pikiloom 0.4.71 → 0.4.73

package/packages/kernel/dist/drivers/claude.js

@@ -1,17 +1,24 @@
 import { spawn } from 'node:child_process';
 import { readFileSync } from 'node:fs';
+import { ClaudeWarmPool } from './claude-pool.js';
 import { claudeTranscriptTailAnchor, discoverClaudeNativeSessions } from './native.js';
 import { attachedFileNote, contextPercent, createLineBuffer, imageMimeForFile, parseJsonLine, sigterm, wireAbort } from './shared.js';
-// Real driver: shells the local `claude` CLI in stream-json mode and normalizes its
-// events into kernel DriverEvents. Faithful to pikiloom's claude.ts event shapes
-// (system / stream_event{message_start,content_block_delta,message_delta} / assistant / result),
-// but fully self-contained. Proves "下层 Claude 不变".
 export class ClaudeDriver {
     bin;
     id = 'claude';
-    capabilities = { steer: true, interact: false, resume: true, tui: true, fork: true };
-    constructor(bin = 'claude') {
+    capabilities = { steer: true, interact: false, resume: true, tui: true, fork: true, rewind: true };
+    pool;
+    constructor(bin = 'claude', opts = {}) {
         this.bin = bin;
+        this.pool = opts.warmPool ? new ClaudeWarmPool() : null;
+    }
+    /** Destroy every parked warm process. Long-lived hosts call this on shutdown. */
+    dispose() {
+        this.pool?.dispose();
+    }
+    /** Parked warm processes right now (tests + telemetry). */
+    warmPoolSize() {
+        return this.pool?.size() ?? 0;
     }
     // Interactive Claude Code TUI (no -p): the kernel spawns this in a PTY and passes
     // the terminal through. Model/resume/BYOK-env come from the kernel's resolution.
@@ -43,7 +50,7 @@ export class ClaudeDriver {
         if (input.effort)
             args.push('--effort', input.effort === 'ultra' ? 'max' : input.effort); // request extended thinking (ultra is a display-only alias for max)
         if (input.sessionId)
-            args.push(...claudeResumeArgs(input.sessionId, input.fork));
+            args.push(...claudeResumeArgs(input.sessionId, input.fork, input.rewind));
         if (input.systemPrompt)
             args.push('--append-system-prompt', input.systemPrompt);
         if (input.mcpConfigPath)
@@ -54,6 +61,17 @@ export class ClaudeDriver {
             args.push('--replay-user-messages'); // parity: mid-turn steer
         if (input.extraArgs?.length)
             args.push(...input.extraArgs);
+        // Warm reuse: a rewind rebranches the session's transcript, so a parked process's
+        // in-memory conversation is stale — destroy it. A fork never touches the parent's
+        // transcript (the parked parent stays valid); the fork turn itself must cold-spawn
+        // for its --fork-session flags. Only a plain continuation may reclaim a process, and
+        // only when the spawn fingerprint still matches what a cold spawn would use now.
+        const fingerprint = claudeProcessFingerprint(this.bin, input);
+        if (input.rewind && input.sessionId)
+            this.pool?.evictSession(input.sessionId);
+        const pooled = (!input.fork && !input.rewind && input.sessionId && this.pool)
+            ? this.pool.take(input.sessionId, fingerprint)
+            : null;
         const state = {
             text: '', reasoning: '', streamedText: false, streamedReasoning: false,
             // Dangling-tool-loop tracking: sawToolResult flips on the first tool_result; textSinceToolResult
@@ -82,6 +100,8 @@ export class ClaudeDriver {
         };
         return new Promise((resolve) => {
             let child;
+            let transport = 'cold';
+            let promptDelivered = false;
             let settled = false;
             // One-shot guard for the truncated-turn recovery injection (see the result handler).
             let truncatedRecoveryAttempted = false;
@@ -101,6 +121,8 @@ export class ClaudeDriver {
             const usageOf = () => this.usage(state);
             const unref = (tm) => { if (tm && typeof tm.unref === 'function')
                 tm.unref(); };
+            let disposeAbort = null;
+            let detachTurnListeners = () => { };
             const clearHoldCap = () => { if (holdCapTimer) {
                 clearTimeout(holdCapTimer);
                 holdCapTimer = null;
@@ -118,18 +140,29 @@ export class ClaudeDriver {
             // kill=false only ends stdin and lets Claude shut down on its own, so any still-running
             // detached background work survives a clean exit (a hard kill mid-flight is exactly what
             // tore the background — and the wake-up — down before). A leak-guard SIGTERM is the backstop.
-            const finish = (r, kill = true) => {
+            const finish = (r, kill = true, park = false) => {
                 if (settled)
                     return;
                 settled = true;
                 clearHoldCap();
                 clearQuiet();
                 clearModelStall();
+                disposeAbort?.();
+                // Park instead of kill: only a CLEAN settle qualifies (ok, no error, no abort, the
+                // process still healthy, session known) — every other exit keeps today's semantics,
+                // so pooling can never leak a wedged or errored process.
+                if (park && this.pool && state.sessionId && r.ok && !r.error && !ctx.signal.aborted
+                    && child && !child.killed && child.exitCode == null) {
+                    detachTurnListeners();
+                    this.pool.put(state.sessionId, fingerprint, child);
+                    resolve({ ...r, transport });
+                    return;
+                }
                 try {
                     child?.stdin?.end();
                 }
                 catch { /* ignore */ }
-                if (!child.killed && child.exitCode == null) {
+                if (child && !child.killed && child.exitCode == null) {
                     if (kill)
                         sigterm(child);
                     else {
@@ -137,12 +170,12 @@ export class ClaudeDriver {
                         unref(guard);
                     }
                 }
-                resolve(r);
+                resolve({ ...r, transport });
             };
             const settleResult = (opts = {}) => finish({
                 ok: opts.ok ?? !state.error, text: state.text, reasoning: state.reasoning || undefined,
                 error: state.error, stopReason: opts.stopReason ?? state.stopReason, sessionId: state.sessionId, anchor: state.anchor, usage: usageOf(),
-            }, opts.kill ?? true);
+            }, opts.kill ?? true, opts.park ?? false);
             // Cap while holding for a still-running background task (stopReason marks it as
             // "still running in the background" so the terminal presentation reads right). Idempotent —
             // the countdown is absolute from the first arm. Sub-agent-backed holds use the longer
@@ -199,14 +232,32 @@ export class ClaudeDriver {
                 }, claudeModelStallMs(input.effort));
                 unref(modelStallTimer);
             };
-            try {
-                child = spawn(this.bin, args, { cwd: input.workdir, env: { ...process.env, ...(input.env || {}) }, stdio: ['pipe', 'pipe', 'pipe'] });
+            // Acquire the process: a reclaimed warm process gets the prompt as its acquisition
+            // probe — a dead pipe throws synchronously here and the turn transparently falls
+            // back to a cold spawn (whose --resume flags are already in `args`).
+            let acquired = null;
+            if (pooled) {
+                try {
+                    pooled.stdin.write(claudeUserMessage(input.prompt, input.attachments) + '\n');
+                    acquired = pooled;
+                    transport = 'warm';
+                    promptDelivered = true;
+                }
+                catch {
+                    sigterm(pooled);
+                }
             }
-            catch (err) {
-                finish({ ok: false, text: '', error: `spawn failed: ${err?.message || err}`, stopReason: 'error' });
-                return;
+            if (!acquired) {
+                try {
+                    acquired = spawn(this.bin, args, { cwd: input.workdir, env: { ...process.env, ...(input.env || {}) }, stdio: ['pipe', 'pipe', 'pipe'] });
+                }
+                catch (err) {
+                    finish({ ok: false, text: '', error: `spawn failed: ${err?.message || err}`, stopReason: 'error' });
+                    return;
+                }
             }
-            wireAbort(ctx.signal, () => sigterm(child));
+            child = acquired;
+            disposeAbort = wireAbort(ctx.signal, () => sigterm(child));
             if (steerable) {
                 ctx.registerSteer(async (prompt, attachments) => {
                     try {
@@ -220,7 +271,7 @@ export class ClaudeDriver {
             }
             const nextLines = createLineBuffer();
             let stderr = '';
-            child.stdout.on('data', (chunk) => {
+            const onStdout = (chunk) => {
                 if (settled)
                     return; // ignore the process's post-settle shutdown chatter
                 for (const line of nextLines(chunk)) {
@@ -299,8 +350,10 @@ export class ClaudeDriver {
                             // DELIVERED live but never lands in the transcript, so the next re-render erases
                             // it (the "对话结束之后就被吞了" shape, caught by the turn audit). Ending stdin
                             // lets the CLI finish writing and exit on its own; the leak-guard SIGTERM is the
-                            // backstop.
-                            settleResult({ kill: false });
+                            // backstop. park: this clean settle is the ONE warm-pool-eligible exit — the
+                            // process (stdin open, transcript flushing in its own time) is parked for the
+                            // session's next turn instead of being shut down; finish() re-checks health.
+                            settleResult({ kill: false, park: true });
                             return;
                         }
                         if (decision === 'hold') {
@@ -332,10 +385,10 @@ export class ClaudeDriver {
                     if (ev.type === 'user' && pending === 0 && claudeUserEventHasToolResult(ev))
                         armModelStall();
                 }
-            });
-            child.stderr.on('data', (c) => { stderr += c.toString('utf8'); });
-            child.on('error', (err) => finish({ ok: false, text: state.text, error: `claude spawn error: ${err.message}`, stopReason: 'error' }));
-            child.on('close', (code) => {
+            };
+            const onStderr = (c) => { stderr += c.toString('utf8'); };
+            const onError = (err) => finish({ ok: false, text: state.text, error: `claude spawn error: ${err.message}`, stopReason: 'error' });
+            const onClose = (code) => {
                 if (settled)
                     return;
                 if (ctx.signal.aborted) {
@@ -349,13 +402,27 @@ export class ClaudeDriver {
                 // dangling check must win over it.)
                 const stopReason = (ok && claudeTurnEndedDangling(state)) ? 'truncated' : state.stopReason;
                 finish({ ok, text: state.text, reasoning: state.reasoning || undefined, error: state.error || (ok ? null : `claude exited ${code}${stderr ? `: ${stderr.slice(0, 300)}` : ''}`), stopReason, sessionId: state.sessionId, anchor: state.anchor, usage: usageOf() }, false);
-            });
-            try {
-                // Send the prompt as a stream-json user message and keep stdin OPEN (do not end it here):
-                // closing it makes Claude exit at the first `result`, before any background task finishes.
-                child.stdin.write(claudeUserMessage(input.prompt, input.attachments) + '\n');
+            };
+            // Parking hands the process to the pool — these turn-scoped listeners must not
+            // outlive the turn (the pool installs its own drain + close bookkeeping).
+            detachTurnListeners = () => {
+                child.stdout?.off('data', onStdout);
+                child.stderr?.off('data', onStderr);
+                child.off('error', onError);
+                child.off('close', onClose);
+            };
+            child.stdout.on('data', onStdout);
+            child.stderr.on('data', onStderr);
+            child.on('error', onError);
+            child.on('close', onClose);
+            if (!promptDelivered) {
+                try {
+                    // Send the prompt as a stream-json user message and keep stdin OPEN (do not end it here):
+                    // closing it makes Claude exit at the first `result`, before any background task finishes.
+                    child.stdin.write(claudeUserMessage(input.prompt, input.attachments) + '\n');
+                }
+                catch { /* ignore */ }
             }
-            catch { /* ignore */ }
         });
     }
     usage(s) {
@@ -367,16 +434,53 @@ export class ClaudeDriver {
         return claudeTranscriptTailAnchor(opts.workdir, opts.sessionId);
     }
 }
+// The spawn-time facts that must still match for a parked warm process to serve a
+// continuation turn as if it were a fresh cold `--resume` — any drift (model switch,
+// effort change, new MCP config, different BYOK env, …) destroys the parked process so
+// the new configuration actually applies. `sessionId` is the pool key, not fingerprint
+// material, and `systemPrompt` is deliberately EXCLUDED: it is a first-turn-only input
+// (the parked process already carries it applied; a cold --resume would not re-send it
+// either), so including it would make every continuation miss the pool. Pure + exported
+// for hermetic testing.
+export function claudeProcessFingerprint(bin, input) {
+    return JSON.stringify([
+        bin, input.workdir, input.model ?? null, input.effort ?? null,
+        input.mcpConfigPath ?? null, input.permissionMode ?? null, !!input.steerable,
+        fingerprintExtraArgs(input.extraArgs),
+        Object.entries(input.env ?? {}).sort(([a], [b]) => (a < b ? -1 : a > b ? 1 : 0)),
+    ]);
+}
+// `--session-id <uuid>` pins a FRESH session's native id and is contributed on the first
+// turn only (continuations resume instead) — it names the session, it doesn't configure
+// the process. Like sessionId itself it must not be fingerprint material, or every
+// continuation of a session born with it would miss the pool forever.
+function fingerprintExtraArgs(extraArgs) {
+    const args = extraArgs ?? [];
+    const out = [];
+    for (let i = 0; i < args.length; i++) {
+        if (args[i] === '--session-id') {
+            i++;
+            continue;
+        }
+        out.push(args[i]);
+    }
+    return out;
+}
 // The --resume flag family for one turn: plain resume appends to the session; a fork branches
 // a NEW session off it (--fork-session), optionally cut at an inclusive keep-boundary record
-// uuid (--resume-session-at). Pure so tests can pin the arg contract without spawning.
-export function claudeResumeArgs(sessionId, fork) {
+// uuid (--resume-session-at); a rewind cuts at that boundary WITHOUT --fork-session, so claude
+// rebranches the SAME session in place (the transcript is a parentUuid tree — the dropped tip
+// becomes a dead sibling and leaves the active path). Pure so tests pin the arg contract.
+export function claudeResumeArgs(sessionId, fork, rewind) {
     const args = ['--resume', sessionId];
     if (fork) {
         args.push('--fork-session');
         if (fork.anchor)
             args.push('--resume-session-at', fork.anchor);
     }
+    else if (rewind?.anchor) {
+        args.push('--resume-session-at', rewind.anchor);
+    }
     return args;
 }
 function claudeUsageOf(s) {
@@ -468,6 +572,15 @@ export function handleClaudeEvent(ev, s, emit) {
         }
         s.model = ev.model ?? s.model;
         s.contextWindow = claudeEffectiveContextWindow(claudeContextWindowFromModel(s.model)) ?? s.contextWindow;
+        // Claude compacted the running context (subtype `compact_boundary`): `trigger` is
+        // `auto` (the context filled up) or `manual` (a `/compact` command). Surface it live
+        // so a terminal can show a "compacting" affordance; the compacted summary itself lands
+        // in the native transcript and settles into a divider separately.
+        if (ev.subtype === 'compact_boundary') {
+            const meta = (ev.compact_metadata ?? {});
+            emit({ type: 'compaction', trigger: meta.trigger === 'manual' ? 'manual' : 'auto', atTokens: typeof meta.pre_tokens === 'number' ? meta.pre_tokens : null });
+            return;
+        }
         // Live thinking progress (system/thinking_tokens, ~every 1.4s of sustained thinking): during
         // extended thinking a subscription account streams no plaintext (signature_delta only) and no
         // usage until the message settles, so without projecting these the terminal shows a dead

package/packages/kernel/dist/drivers/codex.d.ts

@@ -19,8 +19,15 @@ export declare function captureCodexAgentMessage(item: any, s: CodexContentState
 export declare function captureCodexReasoning(text: string, s: CodexContentState, emit: (e: DriverEvent) => void): void;
 export declare function codexFinalText(s: CodexContentState): string;
 export declare function codexFinalReasoning(s: CodexContentState): string;
+export interface CodexLivenessOptions {
+    /** Silence after an accepted steer before the turn is healed (interrupt + replay). */
+    steerStallMs?: number;
+    /** Idle silence (no running tool, no pending HITL request) before the turn is force-closed. */
+    turnStallMs?: number;
+}
 export declare class CodexDriver implements AgentDriver {
     private readonly bin;
+    private readonly liveness;
     readonly id = "codex";
     readonly capabilities: {
         steer: boolean;
@@ -29,7 +36,7 @@ export declare class CodexDriver implements AgentDriver {
         tui: boolean;
         fork: boolean;
     };
-    constructor(bin?: string);
+    constructor(bin?: string, liveness?: CodexLivenessOptions);
     run(input: AgentTurnInput, ctx: DriverContext): Promise<DriverResult>;
     tui(input: TuiInput): TuiSpec;
     listNativeSessions(opts: {

package/packages/kernel/dist/drivers/codex.js

@@ -234,12 +234,53 @@ export function codexFinalText(s) {
 export function codexFinalReasoning(s) {
     return s.reasoning.trim() ? s.reasoning : s.thinkParts.join('\n\n');
 }
+// ── Turn liveness (steer race + silent-stall recovery) ──────────────────────────
+// codex app-server has a turn-boundary race: a `turn/steer` that lands in the instant an
+// item completes is ACCEPTED (recorded into the rollout) but never dispatched — the agent
+// loop goes idle, no further notifications arrive, and `turn/completed` never fires, so the
+// turn hangs forever while the process sits at 0% CPU (observed on codex-cli 0.144.x;
+// same family as openai/codex#15714 / #23807). Two defenses below:
+//
+// 1. A successful steer is treated as ACCEPTED, NOT CONSUMED: it stays pending until a
+//    progress notification proves the loop picked it up. If the turn instead goes silent
+//    (or completes without consuming it), the driver heals in place — interrupt the wedged
+//    turn and restart it with the same input. The steered text is already in the thread
+//    history, so the worst false-positive cost is one duplicated user message; the turn
+//    keeps running under the same run()/task, invisible to upper layers.
+// 2. A generic silence backstop: a turn with no notifications for a long stretch while
+//    nothing is visibly in flight (no running tool call, no server->client request awaiting
+//    a human) is declared stalled and force-closed, so the session ends as a visible error
+//    instead of spinning forever.
+// Thresholds are calibrated against measured rollouts: with reasoning summaries off, a
+// HEALTHY turn shows fully silent thinking stretches of up to ~320s (clustered just above
+// codex core's own 300s stream watchdog, whose retry produces the next event). 600s has
+// never been reached intra-turn in days of observed sessions, so a heal at that point is
+// near-certainly a real wedge — and even a misfire only costs one duplicated user message
+// plus the in-flight thinking (completed items live in the thread history).
+const CODEX_STEER_STALL_MS = 600_000;
+const CODEX_TURN_STALL_MS = 900_000;
+// Notifications that prove the agent loop is making forward progress AFTER a steer.
+// item/completed and tokenUsage are deliberately excluded — they can be the trailing edge
+// of work that finished before the steer landed (the exact race signature). Progress that
+// arrives within CODEX_STEER_CONSUMED_MS of the steer only refreshes the pending marker's
+// clock rather than clearing it: the app-server can flush pre-acceptance stream output in
+// the same write as the steer response, so near-simultaneous events are not proof the
+// loop survived the injection. Progress beyond that window is.
+const CODEX_STEER_CONSUMED_MS = 2_000;
+const CODEX_PROGRESS_METHODS = new Set([
+    'turn/started', 'item/started', 'item/agentMessage/delta',
+    'item/reasoning/textDelta', 'item/reasoning/summaryTextDelta',
+    'item/commandExecution/outputDelta', 'item/fileChange/outputDelta',
+    'item/fileChange/patchUpdated', 'turn/plan/updated', 'rawResponseItem/completed',
+]);
 export class CodexDriver {
     bin;
+    liveness;
     id = 'codex';
     capabilities = { steer: true, interact: false, resume: true, tui: true, fork: true };
-    constructor(bin = 'codex') {
+    constructor(bin = 'codex', liveness = {}) {
         this.bin = bin;
+        this.liveness = liveness;
     }
     async run(input, ctx) {
         // BYOK provider routing arrives as `-c key=value` overrides; pass them through so the
@@ -261,6 +302,15 @@ export class CodexDriver {
         const deltaItems = new Set();
         let lastTextItemId = null;
         let steerRegistered = false;
+        // Liveness state (see the CODEX_STEER_STALL_MS block comment above).
+        const steerStallMs = this.liveness.steerStallMs ?? CODEX_STEER_STALL_MS;
+        const turnStallMs = this.liveness.turnStallMs ?? CODEX_TURN_STALL_MS;
+        let lastEventAt = Date.now();
+        let pendingSteer = null;
+        let pendingServerRequests = 0;
+        let healing = false;
+        let stalled = false;
+        let livenessTimer = null;
         if (!srv.start())
             return { ok: false, text: '', error: 'failed to start codex app-server', stopReason: 'error' };
         let settled = false;
@@ -320,20 +370,78 @@ export class CodexDriver {
                 state.sessionId = threadId;
                 ctx.emit({ type: 'session', sessionId: threadId });
             }
+            // Heal a wedged/lost steer in place: (optionally) interrupt the dead turn, then restart
+            // it with the same input. Runs under the SAME run()/turnDone, so upper layers just see
+            // the turn continue. `healing` suppresses the interrupt's own turn/completed echo.
+            const heal = async (opts) => {
+                if (settled || healing || !pendingSteer)
+                    return;
+                healing = true;
+                const replayInput = pendingSteer.input;
+                pendingSteer = null;
+                if (opts.interrupt && state.sessionId && state.turnId) {
+                    await srv.request('turn/interrupt', { threadId: state.sessionId, turnId: state.turnId }, 5_000);
+                }
+                if (settled) {
+                    healing = false;
+                    return;
+                } // raced with abort / process death
+                const resp = await srv.request('turn/start', {
+                    threadId: state.sessionId,
+                    input: replayInput,
+                    model: input.model || undefined,
+                    effort: input.effort || undefined,
+                });
+                if (settled) {
+                    healing = false;
+                    return;
+                }
+                if (resp.error) {
+                    state.status = 'error';
+                    state.error = `steer recovery failed: ${resp.error.message || 'turn/start failed'}`;
+                    healing = false;
+                    settle();
+                    return;
+                }
+                state.turnId = resp.result?.turn?.id ?? state.turnId;
+                lastEventAt = Date.now();
+                healing = false;
+            };
+            // Codex emits no explicit compaction event, so track peak occupancy: a sharp
+            // mid-turn drop is an auto-compaction, surfaced as a live `compaction` signal.
+            let compactPeakTokens = 0;
             srv.onNotification((method, params) => {
                 if (params?.threadId && params.threadId !== state.sessionId && method !== 'turn/started')
                     return;
+                lastEventAt = Date.now();
+                if (pendingSteer && CODEX_PROGRESS_METHODS.has(method)) {
+                    const now = Date.now();
+                    if (now - pendingSteer.at >= CODEX_STEER_CONSUMED_MS)
+                        pendingSteer = null; // loop provably alive post-steer
+                    else
+                        pendingSteer.progressAt = now; // maybe pre-acceptance flush — keep watching
+                }
                 switch (method) {
                     case 'turn/started':
                         state.turnId = params?.turn?.id ?? null;
                         if (!steerRegistered && state.turnId) {
                             steerRegistered = true;
                             ctx.registerSteer(async (prompt, attachments = []) => {
-                                if (!state.sessionId || !state.turnId)
+                                if (settled || !state.sessionId || !state.turnId)
                                     return false;
-                                const r = await srv.request('turn/steer', { threadId: state.sessionId, expectedTurnId: state.turnId, input: buildTurnInput(prompt, attachments) }, 30_000);
-                                if (r.error)
+                                const steerInput = buildTurnInput(prompt, attachments);
+                                // Arm BEFORE sending — accepted, not yet consumed. Arming after the response
+                                // would race the response's own microtask against progress notifications the
+                                // server flushed in the same write, mis-arming against already-consumed steers.
+                                // Back-to-back steers accumulate so a heal replays everything swallowed.
+                                const armed = { input: [...(pendingSteer?.input ?? []), ...steerInput], at: Date.now(), progressAt: null };
+                                pendingSteer = armed;
+                                const r = await srv.request('turn/steer', { threadId: state.sessionId, expectedTurnId: state.turnId, input: steerInput }, 30_000);
+                                if (r.error) {
+                                    if (pendingSteer === armed)
+                                        pendingSteer = null; // rejected — nothing to watch
                                     return false;
+                                }
                                 state.turnId = r.result?.turnId ?? state.turnId;
                                 return true;
                             });
@@ -444,15 +552,40 @@ export class CodexDriver {
                     case 'thread/tokenUsage/updated': {
                         applyCodexTokenUsage(state, params?.tokenUsage || params?.usage);
                         ctx.emit({ type: 'usage', usage: codexUsageOf(state) });
+                        // No explicit boundary event: a sharp drop from a high peak IS a compaction.
+                        // Conservative thresholds (peak ≥ 50% of window, drop ≥ 25% of window) so
+                        // ordinary turn churn never trips it; re-baseline so it fires once per drop.
+                        {
+                            const cw = state.contextWindow ?? 0;
+                            const used = state.contextUsed ?? 0;
+                            if (cw > 0 && used >= 0) {
+                                if (used > compactPeakTokens)
+                                    compactPeakTokens = used;
+                                else if (compactPeakTokens / cw >= 0.5 && (compactPeakTokens - used) / cw >= 0.25) {
+                                    ctx.emit({ type: 'compaction', trigger: 'auto', atTokens: compactPeakTokens });
+                                    compactPeakTokens = used;
+                                }
+                            }
+                        }
                         break;
                     }
                     case 'turn/completed': {
                         const turn = params?.turn || {};
+                        applyCodexTokenUsage(state, params?.tokenUsage || turn.tokenUsage || turn.usage);
+                        ctx.emit({ type: 'usage', usage: codexUsageOf(state) });
+                        if (healing)
+                            break; // completion echo of the turn heal() just interrupted
+                        if (pendingSteer && !pendingSteer.progressAt && !ctx.signal.aborted && (turn.status ?? 'completed') === 'completed') {
+                            // The other face of the steer race: the turn finished without ever consuming
+                            // the injected input. Replay it as a fresh turn instead of settling — the
+                            // upper layers already dequeued the message on steer-ok, so settling here
+                            // would silently drop it.
+                            void heal({ interrupt: false });
+                            break;
+                        }
                         state.status = turn.status ?? 'completed';
                         if (turn.error)
                             state.error = turn.error.message || turn.error.code || 'turn error';
-                        applyCodexTokenUsage(state, params?.tokenUsage || turn.tokenUsage || turn.usage);
-                        ctx.emit({ type: 'usage', usage: codexUsageOf(state) });
                         settle();
                         break;
                     }
@@ -462,6 +595,7 @@ export class CodexDriver {
             // accept approvals by default (parity with the legacy codex driver). Never throw —
             // an unanswerable request degrades to an empty response, not a JSON-RPC error.
             srv.onRequest(async (method, params, id) => {
+                pendingServerRequests++; // a request awaiting a human legitimately silences the turn
                 try {
                     if (method === 'item/tool/requestUserInput') {
                         const interaction = codexUserInputToInteraction(params, `codex-input-${id}`);
@@ -479,6 +613,9 @@ export class CodexDriver {
                 catch {
                     return {};
                 }
+                finally {
+                    pendingServerRequests--;
+                }
             });
             const turnResp = await srv.request('turn/start', {
                 threadId: state.sessionId,
@@ -488,6 +625,35 @@ export class CodexDriver {
             });
             if (turnResp.error)
                 return { ok: false, text: state.text, error: turnResp.error.message || 'turn/start failed', stopReason: 'error', sessionId: state.sessionId };
+            // Liveness checker: heals a stalled steer, force-closes a silently dead turn. The
+            // silence clock only accumulates while nothing is visibly in flight — a running tool
+            // call or a server->client request parked on a human keeps the turn alive forever.
+            const checkEveryMs = Math.max(25, Math.min(5_000, Math.floor(Math.min(steerStallMs, turnStallMs) / 4)));
+            livenessTimer = setInterval(() => {
+                if (settled || healing)
+                    return;
+                const now = Date.now();
+                if (pendingSteer) {
+                    if (now - (pendingSteer.progressAt ?? pendingSteer.at) >= steerStallMs)
+                        void heal({ interrupt: true });
+                    return;
+                }
+                const busy = pendingServerRequests > 0 || [...toolCalls.values()].some(c => c.status === 'running');
+                if (busy) {
+                    lastEventAt = now;
+                    return;
+                }
+                if (now - lastEventAt < turnStallMs)
+                    return;
+                stalled = true;
+                state.error = state.error || `codex app-server went silent mid-turn (no events for ${Math.round(turnStallMs / 1000)}s); closing the turn`;
+                if (state.sessionId && state.turnId) {
+                    srv.request('turn/interrupt', { threadId: state.sessionId, turnId: state.turnId }, 5_000).finally(() => settle());
+                }
+                else
+                    settle();
+            }, checkEveryMs);
+            livenessTimer.unref?.();
             await turnDone;
             const usage = codexUsageOf(state);
             const ok2 = (state.status === 'completed' || state.status == null) && !state.error && !ctx.signal.aborted;
@@ -497,7 +663,7 @@ export class CodexDriver {
                 text: codexFinalText(state),
                 reasoning: finalReasoning || undefined,
                 error: state.error || (ctx.signal.aborted ? 'Interrupted by user.' : null),
-                stopReason: ctx.signal.aborted ? 'interrupted' : (state.status || 'end_turn'),
+                stopReason: ctx.signal.aborted ? 'interrupted' : stalled ? 'stalled' : (state.status || 'end_turn'),
                 sessionId: state.sessionId,
                 // Fork anchor: the turn id is the inclusive keep-boundary thread/fork's lastTurnId takes.
                 anchor: state.turnId,
@@ -505,6 +671,8 @@ export class CodexDriver {
             };
         }
         finally {
+            if (livenessTimer)
+                clearInterval(livenessTimer);
             srv.kill();
         }
     }

package/packages/kernel/dist/drivers/shared.d.ts

@@ -10,10 +10,7 @@ export declare function parseJsonLine(line: string): any | undefined;
 export declare function wireAbort(signal: AbortSignal, fn: () => void): () => void;
 /** SIGTERM a child, swallowing the already-dead race. */
 export declare function sigterm(proc: ChildProcess | null | undefined): void;
-/** Mime type when the file is an inlineable image, else null. */
-export declare function imageMimeForFile(filePath: string): string | null;
-/** The text note substituted for a non-image attachment. */
-export declare function attachedFileNote(filePath: string): string;
+export { imageMimeForFile, attachedFileNote } from '../attachments.js';
 /**
  * Context-window occupancy as a display percent (one decimal, capped at 99.9).
  * Pass `used` as null when the caller wants "no data" rather than 0%.

package/packages/kernel/dist/drivers/shared.js

@@ -1,4 +1,3 @@
-import { extname } from 'node:path';
 // Driver-internal helpers shared by the concrete drivers (claude/codex/gemini/acp).
 // NOT part of the public API — nothing here is re-exported by any barrel. Each helper
 // exists because the same code appeared verbatim in 3+ drivers.
@@ -43,20 +42,9 @@ export function sigterm(proc) {
     }
     catch { /* ignore */ }
 }
-// Attachment vocabulary: every driver inlines the same image formats (the Anthropic
-// vision set, which the others accept too) and notes non-image files the same way.
-const IMAGE_MIME_BY_EXT = {
-    '.png': 'image/png', '.jpg': 'image/jpeg', '.jpeg': 'image/jpeg',
-    '.gif': 'image/gif', '.webp': 'image/webp',
-};
-/** Mime type when the file is an inlineable image, else null. */
-export function imageMimeForFile(filePath) {
-    return IMAGE_MIME_BY_EXT[extname(filePath).toLowerCase()] ?? null;
-}
-/** The text note substituted for a non-image attachment. */
-export function attachedFileNote(filePath) {
-    return `[Attached file: ${filePath}]`;
-}
+// Attachment vocabulary lives in ../attachments.ts (the Hub also normalizes oversized
+// images there); re-exported so drivers keep one import site for driver-internal helpers.
+export { imageMimeForFile, attachedFileNote } from '../attachments.js';
 /**
  * Context-window occupancy as a display percent (one decimal, capped at 99.9).
  * Pass `used` as null when the caller wants "no data" rather than 0%.

package/packages/kernel/dist/index.d.ts

@@ -9,7 +9,7 @@ export type { SessionStore, CoreSessionRecord, ModelResolver, ModelInjection, To
 export type { LoomIO, PromptInput, Surface, SurfaceCapabilities, Plugin, SpawnContribution, } from './contracts/surface.js';
 export { FsSessionStore, NullModelResolver, NoopToolProvider, PassthroughSystemPromptBuilder, AutoCancelInteractionHandler, DeferToTerminalInteractionHandler, NoopCatalog, defaultBaseDir, } from './ports/defaults.js';
 export { EchoDriver } from './drivers/echo.js';
-export { ClaudeDriver } from './drivers/claude.js';
+export { ClaudeDriver, type ClaudeDriverOptions } from './drivers/claude.js';
 export { CodexDriver } from './drivers/codex.js';
 export { GeminiDriver } from './drivers/gemini.js';
 export { AcpDriver, type AcpDriverConfig } from './drivers/acp.js';