined.add(m[1])
+  if (m[2]) defined.add(m[2])
+  if (m[3]) {
+    for (const part of m[3].split(',')) {
+      const name = part.split(/\s+as\s+/).pop().trim()
+      if (name) defined.add(name)
+    }
+  }
+}
 // Function/method PARAMETERS are locally bound identifiers too, not free calls.
 for (const m of src.matchAll(/function\s*[A-Za-z_$]*\s*\(([^)]*)\)/g)) {
   for (const raw of m[1].split(',')) {