amsd-pipeline-shared 0.5.38 → 0.5.40

package/orchestrations/prompts/templates/set-fidelity-check.json

@@ -1,6 +1,6 @@
 {
   "id": "set-fidelity-check",
-  "version": 1,
+  "version": 2,
   "seams": [],
   "layer": "engine",
   "description": "A judge reads a codeline's minted roster and reusable prompts against the ticket they were minted for and reports what is geared to that ticket rather than to the codeline.",
@@ -8,11 +8,11 @@
     "__CODELINE__",
     "__TICKET_KEY__",
     "__TICKET_TITLE__",
-    "__ROSTER__",
-    "__PROMPTS__"
+    "__ROSTER__"
   ],
   "$why": [
-    "v1 (operator, 2026-10-24: \"We need some post pause 1 checkpoints to ensure no contamination in codeline level prompts and profiles\"): run 13's roster named every agent after the ticket's subject and ~half its reusable prompts described that one ticket; the sets are cached per codeline and reused for every later story. Judged by an agent, not a word list; run by the pre-launch of the leg after pause 1 and reported at pause 1."
+    "v1 (operator, 2026-10-24: \"We need some post pause 1 checkpoints to ensure no contamination in codeline level prompts and profiles\"): run 13's roster named every agent after the ticket's subject and ~half its reusable prompts described that one ticket; the sets are cached per codeline and reused for every later story. Judged by an agent, not a word list; run by the pre-launch of the leg after pause 1 and reported at pause 1.",
+    "v2 (live, 2026-10-24): a codeline's whole prompt set in one call was 908 KB and the judge answered about its own context limit; the prompts are judged one by one by the pipeline's own reviewer as they are built (each copy records its judges), so this judge reads the roster only."
   ],
-  "body": "You judge what a pipeline minted for ONE codeline and will REUSE for every later story on it: its agent roster (names and briefs) and\nits reusable agent prompts. They were minted while this ticket was in scope:\n\nTicket __TICKET_KEY__: __TICKET_TITLE__\nCodeline: __CODELINE__\n\nTHE ROSTER (each agent: name, kind, brief):\n__ROSTER__\n\nTHE PROMPTS (each: id, then its body):\n__PROMPTS__\n\nReport every place where the roster or a prompt is geared to THIS ticket rather than to the codeline: an agent named after the\nticket's subject, component or key; a brief or prompt that takes this ticket's subject, components, files or requirements as its\nsubject or as standing instruction, so that an implementer on a DIFFERENT story of this codeline would be misled. A mention of the\nticket's subject or files only as a marked example of the codeline's conventions is not geared. Quote the words. Report nothing you\ncannot quote. Answer in the structured form you are given; `geared` is empty when nothing is geared."
+  "body": "You judge the agent roster a pipeline minted for ONE codeline and will REUSE for every later story on it: its agents' names and\nbriefs. It was minted while this ticket was in scope:\n\nTicket __TICKET_KEY__: __TICKET_TITLE__\nCodeline: __CODELINE__\n\nTHE ROSTER (each agent: name, kind, brief):\n__ROSTER__\n\nReport every agent that is geared to THIS ticket rather than to the codeline: named after the ticket's subject, component or key, or\nbriefed so that it takes this ticket's subject, components, files or requirements as its subject or as standing instruction — an\nimplementer on a DIFFERENT story of this codeline would be misled. A mention of the ticket's subject or files only as a marked example\nof the codeline's conventions is not geared. Quote the words. Report nothing you cannot quote. Answer in the structured form you are\ngiven; `geared` is empty when nothing is geared."
 }

package/orchestrations/scripts/lib/prelaunch.js

@@ -330,9 +330,13 @@ function checklist({ env, results }) {
 /**
  * NOTHING MINTED FOR A CODELINE IS GEARED TO THE TICKET (operator, 2026-10-24: "post pause 1 checkpoints to ensure no contamination in
  * codeline level prompts and profiles"). After pause 1 the Agents phase has minted each codeline's roster and prompt set, which are kept
- * and reused for every later story on that codeline. A judge agent (template set-fidelity-check, the given `judge` call) reads each
- * codeline's roster and prompts against the ticket and quotes what is geared to it; anything quoted is RED, naming the codeline, the
- * file and the words. Not run when the launch gives no judge (a £0 leg) — then NOT-YET, never GREEN by silence.
+ * and reused for every later story on that codeline. Two parts, each what it needs to be:
+ *   - the ROSTER: a judge agent (template set-fidelity-check, the given `judge` call) reads each codeline's agents against the ticket and
+ *     quotes what is geared to it;
+ *   - the PROMPTS: every recorded copy must carry the record of the reviewer that judged it (judgedWith — the pipeline's own prompt review,
+ *     whose rule is a prompt narrowed to one story); a copy with no record is the exact gap that let run 13's ticket-geared prompts through,
+ *     and is RED by name. (A whole set in one judge call was 908 KB and the judge answered about its context, not the prompts.)
+ * Anything quoted or unjudged is RED, naming the codeline and the words or files. No judge given (a £0 leg): RED, never green by silence.
  */
 const SET_FIDELITY_SCHEMA = { type: 'object', required: ['geared'], properties: { geared: { type: 'array', items: { type: 'object',
   required: ['where', 'quote', 'why'], properties: { where: { type: 'string' }, quote: { type: 'string' }, why: { type: 'string' } } } } } };
@@ -343,26 +347,33 @@ async function setFidelity({ install, projectConfigDir, ticket, codelines = [],
   const rows = []; let examined = 0;
   for (const codeline of codelines) {
     const rp = b.recordedPromptsDir(projectConfigDir, codeline); const mi = b.mintedSnapshotDir(projectConfigDir, codeline);
-    const prompts = fs.existsSync(rp) ? fs.readdirSync(rp).filter((n) => n.endsWith('.json')).sort().map((n) => { const d = readJson(path.join(rp, n), {}); return `=== ${n} ===\n${d.body || Object.values(d.bodies || {}).join('\n')}`; }) : [];
+    const promptFiles = fs.existsSync(rp) ? fs.readdirSync(rp).filter((n) => n.endsWith('.json')).sort() : [];
+    // a generated copy (authority project) with no judging record; a verbatim copy (authority template / copied) carries none by design
+    const unjudged = promptFiles.filter((n) => { const d = readJson(path.join(rp, n), {}); return d && d.authority === 'project' && !d.judgedWith; });
     let roster = '(none recorded)';
     try {
       const ag = readJson(path.join(mi, 'roster.json'), {}).agents || {}; const pr = (readJson(path.join(mi, 'agent-profiles.json'), {}).profiles) || {};
       const names = Object.keys(ag);
       if (names.length) roster = names.map((n) => `- ${n} [${(ag[n] || {}).kind || ''}]\n${String(pr[n] || (ag[n] || {}).persona || '(no brief)')}`).join('\n\n');
     } catch { /* judged as recorded */ }
-    if (!prompts.length && roster === '(none recorded)') { rows.push({ codeline, examined: false, geared: [] }); continue; }
-    const prompt = renderEngineTemplate('set-fidelity-check', { __CODELINE__: codeline, __TICKET_KEY__: ticket.key || '', __TICKET_TITLE__: ticket.title || '',
-      __ROSTER__: roster, __PROMPTS__: prompts.join('\n\n') || '(none recorded)' });
-    const out = await judge({ prompt, schema: SET_FIDELITY_SCHEMA, codeline });
+    if (!promptFiles.length && roster === '(none recorded)') { rows.push({ codeline, examined: false, geared: [], unjudged: [] }); continue; }
+    let geared = [];
+    if (roster !== '(none recorded)') {
+      const prompt = renderEngineTemplate('set-fidelity-check', { __CODELINE__: codeline, __TICKET_KEY__: ticket.key || '', __TICKET_TITLE__: ticket.title || '', __ROSTER__: roster });
+      const out = await judge({ prompt, schema: SET_FIDELITY_SCHEMA, codeline });
+      geared = (out && Array.isArray(out.geared)) ? out.geared : null;
+    }
     examined += 1;
-    rows.push({ codeline, examined: true, geared: (out && Array.isArray(out.geared)) ? out.geared : null });
+    rows.push({ codeline, examined: true, geared, unjudged });
   }
   const unjudged = rows.filter((r) => r.examined && r.geared === null).map((r) => r.codeline);
   const geared = rows.filter((r) => r.examined && Array.isArray(r.geared) && r.geared.length);
-  const ok = examined > 0 && !unjudged.length && !geared.length;
+  const unjudgedPrompts = rows.filter((r) => r.examined && r.unjudged.length);
+  const ok = examined > 0 && !unjudged.length && !geared.length && !unjudgedPrompts.length;
   const why = [geared.length && geared.map((r) => `${r.codeline}: ${r.geared.length} geared — ${r.geared.slice(0, 3).map((g) => `${g.where}: "${String(g.quote).slice(0, 80)}"`).join('; ')}`).join('; '),
+    unjudgedPrompts.length && unjudgedPrompts.map((r) => `${r.codeline}: ${r.unjudged.length} prompt copy(ies) no reviewer is recorded to have judged: ${r.unjudged.slice(0, 4).join(', ')}`).join('; '),
     unjudged.length && `no verdict for ${unjudged.join(', ')}`, !examined && 'nothing minted to examine'].filter(Boolean).join('; ');
-  return res('set-fidelity', ok, ok ? `${examined} codeline set(s) judged against ${ticket.key}: nothing geared to the ticket` : why, { rows });
+  return res('set-fidelity', ok, ok ? `${examined} codeline set(s) judged against ${ticket.key}: no agent geared to the ticket, every prompt copy judged by the reviewer` : why, { rows });
 }
 
 /** The set-fidelity judge a LIVE launch gives setFidelity: the prompt-review seam's own model, ladder and tools (seam-invocation), through the
@@ -375,7 +386,8 @@ function setFidelityJudge({ install, env = process.env, logDir }) {
   return async ({ prompt, schema, codeline }) => {
     const seamEnv = seamInvocationEnv('prompt-review', path.join(orchOf(install), 'agents'));
     if (seamEnv.EPAM_ALLOWED_TOOLS) seamEnv.AI_GATE_ALLOW_TOOLS = '1';
-    seamEnv.EPAM_AGENT_NAME = 'set-fidelity';
+    // the call runs AS the prompt-review seam (llm-handler resolves the seam by agent name); the ledger names the check
+    seamEnv.EPAM_AGENT_NAME = 'prompt-review';
     seamEnv.EPAM_RESPONSE_SCHEMA = spec.schemaEnv({ name: 'set_fidelity', description: 'what is geared to the ticket', parameters: schema });
     const text = await spec.runClaude(promptExec, prompt, path.join(logDir || os.tmpdir(), `set-fidelity-${codeline}.log`), seamEnv, { costAgent: 'set-fidelity' });
     const m = String(text || '').match(/\{[\s\S]*\}/);

package/orchestrations/scripts/spec-mode-runner.js

@@ -4470,13 +4470,29 @@ const TOOL_ROSTER_REVIEW = {
     'checked with your tools. Do not answer in prose.',
   parameters: {
     type: 'object',
-    required: ['verdict', 'findings'],
+    required: ['verdict', 'findings', 'roles'],
     properties: {
       verdict: {
         type: 'string',
         enum: ['sound', 'defects_found', 'nothing_to_review'],
         description: 'sound = every checkable claim held. defects_found = at least one did not.',
       },
+      // A ROLE OUTLIVES THE TICKET (operator, 2026-10-24): the roster is kept with its codelines and reused for every later story, so
+      // every minted role is answered for, one by one — the rule in prose alone was answered "sound" on a roster whose every agent was
+      // named after the ticket. A role that does not outlive the ticket becomes a blocking finding (rolesOutliveTicketFindings).
+      roles: {
+        type: 'array',
+        description: 'One entry for EVERY minted role reviewed.',
+        items: {
+          type: 'object',
+          required: ['name', 'outlivesTicket', 'why'],
+          properties: {
+            name: { type: 'string', description: 'The minted role, named exactly as in the roster.' },
+            outlivesTicket: { type: 'boolean', description: 'true only when this role — its name and its brief — would serve a story on this codeline about something OTHER than the current ticket: named for the kind of engineer/investigator and the codeline, briefed on the codeline, not on this ticket\'s subject, components or files (except as marked examples).' },
+            why: { type: 'string', description: 'Quote the words that gear the role to the ticket, or say what makes it the codeline\'s.' },
+          },
+        },
+      },
       findings: {
         type: 'array',
         items: {
@@ -4885,6 +4901,19 @@ function _readJsonOrNull(f) {
  * Extracted and exported because it could not be tested otherwise: it sat inside the async function
  * that also does the batching and makes the model calls, so exercising it meant paying for a run.
  */
+/** A ROLE THAT DOES NOT OUTLIVE THE TICKET IS A BLOCKING FINDING (operator, 2026-10-24): the review answers for every minted role
+ *  (TOOL_ROSTER_REVIEW.roles); one it marks outlivesTicket:false — named or briefed for the ticket in scope rather than the codeline —
+ *  is turned into the finding the mint's correction loop acts on. The reviewer's own words are the claim. */
+function rolesOutliveTicketFindings(part) {
+  const roles = part && Array.isArray(part.roles) ? part.roles : [];
+  return roles.filter((r) => r && r.name && r.outlivesTicket === false).map((r) => ({
+    agent: String(r.name), severity: 'blocking', claim: String(r.why || '(the reviewer gave no words)'),
+    checked: 'whether this role, as named and briefed, would serve a story on its codeline about something other than the current ticket',
+    found: 'minted for the ticket in scope, not for the codeline it is kept with',
+    remedy: 'a role named for its kind of work and its codeline, briefed on the codeline — its stack, conventions and how to verify there — naming the current ticket only as a marked example',
+  }));
+}
+
 function aggregateRosterReview(results, legalVerdicts) {
   const legal = Array.isArray(legalVerdicts) ? legalVerdicts : [];
   const rows = Array.isArray(results) ? results : [];
@@ -4892,7 +4921,9 @@ function aggregateRosterReview(results, legalVerdicts) {
   const unexamined = rows.filter((r) => !r || !r.part || typeof r.part !== 'object'
     || r.part.verdict === 'nothing_to_review' || !r.part.verdict || illegal(r.part.verdict));
   const examined = rows.filter((r) => !unexamined.includes(r));
-  const findings = rows.flatMap((r) => (r && r.part && Array.isArray(r.part.findings) ? r.part.findings : []));
+  // a role an examined batch says does not outlive the ticket is a finding of that batch, whatever verdict it gave
+  const roleFindings = examined.flatMap((r) => rolesOutliveTicketFindings(r.part));
+  const findings = rows.flatMap((r) => (r && r.part && Array.isArray(r.part.findings) ? r.part.findings : [])).concat(roleFindings);
 
   if (!examined.length) {
     return {
@@ -4903,7 +4934,7 @@ function aggregateRosterReview(results, legalVerdicts) {
     };
   }
 
-  const verdict = examined.some((r) => r.part.verdict === 'defects_found') ? 'defects_found' : 'sound';
+  const verdict = (examined.some((r) => r.part.verdict === 'defects_found') || roleFindings.length) ? 'defects_found' : 'sound';
   return {
     verdict,
     findings,
@@ -11107,7 +11138,7 @@ module.exports = {
   storyAcsBlock,
   specAgentContract,
   rosterReviewVerdict,
-  aggregateRosterReview,
+  aggregateRosterReview, rolesOutliveTicketFindings,
   reviewOutcomeKeepsChange,
   // Exported so the brief lookup can be asserted without a run.
   profilesWithProjectBriefs,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "amsd-pipeline-shared",
-  "version": "0.5.38",
+  "version": "0.5.40",
   "description": "The AMSD pipeline's shared components, one source for the bash pipeline and its Claude workflows: the model-call gateway (llm-handler.sh — retry, ladder, self-heal, cost) and the libraries it runs.",
   "license": "UNLICENSED",
   "bin": {