pi-goal-list-loop-audit 0.27.4 → 0.27.6

package/extensions/goal-settings.ts

@@ -71,8 +71,13 @@ export interface Settings {
    * its quota; "agent-default" restores upstream behavior. Applies to NEW
    * sessions (pi-subagents registers agents at session start). */
   /** v0.26.0: reviewer (post-completion follow-up enqueuer) config —
-   * project-scoped; see extensions/reviewer.ts DEFAULT_REVIEWER_CONFIG. */
+   * project-scoped; see extensions/reviewer.ts DEFAULT_REVIEWER_CONFIG.
+   * v0.27.5: superseded by `postaudit` (same shape, terminology reflects
+   * the auditor-adjacent role). Both keys are read; `postaudit` wins
+   * when both are present. `reviewer` is kept for backwards compat. */
   reviewer?: Record<string, unknown>;
+  /** v0.27.5: post-completion audit config. Same shape as `reviewer`. */
+  postaudit?: Record<string, unknown>;
   subagentModelStrategy?: SubagentModelStrategy;
   /** v0.24.6: per-agent-type model pin, e.g. { "Explore": "minimax/MiniMax-M3" }.
    * Always wins over subagentModelStrategy — the managed override is written
@@ -140,6 +145,7 @@ export const SETTINGS_KEYS: Array<keyof Settings> = [
   "stallEscalationRefires",
   "stallShortWords",
   "stallSimilarityThreshold",
+  "postaudit",
 ];
 
 /** Where each effective setting comes from (for the /glla display). */

package/extensions/loops/goal.ts

@@ -683,7 +683,12 @@ function fireReviewer(
 ): void {
   try {
     const settings = loadSettings(ctx.cwd);
-    const config = resolveReviewerConfig(settings.reviewer as Partial<ReviewerConfig> | undefined);
+    // v0.27.5: dual-read `reviewer` (legacy) and `postaudit` (new) settings
+    // keys. `postaudit` takes precedence when both are present — the
+    // existing settings file shape is preserved; the rename is purely
+    // vocabulary on the user-facing surface.
+    const reviewerBlock = (settings.postaudit ?? settings.reviewer) as Partial<ReviewerConfig> | undefined;
+    const config = resolveReviewerConfig(reviewerBlock);
     if (opts.mode) config.mode = opts.mode;
     const sources: Array<{ name: string; text: string }> = [];
     try {
@@ -728,6 +733,19 @@ function fireReviewer(
     if (!outcome.fired && outcome.suppressedReason && opts.manual) {
       ctx.ui.notify(`Reviewer suppressed: ${outcome.suppressedReason}`, "info");
     }
+    // v0.27.5: surface the silent review to interactive users. The internal
+    // runReviewer notify fires DURING the goal-completion handler, easy to
+    // miss because pi is busy transitioning state. The second notify
+    // arrives AFTER everything settles and points at the file directly.
+    // Skipped when manual=true (manual /review has its own UX already) and
+    // when the runner wasn't fired (suppressed / not applicable).
+    if (!opts.manual && outcome.fired && outcome.reportPath) {
+      const relPath = path.relative(ctx.cwd, outcome.reportPath) || outcome.reportPath;
+      ctx.ui.notify(
+        `↳ review written: ${relPath}${outcome.enqueued ? ` (${outcome.enqueued} enqueued to /list)` : ""}${outcome.proposed ? ` (${outcome.proposed} /goal proposed)` : ""}`,
+        "info",
+      );
+    }
   } catch (err) {
     ctx.ui.notify(`Reviewer failed (non-fatal): ${err instanceof Error ? err.message : String(err)}`, "warning");
   }
@@ -3216,6 +3234,13 @@ async function cmdSettings(args: string, ctx: ExtensionContext): Promise<void> {
     await cmdReviewerSettings(ctx);
     return;
   }
+  // v0.27.5: postaudit is the new vocabulary (the post-completion auditor).
+  // Both keywords open the same config menu and resolve to the same settings
+  // key — the legacy `reviewer` label is kept for backwards compatibility.
+  if (/^postaudit\b/.test(trimmed)) {
+    await cmdReviewerSettings(ctx);
+    return;
+  }
   if (!trimmed) {
     if (ctx.hasUI) {
       await openSettingsUI(ctx);
@@ -3245,6 +3270,9 @@ async function cmdSettings(args: string, ctx: ExtensionContext): Promise<void> {
         fmt("wedgeAlertMinutes", "wedgeAlert"),
         fmt("stallShortWords", "stallShortWords"),
         fmt("stallSimilarityThreshold", "stallSimilarityThreshold"),
+        // v0.27.5: post-completion auditor config — read either the new
+        // `postaudit` key or the legacy `reviewer` key (postaudit wins).
+        `postaudit: ${JSON.stringify(loadSettings(ctx.cwd).postaudit ?? loadSettings(ctx.cwd).reviewer ?? {}) || '(unset — defaults)'}`,
         // v0.25.6: effective per-type subagent model resolution.
         ...["Explore", "Plan", "general-purpose"].map(
           (t) => `subagent ${t}: ${resolveEffectiveSubagentModel(t, loadSettings(ctx.cwd), (ctx.model as any)?.id ? `${(ctx.model as any).provider}/${(ctx.model as any).id}` : undefined)}`,
@@ -3576,7 +3604,8 @@ export default function (pi: ExtensionAPI): void {
       ["stats", "per-project ledger rollups: /glla stats [json|premature|project=<path>]"],
       ["audits", "audit-log browser: /glla audits [N|full] — recent verdicts from .pi-glla/audits.jsonl"],
       ["autoaccept=", "on: drafts activate without the Confirm dialog (unattended rigs)"],
-      ["reviewer", "reviewer config menu (post-completion follow-up enqueuer)"],
+      ["reviewer", "reviewer config menu (alias of postaudit — post-completion follow-up enqueuer)"],
+      ["postaudit", "post-completion audit config menu (the new name for /glla reviewer)"],
       ["project", "write a project override: /glla project key=value"],
     ]),
     handler: settingsHandler,

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "pi-goal-list-loop-audit",
-  "version": "0.27.4",
-  "description": "Goal. Loop. Audit. Done. — a pi-coding-agent extension that supervises long-running work, with isolated auditor on each completion. Beat bamboozling by design: the auditor runs in a fresh session with no extensions, no skills, no editor — only the read tools needed to verify your goal.",
+  "version": "0.27.6",
+  "description": "Goal. Loop. Audit. Done. \u2014 a pi-coding-agent extension that supervises long-running work, with isolated auditor on each completion. Beat bamboozling by design: the auditor runs in a fresh session with no extensions, no skills, no editor \u2014 only the read tools needed to verify your goal.",
   "license": "MIT",
   "author": "dracon",
   "type": "module",
@@ -43,7 +43,9 @@
   },
   "scripts": {
     "check": "tsc --noEmit",
-    "test": "node --experimental-strip-types --test tests/*.test.ts"
+    "test": "bun test",
+    "test:node": "node --experimental-strip-types --test tests/*.test.ts",
+    "test:all": "bun test && tsc --noEmit"
   },
   "peerDependencies": {
     "@earendil-works/pi-agent-core": "*",

package/prompts/goal-loop-continuation.md

@@ -52,6 +52,7 @@ When the agent calls any of these, the orchestrator tracks the call and persists
 - **Default to subagents.** For any task that decomposes into independent chunks, spawn `Agent` subagents. Use `Explore` for read-only research, `general-purpose` for implementation, `Plan` for architecture. Spawn multiple in PARALLEL — don't serialise through your own context. You remain the single writer: synthesize findings and apply edits yourself.
 - **Eager continuation.** When in doubt, KEEP GOING on sub-tasks. If a subagent fails, retry with a different approach. Don't ask permission to continue — just continue. Pause only when you are genuinely blocked on information that does not exist in the repo, or the user explicitly pauses you.
 - **Bound every long command.** Wrap test suites, builds, and dev servers in `timeout <seconds>` (e.g. `timeout 120 bun test src/lib`). An unbounded command that hangs burns an hour; a bounded one burns two minutes and tells you it hung. If a command produces no output for many minutes, treat it as hung: kill it, diagnose why, rerun bounded.
+- **Chunk output near context-full.** When the conversation is heavy (long-running audit, deep debug, big rollout), prefer smaller commits, smaller tool outputs, and focused reasoning — one or two punchy paragraphs, one well-scoped tool call at a time. Don't try to fit a thousand lines of work into one reply. glla's 0.27.2 auto-continue fires on `stop_reason="length"` (the output-token cap) and will reschedule you anyway; pre-empting by chunking is cheaper than recovering from the cap. Save large file writes for their own turns; emit them only when you have the next read step ready to follow.
 
 ## WHEN THE AUDITOR DISAPPROVES