lazycodex-ai 4.16.0 → 4.16.2

package/packages/omo-codex/plugin/components/ultrawork/skills/ultrawork/SKILL.md

@@ -56,9 +56,10 @@ exercises the surface; capture the artifact.
   1. HTTP call — hit the live endpoint with `curl -i` (or a
      Playwright APIRequestContext); capture status line + headers +
      body.
-  2. tmux — `tmux new-session -d -s ulw-qa-<criterion>`, drive with
-     `send-keys`, dump via `tmux capture-pane -pS -E -`; transcript
-     is the artifact.
+  2. Terminal / TUI - drive a real pty and prove it through the
+     xterm.js web terminal (see the TUI visual QA note below). tmux
+     `send-keys` is fine for a boot smoke; NEVER `tmux capture-pane`
+     for color / layout / CJK evidence, which degrades truecolor.
   3. Browser use — in Codex, use `browser:control-in-app-browser`
      first when available and no authenticated/persistent user browser
      profile is required. Otherwise use Chrome to drive the REAL page;
@@ -86,13 +87,13 @@ channel scenario when the behavior is user-facing. `--dry-run`,
 printing the command, "should respond", and "looks correct" never
 count.
 
-For TUI visual QA, terminal transcripts alone are not enough when a
-visual surface is being evaluated. In this repo, prefer
-`node script/qa/web-terminal-visual-qa.mjs --title "<surface>" --from-file <capture.txt> --evidence-dir <dir>`
-or the helper's `--command` tmux-backed PTY connector when available.
-Outside this repo, capture equivalent browser/computer-use rendered
-terminal evidence: screenshot, plain transcript, rendered HTML or action
-log, and cleanup receipt.
+For TUI visual QA, render the terminal through the real xterm.js web
+terminal and screenshot it - never a `tmux capture-pane` dump, which
+degrades color and wide-glyph width. In this repo:
+`node script/qa/web-terminal-visual-qa.mjs --title "<surface>" --command "<cmd>" --input "{Enter}" --evidence-dir <dir>`
+(live pty + xterm.js in Chrome; `--from-file <capture>` replays a raw
+stream). Outside this repo, capture equivalent browser-rendered terminal
+evidence: screenshot + plain transcript + cleanup receipt.
 
 # Bootstrap (DO ALL FOUR BEFORE ANY OTHER WORK — NO SKIPPING)
 
@@ -265,6 +266,7 @@ Every `multi_agent_v1.spawn_agent` message is self-contained and starts with
 handoff. Use `fork_context: false` unless full history is truly
 required; paste only the context the child needs. Full-history forks can
 make the child continue old parent context instead of the delegated task.
+If your tool list has a flat `spawn_agent` with a required `task_name` instead of `multi_agent_v1.*` (`multi_agent_v2`), rewrite: `fork_context: false` becomes `fork_turns: "none"`, `send_input` becomes `send_message`, finished agents end on their own (no `close_agent`; `followup_task` re-tasks, `interrupt_agent` stops), and `wait_agent` takes only `timeout_ms`, returning on any child mailbox activity.
 
 # TOML-backed subagent routing compatibility
 Treat TOML-backed role routing as **routing-unverified**. The
@@ -294,6 +296,13 @@ evidence for that step. Do not start dependent implementation until the
 audit, research, or review result is integrated or explicitly recorded
 as inconclusive. Do not generate a plan before spawned research lanes
 that feed the plan have returned or been closed as inconclusive.
+Spawn every independent child for the current wave first. After the wave
+is launched, run `multi_agent_v1.wait_agent` for each spawned child until
+each reaches terminal status (`completed`, `failed`, `blocked`, or
+explicitly recorded inconclusive) before any dependent `update_plan`
+transition, `create_goal` continuation, implementation tool call, plan
+drafting, approval-gate work, PR handoff, or final response. A timeout is
+not terminal status.
 Do not write the final answer, PR handoff, or completion summary while
 active child agents remain open. Use short `multi_agent_v1.wait_agent` cycles.
 After two silent waits send `TASK STILL ACTIVE: return <deliverable> or

package/packages/omo-codex/plugin/components/ultrawork/skills/ulw-plan/SKILL.md

@@ -67,6 +67,19 @@ Fan out read-only research before deciding. Every spawn names DELIVERABLE / SCOP
 multi_agent_v1.spawn_agent({"message":"TASK: act as an explorer. DELIVERABLE: ... SCOPE: ... VERIFY: ...","agent_type":"explorer","fork_context":false})
 ```
 
+If your tool list has a flat `spawn_agent` with a required `task_name` instead of `multi_agent_v1.*` (`multi_agent_v2`), rewrite: add `"task_name":"<lowercase_digits_underscores>"`, replace `"fork_context":false` with `"fork_turns":"none"`, and `wait_agent` takes only `timeout_ms`, returning on any child mailbox activity (finished agents end on their own).
+
+Spawn every independent child for the current wave first. After the wave
+is launched, use `multi_agent_v1.wait_agent` for each child until each
+reaches terminal status. A timeout is not terminal status. Do not start dependent planning, drafting, approval-gate work, or final handoff until each child result is integrated or recorded as inconclusive.
+
+For work likely to exceed one wait cycle, require the child to send
+`WORKING: <task> - <current phase>` before long passes and
+`BLOCKED: <reason>` only when progress stops. A `multi_agent_v1.wait_agent`
+timeout only means no new mailbox update arrived. Treat a running child as
+alive. Fallback only when the child is completed without the deliverable,
+ack-only after followup, explicitly `BLOCKED:`, or no longer running.
+
 Roles: `explorer` (internal patterns/conventions/tests), `librarian` (external docs/contracts), `metis` (gap analysis), `momus` (high-accuracy plan review). Full spawn/wait/fallback discipline is in `references/full-workflow.md`.
 
 ## Stop rules

package/packages/omo-codex/plugin/components/ultrawork/skills/ulw-plan/references/full-workflow.md

@@ -96,6 +96,8 @@ Every spawn starts with `TASK:`, then DELIVERABLE / SCOPE / VERIFY inside `messa
 multi_agent_v1.spawn_agent({"message":"TASK: act as an explorer. DELIVERABLE: ... SCOPE: ... VERIFY: ...","agent_type":"explorer","fork_context":false})
 ```
 
+If your tool list has a flat `spawn_agent` with a required `task_name` instead of `multi_agent_v1.*` (`multi_agent_v2`), rewrite: add `"task_name":"<lowercase_digits_underscores>"`, replace `"fork_context":false` with `"fork_turns":"none"`, and `wait_agent` takes only `timeout_ms`, returning on any child mailbox activity (finished agents end on their own — skip the close step).
+
 Roles: `explorer`, `librarian`, `metis`, `momus`. Spawn long plan/reviewer agents in the background and poll with short waits; require the child to send `WORKING: <task> - <phase>` before long passes and `BLOCKED: <reason>` only when progress stops. A wait timeout only means no new mailbox update arrived; treat a running child as alive. Fall back only when the child completed without the deliverable, is ack-only after followup, explicitly `BLOCKED:`, or no longer running; then respawn a smaller `fork_context: false` job. Close each agent after integrating its result.
 
 ## Stop rules

package/packages/omo-codex/plugin/components/ultrawork/test/codex-hook.test.ts

@@ -190,7 +190,7 @@ describe("codex ultrawork hook", () => {
 		expect(parsed.hookSpecificOutput.additionalContext).toMatch(/# Manual-QA channels/);
 		expect(parsed.hookSpecificOutput.additionalContext).toMatch(/TESTS ALONE NEVER PROVE DONE/);
 		expect(parsed.hookSpecificOutput.additionalContext).toMatch(/1\. HTTP call/);
-		expect(parsed.hookSpecificOutput.additionalContext).toMatch(/2\. tmux/);
+		expect(parsed.hookSpecificOutput.additionalContext).toMatch(/2\. Terminal \/ TUI/);
 		expect(parsed.hookSpecificOutput.additionalContext).toMatch(/3\. Browser use/);
 		expect(parsed.hookSpecificOutput.additionalContext).toMatch(/4\. Computer use/);
 		expect(parsed.hookSpecificOutput.additionalContext).toMatch(/CLEANUP \(PAIRED/);
@@ -244,6 +244,27 @@ describe("codex ultrawork hook", () => {
 		expect(directive).toMatch(/WORKING:/);
 	});
 
+	it("#given directive #when inspected #then blocks dependent work until spawned planners finish", () => {
+		// given
+		const payload = {
+			hook_event_name: "UserPromptSubmit",
+			prompt: "ulw",
+		};
+
+		// when
+		const output = runUserPromptSubmitHook(payload, { skillFilePath: null });
+		const parsed = parseHookOutput(output);
+
+		// then
+		const directive = parsed.hookSpecificOutput.additionalContext;
+		expect(directive).toMatch(/Subagent-dependent transition barrier/);
+		expect(directive).toMatch(/Spawn every independent child for the current wave first/);
+		expect(directive).toMatch(/After the wave\s+is launched[\s\S]{0,240}wait_agent[\s\S]{0,240}terminal status/);
+		expect(directive).not.toMatch(/Immediately after any `multi_agent_v1\.spawn_agent`/);
+		expect(directive).toMatch(/Do not start dependent implementation/);
+		expect(directive).toMatch(/Do not mark an `update_plan` step `completed`/);
+	});
+
 	it("#given directive #when inspected #then keeps impact-proportional sizing invariants", () => {
 		// given
 		const payload = {

package/packages/omo-codex/plugin/components/ulw-loop/CHANGELOG.md

@@ -2,6 +2,10 @@
 
 ## [0.1.0] - unreleased
 
+- **Memory:** steering ledger entries no longer embed the full plan four times (`before`/`after` at both the audit and entry level). Accepted steers now record a compact `UlwLoopSteeringPlanSnapshot` (plan counters + only the goals the mutation touched), shrinking a measured real-world entry from 189KB to 7.8KB (~24x) and ending quadratic `ledger.jsonl` growth over long runs.
+- **Memory:** steering dedup (`--idempotency-key` / `promptSignature`) streams the ledger line-by-line with a substring pre-filter instead of `JSON.parse`-ing every entry into memory; dedup returns strip legacy full-plan `before`/`after` payloads from re-surfaced audits. `readSteeringLedgerEntries` streams too.
+- **Memory:** `withUlwLoopMutationLock` no longer retains the mutation result (full plan/audit) per `(repo, scope)` in its module-level lock map; settled gates self-evict, so long-lived embedders stop accumulating entries.
+- **Fix:** `omo ulw-loop steer --idempotency-key` was parsed but never forwarded into the proposal, so CLI steers never deduped. It is now wired through.
 - Standalone ultrawork injection (`--with-ultrawork`) now emits the same compact bootstrap pointer as the ultrawork component (opener mandate, `create_goal` with `objective` only, read the bundled `ultrawork` skill at a runtime-resolved absolute path), falling back to the full bundled directive when the plugin skills tree is absent. Keeps the injected payload below Codex App's hook-output truncation budget (code-yeongyu/oh-my-openagent#5828).
 - Initial scaffold of codex-ulw-loop plugin.
 - Per-Criterion Cycle: `EXECUTE` is now **EXECUTE-AS-SCENARIO** — the agent must run the Manual-QA channel scenario the criterion named (HTTP call / tmux / browser use / computer use; see new `## Manual-QA channels` section). Inserted a new **CLEAN (PAIRED, NEVER SKIP)** step that tears down every QA-spawned process / `tmux` session / browser context / container / port / temp dir before recording evidence; the cleanup receipt is embedded in the `--evidence` string. Missing receipt → record BLOCKED, not PASS. Added Constraint #13 and a Stop Rule for leftover state.

package/packages/omo-codex/plugin/components/ulw-loop/directive.md

@@ -49,9 +49,10 @@ exercises the surface; capture the artifact.
   1. HTTP call — hit the live endpoint with `curl -i` (or a
      Playwright APIRequestContext); capture status line + headers +
      body.
-  2. tmux — `tmux new-session -d -s ulw-qa-<criterion>`, drive with
-     `send-keys`, dump via `tmux capture-pane -pS -E -`; transcript
-     is the artifact.
+  2. Terminal / TUI - drive a real pty and prove it through the
+     xterm.js web terminal (see the TUI visual QA note below). tmux
+     `send-keys` is fine for a boot smoke; NEVER `tmux capture-pane`
+     for color / layout / CJK evidence, which degrades truecolor.
   3. Browser use — in Codex, use `browser:control-in-app-browser`
      first when available and no authenticated/persistent user browser
      profile is required. Otherwise use Chrome to drive the REAL page;
@@ -79,13 +80,13 @@ channel scenario when the behavior is user-facing. `--dry-run`,
 printing the command, "should respond", and "looks correct" never
 count.
 
-For TUI visual QA, terminal transcripts alone are not enough when a
-visual surface is being evaluated. In this repo, prefer
-`node script/qa/web-terminal-visual-qa.mjs --title "<surface>" --from-file <capture.txt> --evidence-dir <dir>`
-or the helper's `--command` tmux-backed PTY connector when available.
-Outside this repo, capture equivalent browser/computer-use rendered
-terminal evidence: screenshot, plain transcript, rendered HTML or action
-log, and cleanup receipt.
+For TUI visual QA, render the terminal through the real xterm.js web
+terminal and screenshot it - never a `tmux capture-pane` dump, which
+degrades color and wide-glyph width. In this repo:
+`node script/qa/web-terminal-visual-qa.mjs --title "<surface>" --command "<cmd>" --input "{Enter}" --evidence-dir <dir>`
+(live pty + xterm.js in Chrome; `--from-file <capture>` replays a raw
+stream). Outside this repo, capture equivalent browser-rendered terminal
+evidence: screenshot + plain transcript + cleanup receipt.
 
 # Bootstrap (DO ALL FOUR BEFORE ANY OTHER WORK — NO SKIPPING)
 
@@ -258,6 +259,7 @@ Every `multi_agent_v1.spawn_agent` message is self-contained and starts with
 handoff. Use `fork_context: false` unless full history is truly
 required; paste only the context the child needs. Full-history forks can
 make the child continue old parent context instead of the delegated task.
+If your tool list has a flat `spawn_agent` with a required `task_name` instead of `multi_agent_v1.*` (`multi_agent_v2`), rewrite: `fork_context: false` becomes `fork_turns: "none"`, `send_input` becomes `send_message`, finished agents end on their own (no `close_agent`; `followup_task` re-tasks, `interrupt_agent` stops), and `wait_agent` takes only `timeout_ms`, returning on any child mailbox activity.
 
 # TOML-backed subagent routing compatibility
 Treat TOML-backed role routing as **routing-unverified**. The
@@ -287,6 +289,13 @@ evidence for that step. Do not start dependent implementation until the
 audit, research, or review result is integrated or explicitly recorded
 as inconclusive. Do not generate a plan before spawned research lanes
 that feed the plan have returned or been closed as inconclusive.
+Spawn every independent child for the current wave first. After the wave
+is launched, run `multi_agent_v1.wait_agent` for each spawned child until
+each reaches terminal status (`completed`, `failed`, `blocked`, or
+explicitly recorded inconclusive) before any dependent `update_plan`
+transition, `create_goal` continuation, implementation tool call, plan
+drafting, approval-gate work, PR handoff, or final response. A timeout is
+not terminal status.
 Do not write the final answer, PR handoff, or completion summary while
 active child agents remain open. Use short `multi_agent_v1.wait_agent` cycles.
 After two silent waits send `TASK STILL ACTIVE: return <deliverable> or

package/packages/omo-codex/plugin/components/ulw-loop/dist/cli-commands.js

@@ -24,8 +24,8 @@ export async function ulwLoopCommand(argv) {
     const rest = argv.slice(1);
     const repoRoot = process.cwd();
     const json = hasFlag(rest, "--json");
-    const scope = commandScope(rest);
     try {
+        const scope = commandScope(rest);
         if (!isUlwLoopSubcommand(command)) {
             if (json) {
                 printJsonError(new UlwLoopError(`Unknown ulw-loop subcommand: ${command}.`, "ULW_LOOP_SUBCOMMAND_UNKNOWN", {
@@ -79,7 +79,20 @@ export async function ulwLoopCommand(argv) {
 function unhandledSubcommand(command) {
     throw new UlwLoopError(`Unhandled ulw-loop subcommand: ${String(command)}.`, "ULW_LOOP_SUBCOMMAND_UNHANDLED");
 }
+const SESSION_ID_FLAG = "--session-id";
+function sessionIdFlagPresent(argv) {
+    return hasFlag(argv, SESSION_ID_FLAG) || argv.some((arg) => arg.startsWith(`${SESSION_ID_FLAG}=`));
+}
 function commandScope(argv) {
-    const sessionId = readValue(argv, "--session-id") ?? resolveUlwLoopSessionIdFromEnv();
+    if (sessionIdFlagPresent(argv)) {
+        const sessionId = readValue(argv, SESSION_ID_FLAG)?.trim();
+        if (!sessionId) {
+            throw new UlwLoopError(`${SESSION_ID_FLAG} requires a non-empty value.`, "ULW_LOOP_SESSION_ID_REQUIRED", {
+                details: { flag: SESSION_ID_FLAG },
+            });
+        }
+        return { sessionId };
+    }
+    const sessionId = resolveUlwLoopSessionIdFromEnv();
     return sessionId === null ? undefined : { sessionId };
 }

package/packages/omo-codex/plugin/components/ulw-loop/dist/cli-steering.js

@@ -83,7 +83,8 @@ function neverKind(kind) { return fail(`Unsupported steering kind: ${String(kind
 export async function parseSteeringProposal(argv) {
     const kind = parseSteeringKind(argv);
     const source = parseSteeringSource(argv);
-    const base = { kind, source, evidence: required(argv, "--evidence"), rationale: required(argv, "--rationale") };
+    const idempotencyKey = text(readValue(argv, "--idempotency-key"), "--idempotency-key");
+    const base = { kind, source, evidence: required(argv, "--evidence"), rationale: required(argv, "--rationale"), ...(idempotencyKey === undefined ? {} : { idempotencyKey }) };
     switch (kind) {
         case "add_subgoal": return normalizeSteeringProposal({ ...base, title: required(argv, "--title"), objective: required(argv, "--objective") });
         case "split_subgoal": {

package/packages/omo-codex/plugin/components/ulw-loop/dist/cli.js

@@ -521,7 +521,9 @@ function formatCodexGoalReconciliation(reconciliation) {
 }
 
 // components/ulw-loop/src/plan-io.ts
+import { createReadStream } from "node:fs";
 import { appendFile, mkdir, readFile as readFile4, rename, writeFile } from "node:fs/promises";
+import { createInterface } from "node:readline";
 var LEGACY_OBJECTIVE_PREFIX = `Complete all ulw-loop stories in ${ULW_LOOP_DIR}/${ULW_LOOP_GOALS}: `;
 var LEGACY_OBJECTIVE = `Complete all ulw-loop stories listed in ${ULW_LOOP_DIR}/${ULW_LOOP_GOALS}. Use ${ULW_LOOP_DIR}/${ULW_LOOP_LEDGER} as the durable audit trail.`;
 var locks = new Map;
@@ -540,11 +542,18 @@ async function withUlwLoopMutationLock(repoRoot, scopeOrFn, maybeFn) {
   if (fn === undefined)
     throw new UlwLoopError("Missing ulw-loop mutation body.", "ULW_LOOP_LOCK_BODY_MISSING");
   const lockKey = `${repoRoot}\x00${ulwLoopRelativeDir(scope)}`;
-  const prior = locks.get(lockKey) ?? Promise.resolve();
+  const prior = locks.get(lockKey) ?? Promise.resolve(undefined);
   const run = prior.then(fn, fn);
-  locks.set(lockKey, run.catch(() => {
+  const gate = run.then(() => {
     return;
-  }));
+  }, () => {
+    return;
+  });
+  locks.set(lockKey, gate);
+  gate.then(() => {
+    if (locks.get(lockKey) === gate)
+      locks.delete(lockKey);
+  });
   return run;
 }
 async function readUlwLoopPlan(repoRoot, scope) {
@@ -591,22 +600,36 @@ async function appendLedger(repoRoot, entry, scope) {
   await appendFile(ulwLoopLedgerPath(repoRoot, scope), `${JSON.stringify(entry)}
 `, "utf8");
 }
-async function readSteeringLedgerEntries(repoRoot, scope) {
-  let raw;
+async function* ledgerLines(repoRoot, scope) {
+  const stream = createReadStream(ulwLoopLedgerPath(repoRoot, scope), { encoding: "utf8" });
+  const lines = createInterface({ input: stream, crlfDelay: Number.POSITIVE_INFINITY });
   try {
-    raw = await readFile4(ulwLoopLedgerPath(repoRoot, scope), "utf8");
+    for await (const line of lines) {
+      if (line.trim().length > 0)
+        yield line;
+    }
   } catch (error) {
-    if (hasCode(error, "ENOENT"))
-      return [];
-    throw error;
+    if (!hasCode(error, "ENOENT"))
+      throw error;
+  } finally {
+    lines.close();
+    stream.destroy();
   }
-  const entries = [];
-  for (const line of raw.split(/\r?\n/).filter(Boolean)) {
+}
+async function findAcceptedSteeringLedgerEntry(repoRoot, key, scope) {
+  const probe = JSON.stringify(key);
+  for await (const line of ledgerLines(repoRoot, scope)) {
+    if (!line.includes(probe))
+      continue;
     const entry = JSON.parse(line);
-    if (isSteeringKind(entry.kind))
-      entries.push(entry);
+    if (!isSteeringKind(entry.kind))
+      continue;
+    if (entry.steering?.invariant.accepted !== true)
+      continue;
+    if (entry.idempotencyKey === key || entry.steering.idempotencyKey === key || entry.steering.promptSignature === key)
+      return entry;
   }
-  return entries;
+  return;
 }
 
 // components/ulw-loop/src/evidence.ts
@@ -1254,7 +1277,8 @@ function neverKind(kind) {
 async function parseSteeringProposal(argv) {
   const kind = parseSteeringKind(argv);
   const source = parseSteeringSource(argv);
-  const base = { kind, source, evidence: required2(argv, "--evidence"), rationale: required2(argv, "--rationale") };
+  const idempotencyKey = text(readValue(argv, "--idempotency-key"), "--idempotency-key");
+  const base = { kind, source, evidence: required2(argv, "--evidence"), rationale: required2(argv, "--rationale"), ...idempotencyKey === undefined ? {} : { idempotencyKey } };
   switch (kind) {
     case "add_subgoal":
       return normalizeSteeringProposal({ ...base, title: required2(argv, "--title"), objective: required2(argv, "--objective") });
@@ -1725,6 +1749,30 @@ async function recordFinalReviewBlockers(repoRoot, args, scope) {
   });
 }
 
+// components/ulw-loop/src/steering-snapshot.ts
+function buildSteeringPlanSnapshot(plan, changedGoalIds) {
+  const snapshot = {
+    updatedAt: plan.updatedAt,
+    goalCount: plan.goals.length,
+    goalIds: plan.goals.map((goal) => goal.id),
+    goals: plan.goals.filter((goal) => changedGoalIds.has(goal.id))
+  };
+  return plan.activeGoalId === undefined ? snapshot : { ...snapshot, activeGoalId: plan.activeGoalId };
+}
+function changedGoalIdsBetween(before, after) {
+  const beforeById = new Map(before.goals.map((goal) => [goal.id, goal]));
+  const changed = new Set;
+  for (const goal of after.goals) {
+    const prior = beforeById.get(goal.id);
+    if (prior === undefined || JSON.stringify(prior) !== JSON.stringify(goal))
+      changed.add(goal.id);
+    beforeById.delete(goal.id);
+  }
+  for (const id of beforeById.keys())
+    changed.add(id);
+  return changed;
+}
+
 // components/ulw-loop/src/steering.ts
 var SOURCES2 = ["user_prompt_submit", "finding", "cli"];
 var PROTECTED = new Set(["aggregateCompletion", "codexObjective", "codexObjectiveAliases", "originalConstraints", "qualityGate", "status", "completedAt", "completionStatus"]);
@@ -1982,15 +2030,20 @@ async function steerUlwLoop(repoRoot, proposal, scope) {
   return withUlwLoopMutationLock(repoRoot, scope, async () => {
     const plan = await readUlwLoopPlan(repoRoot, scope);
     const key = proposal.idempotencyKey ?? proposal.promptSignature;
-    const prior = key === undefined ? undefined : (await readSteeringLedgerEntries(repoRoot, scope)).find((entry) => entry.steering?.invariant.accepted === true && (entry.idempotencyKey === key || entry.steering.idempotencyKey === key || entry.steering.promptSignature === key));
-    if (prior?.steering !== undefined)
-      return { plan, accepted: true, audit: { ...prior.steering, deduped: true }, rejectedReasons: [], deduped: true };
+    const prior = key === undefined ? undefined : await findAcceptedSteeringLedgerEntry(repoRoot, key, scope);
+    if (prior?.steering !== undefined) {
+      const { before: _before, after: _after, ...compactPrior } = prior.steering;
+      return { plan, accepted: true, audit: { ...compactPrior, deduped: true }, rejectedReasons: [], deduped: true };
+    }
     const audit = validateUlwLoopSteeringProposal(plan, proposal);
     const accepted = audit.invariant.accepted;
     const next = accepted ? applySteeringMutation(plan, proposal, audit) : plan;
-    const finalAudit = { ...audit, before: plan };
-    if (accepted)
-      finalAudit.after = next;
+    const finalAudit = { ...audit };
+    if (accepted) {
+      const changed = changedGoalIdsBetween(plan, next);
+      finalAudit.before = buildSteeringPlanSnapshot(plan, changed);
+      finalAudit.after = buildSteeringPlanSnapshot(next, changed);
+    }
     if (accepted)
       await writePlan(repoRoot, next, scope);
     await appendLedger(repoRoot, ledgerEntry(proposal, finalAudit, proposal.now?.toISOString() ?? iso()), scope);
@@ -2006,10 +2059,6 @@ function ledgerEntry(proposal, audit, at) {
     entry.criterionId = proposal.criterionId;
   if (proposal.idempotencyKey !== undefined)
     entry.idempotencyKey = proposal.idempotencyKey;
-  if (audit.before !== undefined)
-    entry.before = audit.before;
-  if (audit.after !== undefined)
-    entry.after = audit.after;
   return entry;
 }
 
@@ -2205,8 +2254,8 @@ async function ulwLoopCommand(argv) {
   const rest = argv.slice(1);
   const repoRoot = process.cwd();
   const json = hasFlag(rest, "--json");
-  const scope = commandScope(rest);
   try {
+    const scope = commandScope(rest);
     if (!isUlwLoopSubcommand(command)) {
       if (json) {
         printJsonError(new UlwLoopError(`Unknown ulw-loop subcommand: ${command}.`, "ULW_LOOP_SUBCOMMAND_UNKNOWN", {
@@ -2264,8 +2313,21 @@ async function ulwLoopCommand(argv) {
 function unhandledSubcommand(command) {
   throw new UlwLoopError(`Unhandled ulw-loop subcommand: ${String(command)}.`, "ULW_LOOP_SUBCOMMAND_UNHANDLED");
 }
+var SESSION_ID_FLAG = "--session-id";
+function sessionIdFlagPresent(argv) {
+  return hasFlag(argv, SESSION_ID_FLAG) || argv.some((arg) => arg.startsWith(`${SESSION_ID_FLAG}=`));
+}
 function commandScope(argv) {
-  const sessionId = readValue(argv, "--session-id") ?? resolveUlwLoopSessionIdFromEnv();
+  if (sessionIdFlagPresent(argv)) {
+    const sessionId2 = readValue(argv, SESSION_ID_FLAG)?.trim();
+    if (!sessionId2) {
+      throw new UlwLoopError(`${SESSION_ID_FLAG} requires a non-empty value.`, "ULW_LOOP_SESSION_ID_REQUIRED", {
+        details: { flag: SESSION_ID_FLAG }
+      });
+    }
+    return { sessionId: sessionId2 };
+  }
+  const sessionId = resolveUlwLoopSessionIdFromEnv();
   return sessionId === null ? undefined : { sessionId };
 }
 

package/packages/omo-codex/plugin/components/ulw-loop/dist/plan-io.d.ts

@@ -6,3 +6,9 @@ export declare function readUlwLoopPlan(repoRoot: string, scope?: UlwLoopScope):
 export declare function writePlan(repoRoot: string, plan: UlwLoopPlan, scope?: UlwLoopScope): Promise<void>;
 export declare function appendLedger(repoRoot: string, entry: UlwLoopLedgerEntry, scope?: UlwLoopScope): Promise<void>;
 export declare function readSteeringLedgerEntries(repoRoot: string, scope?: UlwLoopScope): Promise<UlwLoopLedgerEntry[]>;
+/**
+ * First accepted steering entry matching an idempotency key/prompt signature.
+ * A cheap substring probe on the raw line skips JSON.parse for the vast
+ * majority of entries, so dedup stays flat even on legacy multi-MB ledgers.
+ */
+export declare function findAcceptedSteeringLedgerEntry(repoRoot: string, key: string, scope?: UlwLoopScope): Promise<UlwLoopLedgerEntry | undefined>;

package/packages/omo-codex/plugin/components/ulw-loop/dist/plan-io.js

@@ -1,4 +1,6 @@
+import { createReadStream } from "node:fs";
 import { appendFile, mkdir, readFile, rename, writeFile } from "node:fs/promises";
+import { createInterface } from "node:readline";
 import { aggregateCodexObjectiveForScope } from "./goal-status.js";
 import { repoRelative, ulwLoopDir, ulwLoopGoalsPath, ulwLoopLedgerPath, ulwLoopRelativeDir, } from "./paths.js";
 import { iso, ULW_LOOP_DIR, ULW_LOOP_GOALS, ULW_LOOP_LEDGER, UlwLoopError } from "./types.js";
@@ -20,9 +22,17 @@ export async function withUlwLoopMutationLock(repoRoot, scopeOrFn, maybeFn) {
     if (fn === undefined)
         throw new UlwLoopError("Missing ulw-loop mutation body.", "ULW_LOOP_LOCK_BODY_MISSING");
     const lockKey = `${repoRoot}\0${ulwLoopRelativeDir(scope)}`;
-    const prior = locks.get(lockKey) ?? Promise.resolve();
+    const prior = locks.get(lockKey) ?? Promise.resolve(undefined);
     const run = prior.then(fn, fn);
-    locks.set(lockKey, run.catch(() => undefined));
+    // The stored gate resolves to undefined so the map never retains fn's result
+    // (plans/audits), and it removes itself once no newer waiter replaced it —
+    // otherwise a long-lived host leaks one entry per (repo, scope) forever.
+    const gate = run.then(() => undefined, () => undefined);
+    locks.set(lockKey, gate);
+    void gate.then(() => {
+        if (locks.get(lockKey) === gate)
+            locks.delete(lockKey);
+    });
     return run;
 }
 export async function readUlwLoopPlan(repoRoot, scope) {
@@ -69,21 +79,57 @@ export async function appendLedger(repoRoot, entry, scope) {
     await mkdir(ulwLoopDir(repoRoot, scope), { recursive: true });
     await appendFile(ulwLoopLedgerPath(repoRoot, scope), `${JSON.stringify(entry)}\n`, "utf8");
 }
-export async function readSteeringLedgerEntries(repoRoot, scope) {
-    let raw;
+/**
+ * Streams raw ledger lines without materializing the file. Real ledgers reach
+ * many MB (legacy entries embedded full-plan snapshots), so `readFile` here
+ * ballooned every steer/dedup path; line-at-a-time keeps memory O(longest line).
+ */
+async function* ledgerLines(repoRoot, scope) {
+    const stream = createReadStream(ulwLoopLedgerPath(repoRoot, scope), { encoding: "utf8" });
+    const lines = createInterface({ input: stream, crlfDelay: Number.POSITIVE_INFINITY });
     try {
-        raw = await readFile(ulwLoopLedgerPath(repoRoot, scope), "utf8");
+        for await (const line of lines) {
+            if (line.trim().length > 0)
+                yield line;
+        }
     }
     catch (error) {
-        if (hasCode(error, "ENOENT"))
-            return [];
-        throw error;
+        if (!hasCode(error, "ENOENT"))
+            throw error;
+    }
+    finally {
+        lines.close();
+        stream.destroy();
     }
+}
+export async function readSteeringLedgerEntries(repoRoot, scope) {
     const entries = [];
-    for (const line of raw.split(/\r?\n/).filter(Boolean)) {
+    for await (const line of ledgerLines(repoRoot, scope)) {
         const entry = JSON.parse(line);
         if (isSteeringKind(entry.kind))
             entries.push(entry);
     }
     return entries;
 }
+/**
+ * First accepted steering entry matching an idempotency key/prompt signature.
+ * A cheap substring probe on the raw line skips JSON.parse for the vast
+ * majority of entries, so dedup stays flat even on legacy multi-MB ledgers.
+ */
+export async function findAcceptedSteeringLedgerEntry(repoRoot, key, scope) {
+    const probe = JSON.stringify(key);
+    for await (const line of ledgerLines(repoRoot, scope)) {
+        if (!line.includes(probe))
+            continue;
+        const entry = JSON.parse(line);
+        if (!isSteeringKind(entry.kind))
+            continue;
+        if (entry.steering?.invariant.accepted !== true)
+            continue;
+        if (entry.idempotencyKey === key ||
+            entry.steering.idempotencyKey === key ||
+            entry.steering.promptSignature === key)
+            return entry;
+    }
+    return undefined;
+}

package/packages/omo-codex/plugin/components/ulw-loop/dist/steering-snapshot.d.ts

@@ -0,0 +1,15 @@
+import type { UlwLoopPlan } from "./domain-types.js";
+import type { UlwLoopSteeringPlanSnapshot } from "./steering-types.js";
+/**
+ * Compact before/after snapshots for steering ledger entries.
+ *
+ * Ledger entries used to embed the FULL plan (every goal, criterion, and
+ * evidence string) twice per audit and twice more at the entry top level.
+ * That made each accepted steer O(plan size) on disk, so the ledger grew
+ * quadratically over a run, and every dedup scan re-hydrated all of it into
+ * memory. A snapshot instead records plan-level counters plus only the goals
+ * the mutation actually touched, keeping each entry O(changed goals).
+ */
+export declare function buildSteeringPlanSnapshot(plan: UlwLoopPlan, changedGoalIds: ReadonlySet<string>): UlwLoopSteeringPlanSnapshot;
+/** Ids of goals that differ between two plans, including added or removed goals. */
+export declare function changedGoalIdsBetween(before: UlwLoopPlan, after: UlwLoopPlan): Set<string>;

package/packages/omo-codex/plugin/components/ulw-loop/dist/steering-snapshot.js

@@ -0,0 +1,33 @@
+/**
+ * Compact before/after snapshots for steering ledger entries.
+ *
+ * Ledger entries used to embed the FULL plan (every goal, criterion, and
+ * evidence string) twice per audit and twice more at the entry top level.
+ * That made each accepted steer O(plan size) on disk, so the ledger grew
+ * quadratically over a run, and every dedup scan re-hydrated all of it into
+ * memory. A snapshot instead records plan-level counters plus only the goals
+ * the mutation actually touched, keeping each entry O(changed goals).
+ */
+export function buildSteeringPlanSnapshot(plan, changedGoalIds) {
+    const snapshot = {
+        updatedAt: plan.updatedAt,
+        goalCount: plan.goals.length,
+        goalIds: plan.goals.map((goal) => goal.id),
+        goals: plan.goals.filter((goal) => changedGoalIds.has(goal.id)),
+    };
+    return plan.activeGoalId === undefined ? snapshot : { ...snapshot, activeGoalId: plan.activeGoalId };
+}
+/** Ids of goals that differ between two plans, including added or removed goals. */
+export function changedGoalIdsBetween(before, after) {
+    const beforeById = new Map(before.goals.map((goal) => [goal.id, goal]));
+    const changed = new Set();
+    for (const goal of after.goals) {
+        const prior = beforeById.get(goal.id);
+        if (prior === undefined || JSON.stringify(prior) !== JSON.stringify(goal))
+            changed.add(goal.id);
+        beforeById.delete(goal.id);
+    }
+    for (const id of beforeById.keys())
+        changed.add(id);
+    return changed;
+}

package/packages/omo-codex/plugin/components/ulw-loop/dist/steering-types.d.ts

@@ -1,5 +1,5 @@
 import type { UlwLoopSteeringMutationKind, UlwLoopSteeringSource } from "./constants.js";
-import type { UlwLoopPlan } from "./domain-types.js";
+import type { UlwLoopItem, UlwLoopPlan } from "./domain-types.js";
 export interface UlwLoopSteeringInvariantResult {
     accepted: boolean;
     structuralInvariantAccepted: boolean;
@@ -39,13 +39,20 @@ export interface UlwLoopSteeringProposal {
     idempotencyKey?: string;
     now?: Date;
 }
+export interface UlwLoopSteeringPlanSnapshot {
+    readonly updatedAt: string;
+    readonly activeGoalId?: string;
+    readonly goalCount: number;
+    readonly goalIds: readonly string[];
+    readonly goals: readonly UlwLoopItem[];
+}
 export interface UlwLoopSteeringAudit {
     kind: UlwLoopSteeringMutationKind;
     source: UlwLoopSteeringSource;
     targetGoalIds: string[];
     criterionId?: string;
-    before?: unknown;
-    after?: unknown;
+    before?: UlwLoopSteeringPlanSnapshot;
+    after?: UlwLoopSteeringPlanSnapshot;
     evidence: string;
     rationale: string;
     invariant: UlwLoopSteeringInvariantResult;