pi-goal-list-loop-audit 0.39.21 → 0.39.23

package/CHANGELOG.md

@@ -2,6 +2,34 @@
 
 ## Unreleased
 
+## 0.39.23 — terminal summary hardening: agent-proof headings, punctuation, clips (2026-10-10)
+
+### Summary render
+
+- Agent-written `###`/`####` lines inside summary findings demote to
+  bold on chat (the Pi TUI has no heading style); archives keep real
+  headings.
+- Finding outcomes strip stranded leading punctuation, matching the
+  reason husk fix.
+- The headline clipper extends through a closing backtick instead of
+  stranding an unmatched one inside a code span.
+- The `complete_goal` `leftOut` guidance forbids repeating Unresolved
+  facts (Unresolved = open problems; leftOut = deferred scope + why).
+
+## 0.39.22 — stalled-audit UI: dead audits named on every surface (2026-10-10)
+
+### Audit visibility
+
+- Past one hour of worker silence with no in-budget tool, a quiet audit
+  escalates to **stalled** on every surface: the detailed card drops the
+  live-worker claim (`no worker responds`) and offers the retry
+  (`/goal resume retries the claim`); the compact glance card and footer
+  agree (`AUDIT STALLED`, error tone, ✖ glyph) instead of claiming
+  `detached worker` / `No action needed`. The quiet watcher fires its own
+  one-shot stall notice.
+- Elapsed ages read in days past 24h (`15d 00h`, not `360h 00m`).
+- Display-only: no retry or recovery behavior changed.
+
 ## 0.39.21 — 5h retry-window policy, stall-proof test cleanup (2026-10-09)
 
 ### Recovery policy

package/docs/INDEX.md

@@ -31,7 +31,7 @@
 ## Versions and shipped content
 
 [CHANGELOG.md](../CHANGELOG.md) records the release history from
-v0.35.14–v0.39.21 and unreleased work. `/glla version` reports the loaded version
+v0.35.14–v0.39.23 and unreleased work. `/glla version` reports the loaded version
 and registry comparison. A checkout can contain changes not yet published.
 
 The package includes this documentation, [examples](../examples/),

package/docs/RUNTIME-INVENTORY.md

@@ -3,8 +3,8 @@
 Generated by `node scripts/generate-inventory.mjs`. Run with `--check` to detect drift. Counts describe files and literal registrations, not executed coverage.
 
 - Extension/script files: 109
-- Extension/script lines: 67486
-- Top-level test files: 378
+- Extension/script lines: 67574
+- Top-level test files: 380
 - Ambient bridge names: 179 (audit baseline: 179)
 - Commands: `/glla`, `/goal`, `/list`, `/loop`, `/review`
 - Tools: `audit_project_increment`, `block_project_requirement`, `claim_project_task`, `complete_goal`, `complete_task`, `list_activate`, `list_add`, `list_status`, `pause_goal`, `plan_project_increment`, `propose_goal_draft`, `propose_loop_draft`, `propose_loop_refine`, `propose_project_refinement`, `propose_project_requirements`, `propose_task_list`, `record_goal_judgment`, `resume_goal`, `unblock_project_requirement`, `update_task_batch`, `update_task_status`, `wait_for_background`
@@ -14,10 +14,10 @@ Generated by `node scripts/generate-inventory.mjs`. Run with `--check` to detect
 | Module | Lines |
 |---|---:|
 | [extensions/goal-loop-core.ts](../extensions/goal-loop-core.ts) | 5043 |
-| [extensions/loops/goal-tools.ts](../extensions/loops/goal-tools.ts) | 4227 |
+| [extensions/loops/goal-tools.ts](../extensions/loops/goal-tools.ts) | 4229 |
 | [extensions/loops/goal-activation.ts](../extensions/loops/goal-activation.ts) | 3880 |
 | [extensions/goal-commands.ts](../extensions/goal-commands.ts) | 3614 |
-| [extensions/goal-loop-display.ts](../extensions/goal-loop-display.ts) | 2858 |
+| [extensions/goal-loop-display.ts](../extensions/goal-loop-display.ts) | 2884 |
 | [extensions/goal-loop-auditor-process.ts](../extensions/goal-loop-auditor-process.ts) | 2685 |
 | [extensions/loops/goal-auditor-hooks.ts](../extensions/loops/goal-auditor-hooks.ts) | 2654 |
 | [extensions/goal-heartbeat.ts](../extensions/goal-heartbeat.ts) | 2104 |

package/extensions/completion-summary.ts

@@ -509,10 +509,14 @@ function findingPresentation(finding: string, chat: boolean): { outcome: string;
   // evidence tokens ride mid-sentence, so strip a leading `,;:` husk.
   // Content words are untouched — only the join seam is cleaned.
   const normalizedReason = [...new Set(reasonParts)].join(" · ").trim().replace(/^[,;:]+/, "").trim();
+  // v0.39.23 (field 2026-10-10 — Clean-Web card `- , - with …`): the same
+  // stranded-punctuation husk can lead the outcome when the agent doubles
+  // punctuation around the em-dash split. Strip it like the reason husk.
+  const stripHusk = (text: string): string => text.trim().replace(/^[,;:]+/, "").trim();
   return {
     // Chat removes machine receipts; the archive keeps the full evidence
     // prose. The outcome/reason split itself is shared by both surfaces.
-    outcome: chat ? chatNarrative(parsed.outcome) : sanitizeDisplayText(parsed.outcome),
+    outcome: stripHusk(chat ? chatNarrative(parsed.outcome) : sanitizeDisplayText(parsed.outcome)),
     reason: chat ? chatNarrative(normalizedReason) : sanitizeDisplayText(normalizedReason),
     evidence: parsed.evidence,
   };
@@ -1058,6 +1062,15 @@ export function buildRichTerminalParts(args: {
   };
 }
 
+/** Demote agent-written markdown headings to TUI-safe bold for chat.
+ * A `#`-heading line that survives into chat content renders literally
+ * (the Pi TUI has no heading style); `**bold**` renders everywhere the
+ * card does. Archive/non-chat lines bypass this — a .md file renders
+ * headings natively. */
+function terminalizeHeadings(line: string): string {
+  return line.replace(/^(\s*)#{1,6}\s+(.*\S)\s*$/, "$1**$2**");
+}
+
 /** Compose parts + banner + section headers into markdown lines.
  * v0.38.50: the duration line rides directly under the headline.
  * Audit 2026-09-13: findings-first order is pinned — findings, then the
@@ -1081,22 +1094,31 @@ export function composeRichTerminalLines(parts: RichTerminalParts): string[] {
   // the screen literally. Chat/terminal surfaces get plain section labels;
   // the archived markdown keeps its headings (a .md file renders them).
   const head = (text: string): string => (parts.chat ? text : `### ${text}`);
+  // v0.39.23 (field 2026-10-10 — Clean-Web terminal screenshots): agents
+  // write `###`/`####` headers INSIDE finding/remaining/next text and the
+  // Pi TUI renders bold/lists but not `#` headings, so they reached the
+  // screen literally. Chat-facing content demotes markdown headings to
+  // bold (which the TUI renders); the archived markdown keeps them.
+  const contentLine = (line: string): string => {
+    const clean = sanitizeDisplayText(line);
+    return parts.chat ? terminalizeHeadings(clean) : clean;
+  };
   if (parts.summaryLines.length > 0) {
-    lines.push(head("Summary"), ...parts.summaryLines.map((line) => sanitizeDisplayText(line)), "");
+    lines.push(head("Summary"), ...parts.summaryLines.map(contentLine), "");
   }
   if (parts.findingLines.length > 0) {
-    lines.push(head("What Changed"), ...parts.findingLines.map((line) => sanitizeDisplayText(line)), "");
+    lines.push(head("What Changed"), ...parts.findingLines.map(contentLine), "");
   }
   if (parts.remainingLines.length > 0) {
-    lines.push(head("Remaining"), ...parts.remainingLines.map((line) => sanitizeDisplayText(line)), "");
+    lines.push(head("Remaining"), ...parts.remainingLines.map(contentLine), "");
   }
   if (parts.verificationSummaryLine) {
-    lines.push(head("Verification"), sanitizeDisplayText(parts.verificationSummaryLine), "");
+    lines.push(head("Verification"), contentLine(parts.verificationSummaryLine), "");
   } else if (!parts.chat && parts.tableLines.length > 0) {
     lines.push("### Verification Evidence", ...parts.tableLines.map((line) => sanitizeDisplayText(line)), "");
   }
   if (parts.nextLines.length > 0) {
-    lines.push(head("Next"), ...parts.nextLines.map((line) => sanitizeDisplayText(line)), "");
+    lines.push(head("Next"), ...parts.nextLines.map(contentLine), "");
   }
   if (parts.repoLines.length > 0) {
     lines.push(head("Final Repository State"), ...parts.repoLines.map((line) => `- ${sanitizeDisplayText(line)}`), "");

package/extensions/finding-lead.ts

@@ -166,9 +166,24 @@ export function clipSummaryValue(value: string, limit: number): string {
   headUnits.forEach((unit, i) => {
     if (/[,;:·—–(+[\[]/u.test(unit)) boundary = i;
   });
+  // v0.39.23 (field 2026-10-10 — Clean-Web card headline stranded
+  // "lands on `sendThreshold…" with an unmatched backtick): never cut
+  // inside an inline code span. When the kept head holds an unmatched
+  // backtick, extend through the closing backtick when it lands within a
+  // short bound; the capped overrun keeps the span readable instead of
+  // dangling.
+  const finalize = (text: string): string => {
+    const opens = (text.match(/`/g) ?? []).length;
+    if (opens % 2 === 0) return `${text}…`;
+    const consumed = [...text].length;
+    const rest = units.slice(consumed, consumed + 40).join("");
+    const close = rest.indexOf("`");
+    if (close >= 0) return `${text}${rest.slice(0, close + 1)}…`;
+    return `${text}…`;
+  };
   if (boundary >= floor) {
     const cut = headUnits.slice(0, boundary).join("").replace(/[,;:·—–(+[\[\s]+$/u, "");
-    if ([...cut].length >= Math.min(floor, 16)) return `${cut}…`;
+    if ([...cut].length >= Math.min(floor, 16)) return finalize(cut);
   }
   let space = -1;
   headUnits.forEach((unit, i) => {
@@ -177,5 +192,5 @@ export function clipSummaryValue(value: string, limit: number): string {
   const kept = (space > capped / 2 ? headUnits.slice(0, space).join("") : head)
     .trimEnd()
     .replace(/[,;:·—–(+\[]$/u, "");
-  return `${kept}…`;
+  return finalize(kept);
 }

package/extensions/goal-loop-display.ts

@@ -58,7 +58,11 @@ export function fmtElapsed(ms: number): string {
   // liveness signal — minute-only granularity looks frozen on a 1s tick.
   if (m < 60) return `${m}m ${String(s % 60).padStart(2, "0")}s`;
   const h = Math.floor(m / 60);
-  return `${h}h ${String(m % 60).padStart(2, "0")}m`;
+  if (h < 24) return `${h}h ${String(m % 60).padStart(2, "0")}m`;
+  // v0.39.22: day granularity past 24h — a 15-day stalled audit reads
+  // "15d 00h", not "360h 00m". Only affects renders past a day.
+  const d = Math.floor(h / 24);
+  return `${d}d ${String(h % 24).padStart(2, "0")}h`;
 }
 
 export function fmtTokens(n: number): string {
@@ -801,11 +805,17 @@ export interface AuditDisplayProgress {
   sessionPath?: string;
 }
 
-type AuditorDisplayPhase = "queued" | "running" | "quiet" | "blocked" | "awaiting-verdict";
+type AuditorDisplayPhase = "queued" | "running" | "quiet" | "stalled" | "blocked" | "awaiting-verdict";
 /** v0.35.15: exported — the parent-side quiet watcher (goal-ui ticker) uses
  * the same threshold for the one-shot proactive notify so the notification
  * and the status chip can never disagree about when "quiet" begins. */
 export const AUDITOR_QUIET_MS = 3 * 60_000;
+/** v0.39.22 (field 2026-10-10: seo goal quiet 15 days, worker dead):
+ * quiet past this bound with no worker activity and no in-budget tool is
+ * a dead worker, not a slow one. An hour of total silence exceeds any
+ * legitimate inter-tool gap; the stalled presentation names the death and
+ * offers the retry (/goal resume) instead of only the discard. */
+export const AUDITOR_STALLED_MS = 60 * 60_000;
 /** v0.35.15: how long the "silent Xm then resumed" fact stays in the
  * detailed auditor card after a quiet stretch ends — long enough to be seen,
  * short enough not to become permanent noise. */
@@ -911,6 +921,7 @@ export function auditorDisplayPhase(g: Goal, audit: AuditDisplayProgress | null
       toolAgeMs < audit.toolTimeoutMs
     )
       return "running";
+    if (age > AUDITOR_STALLED_MS) return "stalled";
     return "quiet";
   }
   // Same lifecycle scope as above: a stale goal snapshot that still
@@ -959,6 +970,7 @@ function auditorPhaseLabel(phase: AuditorDisplayPhase): string {
     case "queued": return "queued";
     case "running": return "running";
     case "quiet": return "quiet";
+    case "stalled": return "stalled";
     case "blocked": return "blocked";
     case "awaiting-verdict": return "awaiting completion review";
   }
@@ -973,6 +985,7 @@ export function auditorPhaseGlyph(phase: AuditorDisplayPhase): string {
     case "queued": return "⋯";      // waiting to start
     case "running": return "▶";     // worker alive
     case "quiet": return "◌";       // silent — possibly stuck
+    case "stalled": return "✖";      // silent past the death bound — worker dead
     case "blocked": return "⛔";    // infra failure / needs intervention
     case "awaiting-verdict": return "✓"; // worker done, verdict pending
   }
@@ -1508,11 +1521,11 @@ function buildStatusTextBase(state: State, audit?: AuditDisplayProgress | null,
     // v0.35.15: leading phase glyph + draining activity meter — a glance
     // answers "is the audit alive?" without reading the sentence.
     const phaseText = `auditor ${auditorPhaseGlyph(phase)} ${observed}`;
-    const color = phase === "blocked" || phase === "quiet" ? "warning" : live ? "success" : "accent";
+    const color = phase === "stalled" ? "error" : phase === "blocked" || phase === "quiet" ? "warning" : live ? "success" : "accent";
     const activityMeter = auditorActivityMeter(audit, phase, now);
     const label = live
       ? `${paint(theme, "success", phaseText)} ${paint(theme, "accent", activityMeter)} ${activityBadge("AUDITOR · DETACHED · LIVE", now, theme)}`
-      : `${paint(theme, color, phaseText)} ${paint(theme, color === "warning" ? "warning" : "dim", activityMeter)}`;
+      : `${paint(theme, color, phaseText)} ${paint(theme, color === "warning" || color === "error" ? color : "dim", activityMeter)}`;
     // The persistent footer is the liveness surface only: host, phase, and
     // freshness. The auditor card above already owns the transition hint
     // ("next:"), the worker attribution ("detached worker"), and the
@@ -1521,7 +1534,7 @@ function buildStatusTextBase(state: State, audit?: AuditDisplayProgress | null,
     const quietAge = phase === "quiet" ? auditorActivityAge(audit, now) : undefined;
     const quietSuffix = quietAge !== undefined ? ` · silent ${fmtElapsed(quietAge)}` : "";
     const compactPrefix = extras?.compactAuditCard
-      ? `${paint(theme, color, durableLabel ?? (toolWait ? `AUDIT TOOL WAIT · ${sanitizeDisplayText(audit!.currentTool!)} ${fmtElapsed(now - audit!.currentToolStartedAt!)} / ${fmtElapsed(audit!.toolTimeoutMs!)}` : phase === "running" ? "AUDIT RUNNING" : phase === "quiet" ? "AUDIT QUIET — may be stuck" : phase === "blocked" ? "AUDIT BLOCKED" : phase === "queued" ? "AUDIT STARTING" : "AUDIT REVIEW"))} · `
+      ? `${paint(theme, color, durableLabel ?? (toolWait ? `AUDIT TOOL WAIT · ${sanitizeDisplayText(audit!.currentTool!)} ${fmtElapsed(now - audit!.currentToolStartedAt!)} / ${fmtElapsed(audit!.toolTimeoutMs!)}` : phase === "running" ? "AUDIT RUNNING" : phase === "quiet" ? "AUDIT QUIET — may be stuck" : phase === "stalled" ? "AUDIT STALLED — worker dead" : phase === "blocked" ? "AUDIT BLOCKED" : phase === "queued" ? "AUDIT STARTING" : "AUDIT REVIEW"))} · `
       : "";
     const activityAge = auditorActivityAge(audit, now);
     const freshness = extras?.compactAuditCard && phase !== "quiet" && !toolWait
@@ -1983,12 +1996,13 @@ function heldLoopLines(l: LoopState, now: number, theme?: DisplayTheme, width?:
 }
 
 /** Phase colour for the auditing card's lead row. Blocked/quiet are
- * warnings; a verdict that landed or live worker evidence is success;
+ * warnings, stalled is an error; a verdict that landed or live worker evidence is success;
  * queued/running-without-evidence stay neutral accent — a hung provider
  * must never wear success (v0.34.39). Colour always pairs with the phase
  * words (paint wraps the label, never replaces it), so a themeless
  * renderer loses colour but never meaning. */
 function auditorPhaseTone(phase: AuditorDisplayPhase, live: boolean): DisplayColor {
+  if (phase === "stalled") return "error";
   if (phase === "blocked" || phase === "quiet") return "warning";
   if (phase === "awaiting-verdict" || (phase === "running" && live)) return "success";
   return "accent";
@@ -2009,6 +2023,7 @@ function auditorCardModelRef(audit: AuditDisplayProgress | null | undefined, cla
  * row and the closing line can never name different actions. */
 function auditorNextAction(phase: AuditorDisplayPhase): string {
   if (phase === "quiet") return "/goal cancel discards the claim";
+  if (phase === "stalled") return "/goal resume retries the claim";
   if (phase === "blocked") return "/goal resume retries the claim";
   if (phase === "awaiting-verdict") return "completion review applying";
   if (phase === "queued") return "worker starting";
@@ -2082,7 +2097,11 @@ function auditingCardBlock(g: Goal, audit: AuditDisplayProgress | null | undefin
   const staleSeg = activity === undefined && durable?.stale && durable.idleMs !== undefined
     ? ` · no progress ${fmtElapsed(durable.idleMs)}`
     : "";
-  const lead = [`├─ auditor: ${paint(theme, auditorPhaseTone(phase, phaseLive), effectivePhaseLabel)}${detail} · detached worker · ${ageSeg}${staleSeg}`];
+  // v0.39.22: a stalled audit asserts no live worker — printing
+  // "detached worker" would claim a process that has not spoken for over
+  // an hour. The `auditor:` anchor stays; only the worker segment changes.
+  const workerSeg = phase === "stalled" ? "no worker responds" : "detached worker";
+  const lead = [`├─ auditor: ${paint(theme, auditorPhaseTone(phase, phaseLive), effectivePhaseLabel)}${detail} · ${workerSeg} · ${ageSeg}${staleSeg}`];
   // Lead row 2: current tool + time budget, effective model + thinking.
   // Absent facts are omitted, never invented. The tool observation lives
   // ONLY here (it moved out of the tail) so the card keeps its
@@ -2161,7 +2180,9 @@ function auditingCardBlock(g: Goal, audit: AuditDisplayProgress | null | undefin
     tail.push(`${i === 0 ? "├─" : "│ "} ${paint(theme, "dim", observation)}`);
   });
   const last = auditorLastActivity(audit, now);
-  if (phase === "quiet") {
+  if (phase === "stalled") {
+    tail.push(`└─ ${paint(theme, "error", `auditor stalled ${fmtElapsed(activity ?? 0)} — no worker activity, claim durable; /goal resume retries, /goal cancel discards`)}`);
+  } else if (phase === "quiet") {
     const quietMs = activity ?? 0;
     tail.push(`└─ ${paint(theme, "warning", `auditor quiet ${fmtElapsed(quietMs)}${last} — may be stuck; /goal cancel discards the claim`)}`);
   } else if (phase === "blocked") {
@@ -2306,7 +2327,12 @@ function goalLines(g: Goal, state: State, audit: AuditDisplayProgress | null | u
     if (extras?.compactAuditCard) {
       const phase = auditorDisplayPhase(g, audit, now);
       const durable = auditLifecycleProjection(g.pendingCompletion, { now });
-      const needsRecovery = auditRecoveryPending(g) || phase === "quiet" || phase === "blocked" || durable?.phase === "settling";
+      // v0.39.22: stalled needs recovery like quiet/blocked — without it
+      // the glance path falls through to "Audit review pending · detached
+      // worker" + "No action needed", preserving the can't-tell gap on
+      // the production surface. With it, the glance reuses the detailed
+      // lead (stalled · no worker responds) and its retry action row.
+      const needsRecovery = auditRecoveryPending(g) || phase === "quiet" || phase === "stalled" || phase === "blocked" || durable?.phase === "settling";
       const elapsed = auditorElapsedMs(audit, now);
       const finished = audit?.toolCalls?.filter(call => call.finishedAt !== undefined).length;
       const age = auditorActivityAge(audit, now);

package/extensions/loops/goal-tools.ts

@@ -637,6 +637,8 @@ function registerAgentTools(pi: any): void {
           "v0.38.37: what this turn deliberately left out (scope cut, deferred item + why, in plain words). " +
           "Renders under `### Remaining` in the user-facing terminal summary. " +
           "Bounds match the documented 10k-char value guard — longer input clips at a clause boundary, never a refusal. " +
+          "Never repeat a fact already covered by Unresolved: Unresolved names open problems and risks, leftOut names deliberately deferred scope and why. " +
+          "When both would say the same thing, put it in Unresolved once and omit leftOut. " +
           "Omit when nothing was deliberately left out — absent stays absent, never invented.",
       })),
       showVerification: Type.Optional(Type.Boolean({

package/extensions/loops/goal-ui.ts

@@ -914,6 +914,8 @@ function scheduleUIRefresh(): void {
 // suppresses the notify (it is automatic machinery), never the recording.
 let auditorQuietSince: number | null = null;
 let auditorQuietNotified = false;
+let auditorStallSince: number | null = null;
+let auditorStallNotified = false;
 let lastAuditorQuietStretch: { ms: number; endedAt: number } | null = null;
 
 /** Drive the quiet watcher once per UI tick. Returns a warning message to
@@ -929,6 +931,8 @@ export function __auditorQuietWatchTick(now = Date.now()): string | null {
       auditorQuietSince = null;
       auditorQuietNotified = false;
     }
+    auditorStallSince = null;
+    auditorStallNotified = false;
     return null;
   }
   const phase = auditorDisplayPhase(g, latestAuditProgress, now);
@@ -945,6 +949,23 @@ export function __auditorQuietWatchTick(now = Date.now()): string | null {
     }
     return null;
   }
+  // v0.39.22: a stalled worker gets its own one-shot notify — the quiet
+  // notice fired an hour ago and reads as routine by now. The claim stays
+  // durable either way; the notify names the retry, not just the discard.
+  if (phase === "stalled") {
+    if (auditorStallSince === null) {
+      auditorStallSince = typeof latestAuditProgress?.lastActivityAt === "number"
+        ? latestAuditProgress.lastActivityAt
+        : now;
+    }
+    if (!auditorStallNotified && !supervisorPaused(state)) {
+      auditorStallNotified = true;
+      return `glla: the detached auditor is STALLED — no worker activity for ${fmtElapsed(now - auditorStallSince)} and no tool running. The claim is durable; /goal resume retries it, /goal cancel discards it.`;
+    }
+    return null;
+  }
+  if (auditorStallSince !== null) auditorStallSince = null;
+  auditorStallNotified = false;
   // Left the quiet phase: record the ended stretch for the footer summary.
   if (auditorQuietSince !== null) {
     const ms = now - auditorQuietSince;
@@ -1212,6 +1233,8 @@ export function __testOnlyAuditorQuietWatchTick(
 export function __testOnlyResetAuditorQuietWatch(): void {
   auditorQuietSince = null;
   auditorQuietNotified = false;
+  auditorStallSince = null;
+  auditorStallNotified = false;
   lastAuditorQuietStretch = null;
 }
 

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "pi-goal-list-loop-audit",
-	"version": "0.39.21",
+	"version": "0.39.23",
 	"description": "Durable work supervision for Pi: turn outcomes into goals, audited queues and spec-driven builds, recover interruptions, and verify completion against evidence.",
 	"license": "AGPL-3.0-only",
 	"author": "dracon",