Fix malformed Face Mind retry routing

apps/worker/src/index.ts

@@ -544,6 +544,7 @@ async function executeRepoFaceJobWithInterpreter(
   const firstInterpretation = firstForbiddenTools.length > 0
     ? {
         decision: "retry" as const,
+        parseable: true,
         reasons: [
           `Face used forbidden substrate/tool-discovery tool(s): ${firstForbiddenTools.join(", ")}`,
           "Face turns may use retrieval tools only; state and substrate inventory belong to the parent/interpreter.",
@@ -563,6 +564,48 @@ async function executeRepoFaceJobWithInterpreter(
     return dropRepoFaceActionBlocks(firstResponse, firstInterpretation);
   }
 
+  if (!firstInterpretation.parseable) {
+    await auditLog.record({
+      type: "repo_face.parent_interpreter_retry",
+      actorId: job.requester.id,
+      jobId: job.id,
+      provider: job.provider,
+      details: {
+        decision: firstInterpretation.decision,
+        reasons: firstInterpretation.reasons,
+        owner: "parent_interpreter",
+      },
+    });
+    console.warn(
+      `Repo Face parent interpreter retrying its own malformed output for job ${job.id}: ${firstInterpretation.reasons.join("; ")}`,
+    );
+    const secondInterpretation = await interpretRepoFaceTurnOutput(provider, job, firstText, {
+      attempt: 2,
+    });
+    if (secondInterpretation.decision === "route" && secondInterpretation.parseable) {
+      return routeRepoFaceInterpretedOutput(firstResponse, secondInterpretation, {
+        job,
+        faceOutputText: firstText,
+      });
+    }
+
+    await auditLog.record({
+      type: "repo_face.parent_interpreter_drop",
+      actorId: job.requester.id,
+      jobId: job.id,
+      provider: job.provider,
+      details: {
+        decision: secondInterpretation.decision,
+        reasons: secondInterpretation.reasons,
+        owner: "parent_interpreter",
+      },
+    });
+    console.warn(
+      `Repo Face parent interpreter dropped job ${job.id} after malformed interpreter retry: ${secondInterpretation.reasons.join("; ")}`,
+    );
+    return dropRepoFaceActionBlocks(firstResponse, secondInterpretation);
+  }
+
   await auditLog.record({
     type: "repo_face.parent_interpreter_retry",
     actorId: job.requester.id,
@@ -597,6 +640,7 @@ async function executeRepoFaceJobWithInterpreter(
   const retryInterpretation = retryForbiddenTools.length > 0
     ? {
         decision: "drop" as const,
+        parseable: true,
         reasons: [
           `Face used forbidden substrate/tool-discovery tool(s) on retry: ${retryForbiddenTools.join(", ")}`,
           "Dropping action blocks rather than routing a turn that crossed the child/tool boundary.",
@@ -630,6 +674,7 @@ async function executeRepoFaceJobWithInterpreter(
 
 interface RepoFaceParentInterpretation {
   decision: "route" | "retry" | "drop";
+  parseable: boolean;
   reasons: string[];
   routedOutput?: string;
 }
@@ -714,8 +759,9 @@ function parseRepoFaceParentInterpretation(
 ): RepoFaceParentInterpretation {
   const block = parseInterpretationBlock(interpretationText);
   const decision = block.decision?.trim().toLowerCase();
+  const parseable = decision === "route" || decision === "retry" || decision === "drop";
   const parsedDecision =
-    decision === "route" || decision === "retry" || decision === "drop"
+    parseable
       ? decision
       : attempt === 1
         ? "retry"
@@ -728,7 +774,12 @@ function parseRepoFaceParentInterpretation(
 
   return {
     decision: parsedDecision,
-    reasons: reasons.length > 0 ? reasons : ["parent interpreter did not provide a parseable reason"],
+    parseable,
+    reasons: reasons.length > 0
+      ? reasons
+      : [parseable
+        ? "parent interpreter did not provide a parseable reason"
+        : "parent interpreter did not provide a parseable INTERPRETATION decision"],
     routedOutput: extractRoutedRepoFaceOutput(interpretationText),
   };
 }
@@ -886,8 +937,10 @@ function dropRepoFaceActionBlocks(
   interpretation: RepoFaceParentInterpretation,
 ): ProviderResponse {
   const text = fitDiscordMessage(response.outputText ?? response.summary);
-  const privateSummary = stripRepoIdentityPostIntents(text) ||
-    `Parent interpreter dropped repo Face action blocks: ${interpretation.reasons.join("; ")}`;
+  const stripped = stripRepoIdentityPostIntents(text);
+  const privateSummary = stripped !== text && stripped.trim().length > 0
+    ? stripped
+    : `Parent interpreter dropped repo Face output: ${interpretation.reasons.join("; ")}`;
   return {
     ...response,
     outputText: privateSummary,

notes/fresh-workspace-handoff.md

@@ -17,7 +17,7 @@ Git history and smoke artifacts carry routine proof. `state/evidence.jsonl` shou
 ## Current Shape
 
 - Discord bot and worker may be live, but the unattended repo Face / CTB swarm brake is active as of 2026-05-26.
-- The swarm was paused because live Face turns leaked owner-reply/Mind scaffolding into public Discord: Mind was being wrapped by the owner Discord reply prompt, then a worker fallback treated plain malformed Mind prose as public `SAY`. The fixed worker stack was restarted on 2026-05-26, typecheck/build pass, `agents:interpreter-loop-smoke` passes, and `agents:heartbeat-model-smoke -- --model gpt-5.4 --reasoning-effort none` passes. The Mind prompt has now been slimmed from incident-by-incident scar tissue into semantic route/retry/drop judgment plus a hard STATE NOTE/ARTICLE/SAY DSL contract. Keep `state/agent-swarm-paused.json` active until the operator explicitly asks to thaw; before thawing, run one controlled current-room turn and inspect model-output logs for `INTERPRETATION` plus explicit DSL actions only. Bifrost dispatch remains disabled by default behind `BIFROST_DISPATCH_ENABLED=false`.
+- The swarm is paused because live Face turns leaked owner-reply/Mind scaffolding into public Discord twice on 2026-05-26. The first failure wrapped parent Mind in the owner Discord reply prompt and let malformed plain Mind prose become public `SAY`. The second thaw attempt showed the deeper split: malformed parent Mind output caused the worker to ask the Face to revise, and that retry path turned owner-question boilerplate into Bifrost/Nibu `SAY`. The current worker retries malformed Mind output as Mind-owned interpretation, drops to diagnostic private summary instead of re-prompting the Face when Mind stays malformed, and never preserves raw dropped prose as the job summary. `scripts/inspect-model-output-logs.mjs` now flags repo-face parent interpreter records missing an `INTERPRETATION` decision. Keep `state/agent-swarm-paused.json` active until a new controlled thaw explicitly proves the live worker, parent Mind, model-output inspection, and Discord side effects all agree.
 - The teardown authority remains `notes/agent-swarm-teardown-plan-2026-05-21.md`, but it is now a restart checklist and smell ledger rather than a stop sign. Keep using it when a Face/Bifrost/CTB boundary smells like prompt/parser/transport Jenga.
 - The repo-controlled pause flag is `state/agent-swarm-paused.json`. While it is missing or has `"paused": false`, swarm runners may operate. While it has `"paused": true` or is malformed, the Face heartbeat runner skips and the orchestrator skips agent-swarm organs. This is the brake. Do not route around it.
 - Postgres owns jobs, audit events, interaction memory, and rate-limit state.

scripts/inspect-model-output-logs.mjs

@@ -64,6 +64,11 @@ function inspect(records) {
   const speechSmells = records.filter((record) =>
     /repo-face heartbeat|heartbeat from|bright bridge note|tiny fish sorting note|librarian note|maintenance pass/i.test(String(record.finalMessage ?? "")),
   );
+  const malformedRepoFaceInterpreter = records.filter((record) =>
+    record.command === "repo-face-rumination" &&
+    record.promptMarker === "repo-face-turn-interpreter" &&
+    !/\bINTERPRETATION\b[\s\S]*?\bdecision:\s*(route|retry|drop)\b/i.test(String(record.finalMessage ?? "")),
+  );
   const quotaSignals = records.filter((record) =>
     /quota|rate limit|rate-limit|usage limit|capacity|too many requests|(?:http|status|code|error)\s*429|429\s*(?:too many requests|rate)|insufficient_quota|model.*unavailable|model.*access|limit exceeded/i.test(
       `${record.stderrTail ?? ""}\n${record.stdoutTail ?? ""}\n${record.handoffReason ?? ""}`,
@@ -74,6 +79,7 @@ function inspect(records) {
   pushWarning(warnings, noFinalMessage.length, "model runs with no final message", noFinalMessage);
   pushWarning(warnings, forbiddenToolRecords.length, "Face/model runs using forbidden substrate tools", forbiddenToolRecords);
   pushWarning(warnings, speechSmells.length, "robotic/provenance speech smells", speechSmells);
+  pushWarning(warnings, malformedRepoFaceInterpreter.length, "repo Face parent interpreter outputs missing INTERPRETATION decision", malformedRepoFaceInterpreter);
   pushWarning(warnings, quotaSignals.length, "quota/rate/capacity fallback signals", quotaSignals);
 
   return {
@@ -84,6 +90,7 @@ function inspect(records) {
     noFinalMessage: summarizeRecords(noFinalMessage),
     forbiddenToolRecords: summarizeRecords(forbiddenToolRecords),
     speechSmells: summarizeRecords(speechSmells),
+    malformedRepoFaceInterpreter: summarizeRecords(malformedRepoFaceInterpreter),
     quotaSignals: summarizeRecords(quotaSignals),
     warnings,
     recent: summarizeRecords(records.slice(-12)),

state/agent-swarm-paused.json

@@ -1,5 +1,5 @@
 {
   "paused": true,
-  "reason": "Paused after live Face turns leaked owner-Codex/bridge artifact text into public Discord despite Face stdout choosing private silence. Fix side-effect extraction/routing before thawing.",
-  "updatedAt": "2026-05-26T11:29:00Z"
+  "reason": "Re-paused after live Bifrost Face post leaked owner-question wording into public Discord during thaw validation.",
+  "updatedAt": "2026-05-26T19:45:00.000Z"
 }

state/map.yaml

@@ -14,13 +14,17 @@ objective:
     Aetheria lore, while keeping the system understandable enough to modify
     safely.
   current: >
-    Keep the repo Face swarm paused after the 2026-05-26 Mind/owner-reply leak
+    Keep the repo Face swarm paused after the 2026-05-26 Mind/owner-reply leaks
     until a controlled thaw verifies the Face/Mind/speech boundary in current
-    room context. Bifrost work routing remains paused because the governance
-    bridge proved too disorienting for live social traffic. Parent Mind now has
-    a slimmer semantic prompt with a hard DSL contract: interpret natural Face
-    thought, decide route/retry/drop, and emit only explicit STATE NOTE, ARTICLE,
-    or SAY blocks for side effects.
+    room context. The second thaw attempt proved malformed parent Mind output
+    was being treated as a Face retry, allowing owner-question boilerplate to
+    become public Bifrost/Nibu SAY. The worker now treats malformed Mind output
+    as Mind-owned retry/drop and the model-output inspector flags interpreter
+    records missing INTERPRETATION decisions. Bifrost work routing remains
+    paused because the governance bridge proved too disorienting for live social
+    traffic. Parent Mind has a slimmer semantic prompt with a hard DSL contract:
+    interpret natural Face thought, decide route/retry/drop, and emit only
+    explicit STATE NOTE, ARTICLE, or SAY blocks for side effects.
 
 constraints:
   - Public traffic still rides one remote Qwen box plus local embedding and ops infrastructure.
@@ -57,7 +61,7 @@ data_flow:
     - keep persistent state lean enough that re-entry sharpens the next turn instead of drowning it
 
 active_subgoals:
-  - Treat `notes/agent-swarm-teardown-plan-2026-05-21.md` as the historical teardown authority and the restart checklist. The repo Face swarm is currently paused after the 2026-05-26 owner-reply/Mind leak, and should thaw only through controlled evidence that the scheduler, prompt, model-output logs, parent Mind, and worker side-effect rails still have explicit observable boundaries. Keep watching for prompt/parser/transport Jenga before adding new swarm features.
+  - Treat `notes/agent-swarm-teardown-plan-2026-05-21.md` as the historical teardown authority and the restart checklist. The repo Face swarm is currently paused after the 2026-05-26 owner-reply/Mind leaks, including the second thaw failure where malformed parent Mind output was converted into public owner-question SAY. It should thaw only through controlled evidence that the scheduler, prompt, model-output logs, parent Mind retry/drop ownership, and worker side-effect rails still have explicit observable boundaries. Keep watching for prompt/parser/transport Jenga before adding new swarm features.
   - Make the new project-state surfaces the default continuity path for nontrivial VoidBot work.
   - Keep the rebuilt moderation/self-state foundation on canonical CultCache `.cc` typed operations, never an editable JSON working projection.
   - Keep the typed moderation, mood, repo Face, and watchdog organs behind the single `GameCult Local Orchestrator` scheduled pulse; the per-organ scheduled tasks are legacy installer surfaces and should stay disabled unless a deliberate isolation test needs one. Bifrost dispatch is disabled by default behind `BIFROST_DISPATCH_ENABLED=false` until the governance bridge is rebuilt cleanly. As of the 2026-05-21 restart, the orchestrator task is enabled and is the live swarm pulse owner.