@opengsd/gsd-pi 1.2.0-dev.5457a158 → 1.2.0-dev.822c9439

package/dist/headless-events.js

@@ -11,6 +11,7 @@
  * so wording stays in lockstep with detection.
  */
 import { isBlockedNoticeMessage, isManualResolutionNotice, isPauseNotice, isTerminalNotice, } from './resources/extensions/gsd/stop-notice.js';
+import { canonicalToolName } from './resources/extensions/gsd/engine-hook-contract.js';
 // ---------------------------------------------------------------------------
 // Exit Code Constants
 // ---------------------------------------------------------------------------
@@ -76,12 +77,13 @@ export const IDLE_TIMEOUT_MS = 15_000;
 // longer idle timeout to avoid killing the session prematurely (#808).
 export const NEW_MILESTONE_IDLE_TIMEOUT_MS = 120_000;
 const INTERACTIVE_HEADLESS_TOOLS = new Set(['ask_user_questions', 'secure_env_collect']);
+// Delegates to the shared normalizer seam (engine-hook-contract.ts) instead of
+// a hand-rolled parser. Behavior differs from the old parser only on malformed
+// MCP names: `mcp____tool` (empty server) and `mcp__server__` (empty tool) are
+// now returned unchanged rather than partially stripped — neither can match a
+// real tool name, so detection behavior is unaffected.
 export function canonicalHeadlessToolName(toolName) {
-    const name = String(toolName ?? '');
-    if (!name.startsWith('mcp__'))
-        return name;
-    const toolSeparator = name.indexOf('__', 'mcp__'.length);
-    return toolSeparator >= 0 ? name.slice(toolSeparator + 2) : name;
+    return canonicalToolName(String(toolName ?? ''));
 }
 function getCommandBlockContent(event) {
     if (event.type !== 'message_start' && event.type !== 'message_end')

package/dist/mcp-server.js

@@ -29,6 +29,7 @@ function isPlainObject(value) {
 // built via a template string so TypeScript's NodeNext resolver treats them as
 // `any` and skips static checking.
 const MCP_PKG = '@modelcontextprotocol/sdk';
+import { sanitizeSchemaForMoonshot } from '@gsd/pi-ai';
 export function mcpSdkSpecifier(subpath) {
     return `${MCP_PKG}/${subpath}.js`;
 }
@@ -65,7 +66,7 @@ export async function startMcpServer(options) {
         tools: tools.map((t) => ({
             name: t.name,
             description: t.description,
-            inputSchema: t.parameters,
+            inputSchema: sanitizeSchemaForMoonshot(t.parameters),
         })),
     }));
     // tools/call — execute the requested tool and return content blocks.

package/dist/resource-loader.d.ts

@@ -22,6 +22,8 @@ export declare function readManagedResourceVersion(agentDir: string): string | n
  * directory (e.g. pre-install verification).
  */
 export declare function computeResourceFingerprint(rootDir?: string): string;
+export declare function collectGsdBrowserPackageSkillReferences(content: string): string[];
+export declare function hasStaleGsdBrowserPackageSkill(skillsDir: string): boolean;
 export declare function getNewerManagedResourceVersion(agentDir: string, currentVersion: string): string | null;
 /**
  * Syncs a single bundled resource directory into the agent directory.
@@ -49,18 +51,20 @@ export declare function reconcileMergedNodeModules(agentNodeModules: string, hoi
 /** Build a cache fingerprint from packageRoot + sorted entry names of both directories */
 export declare function mergedFingerprint(hoisted: string, internal: string): string;
 /**
- * Syncs all bundled resources to agentDir (~/.gsd/agent/) on every launch.
+ * Initializes managed resources under agentDir (~/.gsd/agent/).
  *
  * - extensions/ → ~/.gsd/agent/extensions/   (overwrite when version changes)
  * - shared/     → ~/.gsd/agent/shared/       (overwrite when version changes)
  * - agents/     → ~/.gsd/agent/agents/        (overwrite when version changes)
  * - skills/     → ~/.gsd/agent/skills/        (overwrite when version changes)
+ * - gsd-browser skill → ~/.gsd/agent/skills/gsd-browser/ from @opengsd/gsd-browser
  * - GSD-WORKFLOW.md → ~/.gsd/agent/GSD-WORKFLOW.md (fallback for env var miss)
  *
- * Skips the copy when the managed-resources.json version matches the current
- * GSD version, avoiding ~128ms of synchronous cpSync on every startup.
- * After `npm update -g @glittercowboy/gsd`, versions will differ and the
- * copy runs once to land the new resources.
+ * Skips the full copy only when the managed-resources.json version, content
+ * fingerprint, and package-owned gsd-browser skill all match the current
+ * install, avoiding ~128ms of synchronous cpSync on steady-state startup.
+ * After `npm update -g @opengsd/gsd-pi`, versions will differ and the copy
+ * runs once to land the new resources.
  *
  * Inspectable: `ls ~/.gsd/agent/extensions/`
  */

package/dist/resource-loader.js

@@ -1,6 +1,7 @@
 import { createHash } from 'node:crypto';
 import { homedir } from 'node:os';
 import { chmodSync, copyFileSync, cpSync, existsSync, lstatSync, mkdirSync, readFileSync, readlinkSync, readdirSync, rmSync, symlinkSync, unlinkSync, writeFileSync } from 'node:fs';
+import { createRequire } from 'node:module';
 import { basename, dirname, join, relative, resolve } from 'node:path';
 import { fileURLToPath } from 'node:url';
 import { compareSemver } from './update-check.js';
@@ -24,6 +25,9 @@ const resourcesDir = resolveBundledResourcesDirFromPackageRoot(packageRoot);
 const bundledExtensionsDir = join(resourcesDir, 'extensions');
 const resourceVersionManifestName = 'managed-resources.json';
 const resourceFingerprintFileName = '.managed-resources-content-hash';
+const gsdBrowserSkillName = 'gsd-browser';
+const requireFromResourceLoader = createRequire(import.meta.url);
+const gsdBrowserSkillReferenceDirs = ['docs', 'scripts', 'gsd-browser-skill'];
 export { discoverExtensionEntryPaths } from './extension-discovery.js';
 export function getExtensionKey(entryPath, extensionsDir) {
     const relPath = relative(extensionsDir, entryPath);
@@ -153,6 +157,105 @@ function getCurrentResourceFingerprint() {
     }
     return computeResourceFingerprint();
 }
+function resolveGsdBrowserPackageSkillPath() {
+    try {
+        return requireFromResourceLoader.resolve('@opengsd/gsd-browser/SKILL.md');
+    }
+    catch {
+        return null;
+    }
+}
+export function collectGsdBrowserPackageSkillReferences(content) {
+    const refs = new Set();
+    const referencePattern = /`((?:\.\/)?(?:docs|scripts|gsd-browser-skill)\/[^`\s]+)`/g;
+    for (const match of content.matchAll(referencePattern)) {
+        const ref = normalizeGsdBrowserPackageSkillReference(match[1]);
+        if (ref)
+            refs.add(ref);
+    }
+    return [...refs].sort();
+}
+function normalizeGsdBrowserPackageSkillReference(ref) {
+    if (!ref)
+        return null;
+    const normalized = ref
+        .replace(/^\.\//, '')
+        .replace(/[),.;:]+$/, '');
+    if (normalized.includes('..') || normalized.startsWith('/'))
+        return null;
+    if (!gsdBrowserSkillReferenceDirs.some(dir => normalized === dir || normalized.startsWith(`${dir}/`))) {
+        return null;
+    }
+    return normalized;
+}
+function readGsdBrowserPackageSkillBundle(sourceSkillPath) {
+    const skillContent = readFileSync(sourceSkillPath);
+    const files = new Map([['SKILL.md', skillContent]]);
+    const supportRefs = collectGsdBrowserPackageSkillReferences(skillContent.toString('utf-8'));
+    const sourceDir = dirname(sourceSkillPath);
+    for (const relPath of supportRefs) {
+        const sourcePath = join(sourceDir, relPath);
+        if (existsSync(sourcePath)) {
+            files.set(relPath, readFileSync(sourcePath));
+        }
+    }
+    return files;
+}
+export function hasStaleGsdBrowserPackageSkill(skillsDir) {
+    const targetDir = join(skillsDir, gsdBrowserSkillName);
+    const sourceSkillPath = resolveGsdBrowserPackageSkillPath();
+    // Package unresolvable. syncGsdBrowserPackageSkill is a no-op in this case
+    // (it preserves any existing managed skill), so reporting "stale" here
+    // would drive a full resource resync on every launch that the sync cannot
+    // actually satisfy. Keep the existing managed skill in place until the
+    // package becomes resolvable again or the manifest fingerprint changes.
+    if (!sourceSkillPath)
+        return false;
+    try {
+        const sourceDir = dirname(sourceSkillPath);
+        const skillContent = readFileSync(sourceSkillPath, 'utf-8');
+        for (const [relPath, content] of readGsdBrowserPackageSkillBundle(sourceSkillPath)) {
+            const targetPath = join(targetDir, relPath);
+            if (!existsSync(targetPath) || !readFileSync(targetPath).equals(content)) {
+                return true;
+            }
+        }
+        // Only flag SKILL.md references when sync can actually correct them.
+        // The bundle loop above already covers "source has it, target missing".
+        // A reference the package itself does not ship (source missing) is only
+        // worth a resync if the target still has a stale copy that sync should
+        // clean up; missing on both sides is the steady state when upstream
+        // omits files referenced from SKILL.md and is not actionable here.
+        for (const relPath of collectGsdBrowserPackageSkillReferences(skillContent)) {
+            const sourcePath = join(sourceDir, relPath);
+            const targetPath = join(targetDir, relPath);
+            if (!existsSync(sourcePath) && existsSync(targetPath)) {
+                return true;
+            }
+        }
+        return false;
+    }
+    catch {
+        return true;
+    }
+}
+function syncGsdBrowserPackageSkill(skillsDir) {
+    const targetDir = join(skillsDir, gsdBrowserSkillName);
+    const sourceSkillPath = resolveGsdBrowserPackageSkillPath();
+    if (!sourceSkillPath)
+        return;
+    makeTreeWritable(targetDir);
+    rmSync(targetDir, { recursive: true, force: true });
+    for (const [relPath, content] of readGsdBrowserPackageSkillBundle(sourceSkillPath)) {
+        const targetPath = join(targetDir, relPath);
+        mkdirSync(dirname(targetPath), { recursive: true });
+        writeFileSync(targetPath, content);
+        if (relPath.startsWith('scripts/') && relPath.endsWith('.sh')) {
+            chmodSync(targetPath, 0o755);
+        }
+    }
+    makeTreeWritable(targetDir);
+}
 function collectFileEntries(dir, root, out) {
     if (!existsSync(dir))
         return;
@@ -557,18 +660,20 @@ function pruneRemovedBundledExtensions(manifest, agentDir) {
     removeFileIfStale('env-utils.js');
 }
 /**
- * Syncs all bundled resources to agentDir (~/.gsd/agent/) on every launch.
+ * Initializes managed resources under agentDir (~/.gsd/agent/).
  *
  * - extensions/ → ~/.gsd/agent/extensions/   (overwrite when version changes)
  * - shared/     → ~/.gsd/agent/shared/       (overwrite when version changes)
  * - agents/     → ~/.gsd/agent/agents/        (overwrite when version changes)
  * - skills/     → ~/.gsd/agent/skills/        (overwrite when version changes)
+ * - gsd-browser skill → ~/.gsd/agent/skills/gsd-browser/ from @opengsd/gsd-browser
  * - GSD-WORKFLOW.md → ~/.gsd/agent/GSD-WORKFLOW.md (fallback for env var miss)
  *
- * Skips the copy when the managed-resources.json version matches the current
- * GSD version, avoiding ~128ms of synchronous cpSync on every startup.
- * After `npm update -g @glittercowboy/gsd`, versions will differ and the
- * copy runs once to land the new resources.
+ * Skips the full copy only when the managed-resources.json version, content
+ * fingerprint, and package-owned gsd-browser skill all match the current
+ * install, avoiding ~128ms of synchronous cpSync on steady-state startup.
+ * After `npm update -g @opengsd/gsd-pi`, versions will differ and the copy
+ * runs once to land the new resources.
  *
  * Inspectable: `ls ~/.gsd/agent/extensions/`
  */
@@ -603,11 +708,13 @@ export function initResources(agentDir, skillsDir = join(agentDir, 'skills')) {
         const hasStaleExtensionFiles = hasStaleCompiledExtensionSiblings(extensionsDir, bundledExtensionsDir);
         const hasMissingSharedFiles = hasMissingBundledResourceFiles(join(agentDir, 'shared'), join(resourcesDir, 'shared'));
         const hasMissingSkillFiles = hasMissingBundledResourceFiles(skillsDir, join(resourcesDir, 'skills'));
+        const hasStaleGsdBrowserSkill = hasStaleGsdBrowserPackageSkill(skillsDir);
         if (manifest.contentHash &&
             manifest.contentHash === currentHash &&
             !hasStaleExtensionFiles &&
             !hasMissingSharedFiles &&
-            !hasMissingSkillFiles) {
+            !hasMissingSkillFiles &&
+            !hasStaleGsdBrowserSkill) {
             return;
         }
     }
@@ -616,6 +723,7 @@ export function initResources(agentDir, skillsDir = join(agentDir, 'skills')) {
     syncResourceDir(join(resourcesDir, 'shared'), join(agentDir, 'shared'));
     syncResourceDir(join(resourcesDir, 'agents'), join(agentDir, 'agents'));
     syncResourceDir(join(resourcesDir, 'skills'), skillsDir);
+    syncGsdBrowserPackageSkill(skillsDir);
     // Sync GSD-WORKFLOW.md to agentDir as a fallback for when GSD_WORKFLOW_PATH
     // env var is not set (e.g. fork/dev builds, alternative entry points).
     const workflowSrc = join(resourcesDir, 'GSD-WORKFLOW.md');

package/dist/resources/.managed-resources-content-hash

@@ -1 +1 @@
-eb84bf3c0ea71702
+dab8e4d2f1341f51

package/dist/resources/GSD-WORKFLOW.md

@@ -470,13 +470,13 @@ key_decisions: []
 
 **After a slice completes:**
 1. Write slice `S##-SUMMARY.md` (compresses all task summaries).
-2. Write slice `S##-UAT.md` — a non-blocking human test script derived from the slice's must-haves and demo sentence. The agent does NOT wait for UAT results.
+2. Write slice `S##-UAT.md` — a human test script derived from the slice's must-haves and demo sentence.
 3. Mark the slice checkbox in `M###-ROADMAP.md` as `[x]`.
 4. Update `STATE.md` with new position.
 5. Update milestone `M###-SUMMARY.md` with the completed slice's contributions.
-6. Continue to next slice immediately. The user tests the UAT whenever convenient.
-7. If the user reports UAT failures later, create fix tasks in the current or a new slice.
-8. If all slices done → milestone complete.
+6. Continue to next slice immediately. UAT can run after slice completion; automatic milestone closure requires each slice assessment to record `PASS`.
+7. If UAT is missing or non-PASS at milestone closeout, run `/gsd dispatch uat`, request a slice-specific UAT rerun when needed, or create remediation work with `/gsd dispatch reassess`.
+8. If all slices are done and UAT plus milestone validation pass → milestone complete.
 
 ---
 
@@ -605,6 +605,7 @@ Commit types: `feat`, `fix`, `test`, `refactor`, `docs`, `perf`, `chore`
 |---------|-----|
 | Bad task | Revert the task commit on the active branch |
 | Bad milestone squash | Revert the squash commit on the integration branch |
+| UAT missing or non-PASS at closeout | Run `/gsd dispatch uat`, request a slice-specific UAT rerun when needed, or create remediation work with `/gsd dispatch reassess` |
 | UAT failure after merge | Create follow-up fix tasks in the current or next milestone |
 
 ---

package/dist/resources/extensions/async-jobs/async-bash-tool.js

@@ -5,9 +5,9 @@
  * immediately. The LLM can continue working and check results later
  * with await_job.
  */
-import { getShellConfig, sanitizeCommand, DEFAULT_MAX_BYTES, DEFAULT_MAX_LINES, } from "@gsd/pi-coding-agent";
+import { getShellConfig, sanitizeCommand, killProcessTree, SIGKILL_GRACE_MS, HARD_DEADLINE_MS, DEFAULT_MAX_BYTES, DEFAULT_MAX_LINES, } from "@gsd/pi-coding-agent";
 import { Type } from "@sinclair/typebox";
-import { spawn, spawnSync } from "node:child_process";
+import { spawn } from "node:child_process";
 import { createWriteStream } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
@@ -22,37 +22,6 @@ function getTempFilePath() {
     const id = randomBytes(8).toString("hex");
     return join(tmpdir(), `pi-async-bash-${id}.log`);
 }
-/**
- * Kill a process and its children (cross-platform).
- * Uses process group kill on Unix; taskkill /F /T on Windows.
- */
-function killTree(pid) {
-    if (process.platform === "win32") {
-        try {
-            spawnSync("taskkill", ["/F", "/T", "/PID", String(pid)], {
-                timeout: 5_000,
-                stdio: "ignore",
-            });
-        }
-        catch {
-            try {
-                process.kill(pid, "SIGTERM");
-            }
-            catch { /* already exited */ }
-        }
-    }
-    else {
-        try {
-            process.kill(-pid, "SIGTERM");
-        }
-        catch {
-            try {
-                process.kill(pid, "SIGTERM");
-            }
-            catch { /* already exited */ }
-        }
-    }
-}
 export function createAsyncBashTool(getManager, getCwd) {
     return {
         name: "async_bash",
@@ -62,7 +31,7 @@ export function createAsyncBashTool(getManager, getCwd) {
             `Output is truncated to the last ${DEFAULT_MAX_LINES} lines or ${DEFAULT_MAX_BYTES / 1024}KB.`,
         promptSnippet: "Run a bash command in the background, returning a job ID immediately.",
         promptGuidelines: [
-            "Use async_bash for commands that take more than a few seconds (builds, tests, installs, large git operations).",
+            "Use async_bash for long-running builds, tests, installs, or operations that should run in the background so you can continue other work (the job ID lets you await_job later). Sync bash is uncapped — use async_bash when you want non-blocking behavior, not because of a timeout concern.",
             "After starting async jobs, continue with other work and use await_job when you need the results.",
             "await_job has a configurable timeout (default 120s) to prevent indefinite blocking — if it times out, jobs keep running and you can check again later.",
             "For long-running processes (SSH, deploys, training) that may take minutes+, prefer async_bash with periodic await_job polling over a single long await.",
@@ -122,36 +91,25 @@ function executeBashInBackground(command, cwd, signal, timeout) {
         });
         let timedOut = false;
         let timeoutHandle;
-        let sigkillHandle;
         let hardDeadlineHandle;
-        /** Grace period (ms) between SIGTERM and SIGKILL. */
-        const SIGKILL_GRACE_MS = 5_000;
-        /** Hard deadline (ms) after SIGKILL to force-resolve the promise. */
-        const HARD_DEADLINE_MS = 3_000;
         if (timeout !== undefined && timeout > 0) {
             timeoutHandle = setTimeout(() => {
                 timedOut = true;
+                // killProcessTree owns the SIGTERM -> grace -> SIGKILL escalation, so a
+                // SIGTERM-immune child is actually force-killed rather than left running.
                 if (child.pid)
-                    killTree(child.pid);
-                // If the process ignores SIGTERM, escalate to SIGKILL
-                sigkillHandle = setTimeout(() => {
-                    if (child.pid) {
-                        // killTree already uses taskkill /F /T on Windows
-                        killTree(child.pid);
-                    }
-                    // Hard deadline: if even SIGKILL doesn't trigger 'close',
-                    // force-resolve so the job doesn't hang forever (#2186).
-                    hardDeadlineHandle = setTimeout(() => {
-                        const output = Buffer.concat(chunks).toString("utf-8");
-                        safeResolve(output
-                            ? `${output}\n\nCommand timed out after ${timeout} seconds (force-killed)`
-                            : `Command timed out after ${timeout} seconds (force-killed)`);
-                    }, HARD_DEADLINE_MS);
-                    if (typeof hardDeadlineHandle === "object" && "unref" in hardDeadlineHandle)
-                        hardDeadlineHandle.unref();
-                }, SIGKILL_GRACE_MS);
-                if (typeof sigkillHandle === "object" && "unref" in sigkillHandle)
-                    sigkillHandle.unref();
+                    killProcessTree(child.pid);
+                // Hard deadline: a D-state (uninterruptible-I/O) child never emits 'close'
+                // even after SIGKILL, so force-resolve the promise rather than hang (#2186).
+                // Fires after the full grace window so SIGKILL has had its chance first.
+                hardDeadlineHandle = setTimeout(() => {
+                    const output = Buffer.concat(chunks).toString("utf-8");
+                    safeResolve(output
+                        ? `${output}\n\nCommand timed out after ${timeout} seconds (force-killed)`
+                        : `Command timed out after ${timeout} seconds (force-killed)`);
+                }, SIGKILL_GRACE_MS + HARD_DEADLINE_MS);
+                if (typeof hardDeadlineHandle === "object" && "unref" in hardDeadlineHandle)
+                    hardDeadlineHandle.unref();
             }, timeout * 1000);
         }
         const chunks = [];
@@ -182,7 +140,19 @@ function executeBashInBackground(command, cwd, signal, timeout) {
             child.stderr.on("data", onData);
         const onAbort = () => {
             if (child.pid)
-                killTree(child.pid);
+                killProcessTree(child.pid);
+            // Arm the same hard-deadline force-resolve the timeout path uses, so a
+            // cancelled D-state child (never emits 'close' even after SIGKILL) can't
+            // hang the job promise forever. safeResolve is idempotent, so the real
+            // 'close' still wins if the child does exit.
+            if (!hardDeadlineHandle) {
+                hardDeadlineHandle = setTimeout(() => {
+                    const output = Buffer.concat(chunks).toString("utf-8");
+                    safeResolve(output ? `${output}\n\nCommand aborted (force-killed)` : "Command aborted (force-killed)");
+                }, SIGKILL_GRACE_MS + HARD_DEADLINE_MS);
+                if (typeof hardDeadlineHandle === "object" && "unref" in hardDeadlineHandle)
+                    hardDeadlineHandle.unref();
+            }
         };
         if (signal.aborted) {
             onAbort();
@@ -193,8 +163,6 @@ function executeBashInBackground(command, cwd, signal, timeout) {
         child.on("error", (err) => {
             if (timeoutHandle)
                 clearTimeout(timeoutHandle);
-            if (sigkillHandle)
-                clearTimeout(sigkillHandle);
             if (hardDeadlineHandle)
                 clearTimeout(hardDeadlineHandle);
             signal.removeEventListener("abort", onAbort);
@@ -203,8 +171,6 @@ function executeBashInBackground(command, cwd, signal, timeout) {
         child.on("close", (code) => {
             if (timeoutHandle)
                 clearTimeout(timeoutHandle);
-            if (sigkillHandle)
-                clearTimeout(sigkillHandle);
             if (hardDeadlineHandle)
                 clearTimeout(hardDeadlineHandle);
             signal.removeEventListener("abort", onAbort);

package/dist/resources/extensions/async-jobs/await-tool.js

@@ -21,7 +21,7 @@ export function createAwaitTool(getManager) {
         label: "Await Background Job",
         description: "Wait for background jobs to complete. Provide specific job IDs or omit to wait for the next job that finishes. Returns results of completed jobs.",
         parameters: schema,
-        async execute(_toolCallId, params, _signal, _onUpdate, _ctx) {
+        async execute(_toolCallId, params, signal, _onUpdate, _ctx) {
             const manager = getManager();
             const { jobs: jobIds, timeout } = params;
             const timeoutMs = ((timeout ?? DEFAULT_TIMEOUT_SECONDS) * 1000);
@@ -54,37 +54,75 @@ export function createAwaitTool(getManager) {
                     };
                 }
             }
-            // Suppress follow-up notifications for all watched jobs upfront.
+            // If all watched jobs are already done, suppress follow-up and return immediately.
             // suppressFollowUp() cancels the pending delivery timer (if any), which
             // handles both the within-turn case (job completes while we await) and
             // the cross-turn case (job already completed before await_job was called).
             // Previously this only set j.awaited = true, which missed the cross-turn
             // case because the queueMicrotask had already fired (#3787).
-            for (const j of watched)
-                manager.suppressFollowUp(j.id);
-            // If all watched jobs are already done, return immediately
             const running = watched.filter((j) => j.status === "running");
             if (running.length === 0) {
-                const result = formatResults(watched);
-                return { content: [{ type: "text", text: result }], details: undefined };
+                for (const j of watched)
+                    manager.suppressFollowUp(j.id);
+                return { content: [{ type: "text", text: renderCompleted(watched) }], details: undefined };
             }
-            // Wait for at least one to complete, or timeout
+            // Wait for at least one to complete, timeout, or abort signal
             const TIMEOUT_SENTINEL = Symbol("timeout");
+            const ABORT_SENTINEL = Symbol("abort");
+            // The race timer and abort listener are explicitly torn down once the race
+            // settles (below) so a completion- or timeout-won race never leaks a pending
+            // timer or a lingering abort listener — the listener holds a closure over the
+            // race resolver, so { once: true } alone (which only detaches on fire) is not
+            // enough when ESC never happens.
+            let raceTimer;
+            let abortListener;
             const timeoutPromise = new Promise((resolve) => {
-                const timer = setTimeout(() => resolve(TIMEOUT_SENTINEL), timeoutMs);
+                raceTimer = setTimeout(() => resolve(TIMEOUT_SENTINEL), timeoutMs);
                 // Allow the process to exit even if the timer is pending
-                if (typeof timer === "object" && "unref" in timer)
-                    timer.unref();
+                if (typeof raceTimer === "object" && "unref" in raceTimer)
+                    raceTimer.unref();
+            });
+            const abortPromise = new Promise((resolve) => {
+                if (!signal || signal.aborted) {
+                    resolve(ABORT_SENTINEL);
+                }
+                else {
+                    abortListener = () => resolve(ABORT_SENTINEL);
+                    signal.addEventListener("abort", abortListener, { once: true });
+                }
             });
             const raceResult = await Promise.race([
                 Promise.race(running.map((j) => j.promise)).then(() => "completed"),
                 timeoutPromise,
+                abortPromise,
             ]);
+            // Tear down race resources now that a winner is decided.
+            if (raceTimer)
+                clearTimeout(raceTimer);
+            if (abortListener && signal)
+                signal.removeEventListener("abort", abortListener);
+            const aborted = raceResult === ABORT_SENTINEL;
             const timedOut = raceResult === TIMEOUT_SENTINEL;
             // Collect all completed results (more may have finished while waiting)
             const completed = watched.filter((j) => j.status !== "running");
             const stillRunning = watched.filter((j) => j.status === "running");
-            let result = formatResults(completed);
+            // Suppress follow-up ONLY for completed jobs — leave stillRunning unsuppressed
+            // so deliverResult/onJobComplete can resurface their results later.
+            for (const j of completed)
+                manager.suppressFollowUp(j.id);
+            if (aborted) {
+                // ESC ended the wait, not the jobs: still-running jobs keep going and
+                // resurface via onJobComplete (they were deliberately not suppressed above).
+                const runningDesc = stillRunning.map((j) => `${j.id} (${j.label})`).join(", ");
+                const interrupt = stillRunning.length > 0
+                    ? `Wait interrupted. Still running: ${runningDesc} — results will surface when complete.`
+                    : "Wait interrupted.";
+                const text = completed.length > 0
+                    ? `${renderCompleted(completed)}\n\n${interrupt}`
+                    : interrupt;
+                return { content: [{ type: "text", text }], details: undefined };
+            }
+            let result = renderCompleted(completed);
             if (stillRunning.length > 0) {
                 result += `\n\n**Still running:** ${stillRunning.map((j) => `${j.id} (${j.label})`).join(", ")}`;
             }
@@ -97,6 +135,36 @@ export function createAwaitTool(getManager) {
         },
     };
 }
+/**
+ * Render completed jobs for the await_job result, de-duplicating against
+ * follow-ups that have already been delivered to context.
+ *
+ * A job's follow-up fires ~immediately (setTimeout(0)) once it settles, so when
+ * await_job runs in a LATER turn the result is already in context. Reprinting it
+ * inline produces the same output twice. Jobs already `delivered` are therefore
+ * acknowledged on a single line instead of having their full output reprinted;
+ * not-yet-delivered jobs (the within-turn case, where suppressFollowUp won the
+ * race) are rendered in full as before.
+ */
+function renderCompleted(jobs) {
+    if (jobs.length === 0)
+        return "No completed jobs.";
+    const fresh = jobs.filter((j) => !j.delivered);
+    const alreadyDelivered = jobs.filter((j) => j.delivered);
+    const sections = [];
+    if (fresh.length > 0)
+        sections.push(formatResults(fresh));
+    if (alreadyDelivered.length > 0) {
+        const names = alreadyDelivered
+            .map((j) => `${j.id} (${j.label})`)
+            .join(", ");
+        const sentence = alreadyDelivered.length === 1
+            ? `This job already finished and its result was shown above when it completed, so there is nothing new to report: ${names}.`
+            : `These jobs already finished and their results were shown above when they completed, so there is nothing new to report: ${names}.`;
+        sections.push(sentence);
+    }
+    return sections.join("\n\n");
+}
 function formatResults(jobs) {
     if (jobs.length === 0)
         return "No completed jobs.";

package/dist/resources/extensions/async-jobs/index.js

@@ -98,9 +98,74 @@ export default function AsyncJobs(pi) {
                 });
                 return;
             }
+            const ctx = _ctx;
             const running = manager.getRunningJobs();
             const recent = manager.getRecentJobs(10);
             const completed = recent.filter((j) => j.status !== "running");
+            // Interactive kill-picker when there are running jobs and a UI is available
+            if (running.length > 0 && ctx.hasUI) {
+                // Kill-picker loop: each iteration shows live running jobs.
+                // Step 1 picks a job (neutral label — selecting does NOT cancel yet);
+                // step 2 confirms before the destructive cancel. Labels deliberately
+                // omit a live elapsed time: ctx.ui.select renders the option strings
+                // once and never refreshes them, so a "(24s)" baked into the label
+                // would freeze and mislead. Accurate elapsed times are shown in the
+                // post-picker summary (rebuilt fresh) instead.
+                const DONE = "Close";
+                while (true) {
+                    const liveJobs = manager.getRunningJobs();
+                    if (liveJobs.length === 0)
+                        break;
+                    // Map display label -> job id so we never parse ids back out of
+                    // free-form label text.
+                    const labelToId = new Map();
+                    for (const j of liveJobs) {
+                        labelToId.set(`${j.id} — ${j.label} (running)`, j.id);
+                    }
+                    const options = [...labelToId.keys(), DONE];
+                    const choice = await ctx.ui.select("Background jobs — pick one to cancel, Escape to close", options);
+                    // ESC returns undefined; headless may return string[]; DONE closes
+                    if (!choice || typeof choice !== "string" || choice === DONE)
+                        break;
+                    const id = labelToId.get(choice);
+                    if (!id)
+                        break;
+                    const job = liveJobs.find((j) => j.id === id);
+                    const confirmed = await ctx.ui.confirm("Cancel background job?", `This will stop ${id}${job ? ` — ${job.label}` : ""}. Other jobs keep running.`);
+                    if (!confirmed)
+                        continue;
+                    const r = manager.cancel(id);
+                    ctx.ui.notify(r === "cancelled" ? `Job ${id} cancelled.` : `Job ${id}: ${r}`, r === "cancelled" ? "success" : "warning");
+                }
+                // After picker, send a summary of any still-running jobs
+                const stillRunning = manager.getRunningJobs();
+                const summaryLines = ["## Background Jobs"];
+                if (stillRunning.length > 0) {
+                    summaryLines.push("", "### Running");
+                    for (const job of stillRunning) {
+                        const elapsed = ((Date.now() - job.startTime) / 1000).toFixed(0);
+                        summaryLines.push(`- **${job.id}** — ${job.label} (${elapsed}s)`);
+                    }
+                }
+                const recentCompleted = manager.getRecentJobs(10).filter((j) => j.status !== "running");
+                if (recentCompleted.length > 0) {
+                    summaryLines.push("", "### Recent");
+                    for (const job of recentCompleted) {
+                        const elapsed = ((Date.now() - job.startTime) / 1000).toFixed(1);
+                        summaryLines.push(`- **${job.id}** — ${job.label} (${job.status}, ${elapsed}s)`);
+                    }
+                }
+                if (stillRunning.length === 0 && recentCompleted.length === 0) {
+                    summaryLines.push("", "No background jobs.");
+                }
+                pi.sendMessage({
+                    customType: "async_jobs_list",
+                    content: summaryLines.join("\n"),
+                    display: true,
+                });
+                return;
+            }
+            // Text-only display: headless/RPC mode, or no running jobs
             const lines = ["## Background Jobs"];
             if (running.length === 0 && completed.length === 0) {
                 lines.push("", "No background jobs.");