@mjasnikovs/pi-task 0.13.32 → 0.13.34

package/dist/task/auto-orchestrator.js

@@ -13,9 +13,9 @@ import { renderInlineMarkdown, stripInlineMarkdown } from './inline-markdown.js'
 import { AUTO_CLARIFY_PROMPT, AUTO_DECOMPOSE_PROMPT } from './auto-prompts.js';
 import { isDuplicateQuestion, MAX_DUP_STRIKES, DUP_REPROMPT_HINT } from './question-dedup.js';
 import { allocateAutoId, buildAutoBody, parseDecomposeList, parseTaskList, checkOffTask, stampTaskInProgress, findResumableAuto } from './auto-io.js';
-import { writeTaskFile, readTaskFile, updateTaskFrontMatter, taskFilePath } from './task-io.js';
+import { writeTaskFile, readTaskFile, updateTaskFrontMatter, taskFilePath, tasksDir } from './task-io.js';
 import { gitCommitAll } from './auto-commit.js';
-import { runGuidelineEnforcement, classifyEnforceChildFailure, ENFORCE_TIMEOUT_MS } from './enforce-guidelines.js';
+import { runGuidelineEnforcement, classifyEnforceChildFailure, ENFORCE_TIMEOUT_MS, ENFORCE_LOOP } from './enforce-guidelines.js';
 import { runWorker } from '../workers/pi-worker-core.js';
 import { runPhaseChild, prependHint, USER_CANCELLED } from './child-runner.js';
 import { SessionUI, registerBridgeCommand } from '../remote/bridge.js';
@@ -328,6 +328,20 @@ function defaultDeps(ctx, cwd, signal, title) {
                     lastLine = undefined;
                     contextUsage = undefined;
                     const startedAt = Date.now();
+                    // Per-pass debug log. The enforce child is otherwise
+                    // unobservable (it has no per-task file like the impl runner),
+                    // so a loop/timeout in the verifier was undiagnosable. One
+                    // rolling file under .pi-tasks/; each pass brackets its
+                    // tool/text lines with start/end markers naming the task and
+                    // the terminal outcome, so a kill shows the exact repeated
+                    // calls that tripped the loop detector. Fire-and-forget.
+                    const enforceLogPath = path.join(tasksDir(cwd2), 'enforce-debug.log');
+                    const logEnforce = (msg) => {
+                        void fsp
+                            .appendFile(enforceLogPath, `${new Date().toISOString()} ${msg}\n`)
+                            .catch(() => { });
+                    };
+                    logEnforce(`=== enforce start: ${taskTitle} ===`);
                     const stopLoader = startAutoLoader(enforceCtx, () => ({
                         title: taskTitle,
                         kind: 'enforce',
@@ -345,12 +359,16 @@ function defaultDeps(ctx, cwd, signal, title) {
                             signal: sig,
                             tools,
                             timeoutMs: ENFORCE_TIMEOUT_MS,
+                            loop: ENFORCE_LOOP,
                             onLine: line => {
                                 lastLine = line;
-                                phaseDeps.logDebug?.(`enforce: ${line}`);
+                                logEnforce(line);
                             }
                         });
                         const failure = classifyEnforceChildFailure(r);
+                        logEnforce(failure ?
+                            `=== enforce end: FAIL — ${failure} ===`
+                            : '=== enforce end: verdict captured ===');
                         if (failure)
                             throw new Error(failure);
                         return r.text;

package/dist/task/enforce-guidelines.d.ts

@@ -7,6 +7,21 @@ declare const ENFORCE_TOOLS = "read,grep,find,ls,edit,write";
  *  fix pass reads several files and may rewrite a few; give it room but never
  *  let a wedged child hang the whole run. */
 declare const ENFORCE_TIMEOUT_MS = 600000;
+/**
+ * Loop-guard tuning for the enforcement child. Unlike a read-only research
+ * worker, this pass's JOB is to rework files in place, so it legitimately reads,
+ * edits, and re-greps the SAME file many times — the default path-revisit
+ * heuristic (5 ops on one path) mislabels that as a runaway. It is disabled here
+ * (pathThreshold = Infinity); only the exact-match guard remains (a byte-for-byte
+ * identical call repeated, raised to 12 over a 40-call window), with the 10-min
+ * wall-clock timeout as the real runaway backstop. Validated against a real
+ * loop-kill on mx5 TASK_0002, where a single-file fix pass on queries.ts tripped
+ * the default detector at 5 revisits. */
+export declare const ENFORCE_LOOP: {
+    readonly window: 40;
+    readonly threshold: 12;
+    readonly pathThreshold: number;
+};
 export interface GuidelineDoc {
     /** Filenames found, in discovery order (e.g. ['AGENTS.md', 'CLAUDE.md']). */
     files: string[];

package/dist/task/enforce-guidelines.js

@@ -28,6 +28,21 @@ const ENFORCE_TOOLS = 'read,grep,find,ls,edit,write';
  *  fix pass reads several files and may rewrite a few; give it room but never
  *  let a wedged child hang the whole run. */
 const ENFORCE_TIMEOUT_MS = 600_000;
+/**
+ * Loop-guard tuning for the enforcement child. Unlike a read-only research
+ * worker, this pass's JOB is to rework files in place, so it legitimately reads,
+ * edits, and re-greps the SAME file many times — the default path-revisit
+ * heuristic (5 ops on one path) mislabels that as a runaway. It is disabled here
+ * (pathThreshold = Infinity); only the exact-match guard remains (a byte-for-byte
+ * identical call repeated, raised to 12 over a 40-call window), with the 10-min
+ * wall-clock timeout as the real runaway backstop. Validated against a real
+ * loop-kill on mx5 TASK_0002, where a single-file fix pass on queries.ts tripped
+ * the default detector at 5 revisits. */
+export const ENFORCE_LOOP = {
+    window: 40,
+    threshold: 12,
+    pathThreshold: Number.POSITIVE_INFINITY
+};
 /**
  * Read the guideline files that live directly in `cwd` (AGENTS.md, CLAUDE.md).
  * Returns null when none exist or all are empty — the caller treats that as a

package/dist/workers/pi-worker-core.d.ts

@@ -13,6 +13,17 @@ export interface RunWorkerInput {
     onLine?: (line: string) => void;
     /** Per-worker wall-clock timeout in ms. Defaults to RESEARCH_WORKER_TIMEOUT_MS. */
     timeoutMs?: number;
+    /**
+     * Per-worker loop-detector tuning. Defaults to the read-only research/impl
+     * guard (LOOP_WINDOW / LOOP_THRESHOLD, path threshold = exact threshold). An
+     * edit/fix pass legitimately revisits one file, so it can raise (or disable
+     * via Infinity) `pathThreshold` to keep only the exact-match guard.
+     */
+    loop?: {
+        window?: number;
+        threshold?: number;
+        pathThreshold?: number;
+    };
 }
 export interface RunWorkerResult {
     text: string;

package/dist/workers/pi-worker-core.js

@@ -71,7 +71,9 @@ export async function runWorker(input) {
         const invocation = getPiInvocation([...baseArgs, prompt]);
         const tStart = Date.now();
         let tFirstByte = null;
-        const loopDetector = new LoopDetector(LOOP_WINDOW, LOOP_THRESHOLD);
+        const loopWindow = input.loop?.window ?? LOOP_WINDOW;
+        const loopThreshold = input.loop?.threshold ?? LOOP_THRESHOLD;
+        const loopDetector = new LoopDetector(loopWindow, loopThreshold, input.loop?.pathThreshold ?? loopThreshold);
         // Capture the hit the detector reports (it also returns it to the unified
         // runner, which kills the child on a hit). Without capturing it here the
         // SIGTERM that kill produces would surface as a bare non-zero exit the

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@mjasnikovs/pi-task",
-    "version": "0.13.32",
+    "version": "0.13.34",
     "description": "Deterministic spec-orchestration for local models, with a bundled real-time remote web view and web/docs/fetch/worker subagent tools.",
     "type": "module",
     "main": "./dist/index.js",