chore: rolling promotion dev -> main

[namastex888]

Rolling Promotion PR

Auto-maintained rolling promotion PR from dev to main.

Process:

• This PR is automatically created and kept open
• Human reviews and merges when ready
• Label ready-to-merge added when all checks pass

IMPORTANT: Merge with "Create a merge commit" — NEVER squash.
Squash merging breaks history sync between dev and main,
causing the next rolling PR to show all commits again.

Human approval required for merge to production.

Summary by CodeRabbit

• Chores

  • Package/plugin version bumped to 4.260407.2

• New Features

  • SendMessage can be routed via the omni bridge in turn-based mode, publishing replies and blocking native handling when applicable.

• Improvements

  • Bridge now handles session reset events during startup, evicts cancelled sessions, and drains queued messages after resets.
  • Turn-based prompt updated to use SendMessage/omni workflow.

• Tests

  • Added/expanded tests covering session reset handling and SendMessage interception.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 6bc97f84-f72c-468b-b47b-43d09a58299c

📥 Commits

Reviewing files that changed from the base of the PR and between 7cb9692 and 2ae5f39.

📒 Files selected for processing (5)

• .claude-plugin/marketplace.json
• package.json
• plugins/genie/.claude-plugin/plugin.json
• plugins/genie/package.json
• src/lib/providers/__tests__/claude-sdk.test.ts

---

📝 Walkthrough

Walkthrough

Bumps genie package/plugin versions to 4.260407.2; adds a SendMessage PreToolUse hook in the Claude executor to publish Omni replies and deny local handling; expands OmniBridge to handle omni.session.reset.> (including spawning placeholders), shutdown/eviction, and queue draining; adds extensive tests and updates agent run/state records.

Changes

Cohort / File(s)	Summary
Manifest bumps \.claude-plugin/marketplace.json, package.json, plugins/genie/.claude-plugin/plugin.json, plugins/genie/package.json	Version strings updated from 4.260406.4 → 4.260407.2 across root and plugin manifests.
Omni bridge runtime src/services/omni-bridge.ts	Adds cancelled?: boolean on session entries; subscribes to omni.session.reset.>; adds processSessionResetEvents, handleSessionReset, spawning-placeholder matching, spawn cancellation, executor shutdown paths, and queue draining.
Claude executor & hook src/services/executors/claude-sdk.ts, src/services/executors/__tests__/claude-sdk.test.ts	Adds exported createSendMessageOmniHook and wires a PreToolUse hook to intercept SendMessage in turn-based mode: publishes Omni NATS replies, returns permissionDecision: 'deny' on handled/invalid conditions; extensive tests for hook behavior and integration.
Turn-based prompt src/services/executors/turn-based-prompt.ts	Rewrites prompt guidance to prefer SendMessage(recipient: "omni", ...) plus omni done flow; removes older omni say/other commands.
Omni bridge tests src/services/__tests__/omni-bridge.test.ts	Large new test suite validating session reset handling, spawn-time cancellation, queue draining, subject parsing (dot-containing chatIds), malformed payload tolerance, and subscription to omni.session.reset.>.
Executor/provider tests (mocks) src/lib/providers/__tests__/claude-sdk.test.ts	Switched to shared SDK query mock utilities (_sdk-mocks) and reset lifecycle; minor test import/arrangement changes.
Agent state / runs .genie/agents/metrics-updater/runs.jsonl, .genie/agents/metrics-updater/state.json	Added a failed metrics-updater runs.jsonl entry and updated state.json last_run timestamp and last_run_status to failed.

Sequence Diagram(s)

sequenceDiagram
  autonumber
  participant NATS as NATS (omni.session.reset.>)
  participant Bridge as OmniBridge
  participant Map as SessionMap
  participant Spawn as SpawningPlaceholder
  participant Exec as Executor
  participant Queue as MessageQueue

  NATS->>Bridge: publish reset subject (omni.session.reset.{inst}.{chat})
  Bridge->>Map: findSessionKey(instance, chat) (match spawning placeholders or live key)
  alt placeholder exists & spawning
    Bridge->>Spawn: mark cancelled = true, clear buffered messages, close turn tracker
    Bridge->>Map: remove session entry
    Bridge->>Queue: drain queued messages (free slot -> spawn new)
    Note right of Spawn: when spawn resolves
    Spawn->>Exec: created executor detected as cancelled -> shutdown executor
  else live session exists
    Bridge->>Exec: close turn tracker, executor.shutdown(...)
    Bridge->>Map: remove session entry
    Bridge->>Queue: drain queued messages
  else no session found
    Bridge-->>NATS: no-op (no shutdown)
  end

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

Possibly related PRs

• chore: rolling promotion dev -> main #1078 — Overlaps in Omni/NATS integration and executor hook areas; strong code-level overlap with SendMessage hook and natsPublish changes.
• chore: rolling promotion dev -> main #1059 — Related modifications to omni bridge and Claude SDK executor session/turn handling; touches lifecycle/shutdown/drain logic.
• chore: promote dev to main — skills sync fix #624 — Related manifest/package version bumps for the genie plugin/package.

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 66.67% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'chore: rolling promotion dev -> main' accurately describes the main change—a rolling promotion PR that brings changes from dev to main.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch dev

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[gemini-code-assist]

Code Review

This pull request updates the version of the genie plugin and its associated packages from 4.260406.4 to 4.260406.5 across multiple configuration files, including marketplace.json and package.json. I have no feedback to provide as the changes are consistent version increments.

[namastex888]

@codex review this pr

[coderabbitai]

Actionable comments posted: 7

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In @.genie/agents/metrics-updater/runs.jsonl:
- Line 22: The state-load step for LAST_METRICS is silently discarding Python
stderr via `2>/dev/null || true`, which hides parse/load failures and prevents
fallback; update the state-load invocation in run-metrics.sh (the command that
sets LAST_METRICS) to remove the blind redirection and instead capture the
Python command's stderr and exit status, log a clear error message to stderr or
the script logger including the stderr output and the failing state file name,
and on non-zero exit set a fallback flag or ensure LAST_METRICS is left
unset/empty but record `fallback=true` so cached metrics are used; locate the
block around the LAST_METRICS assignment and replace the `2>/dev/null || true`
pattern with explicit error handling that logs the Python error and handles the
exit code.

In `@src/services/__tests__/omni-bridge.test.ts`:
- Around line 899-927: The test uses a fixed sleep to wait for the placeholder
session to be installed which is timing-sensitive; change the test to use a
deterministic barrier by having the provided spawnFn signal when it has reached
the blocked state (e.g., resolve a "spawnStarted" promise inside spawnFn) and
have the test await that promise before calling bridge.reset, then release the
original spawnGate via releaseSpawn to continue; update the concurrency
assertions to use Promise.allSettled() for any parallel operations (references:
spawnFn, spawnGate/releaseSpawn, routeMessage on bridge, and the OmniBridge
sessions map) so the test no longer relies on setTimeout timing.

In `@src/services/executors/__tests__/claude-sdk.test.ts`:
- Around line 694-719: The test hides the real failure because
waitForDeliveries() only waits the caught queue promise and swallows
_processDelivery() rejections; update the test to reset shared mocks
(resetAllMocks or clearMocks) at the start of this describe and replace the
broad waitForDeliveries() usage with an explicit delivery-specific signal: after
calling executor.deliver(...) await a promise that resolves when that delivery's
processing completes (for example by awaiting a delivery-specific event/Promise
returned by executor.deliver or by spying on executor._processDelivery and
awaiting its resolved/rejected call), then assert queryMock was called; ensure
you reference executor.spawn, executor.deliver, executor._processDelivery and
queryMock when locating the code to change.

In `@src/services/executors/claude-sdk.ts`:
- Around line 208-212: The executor currently returns an empty object when
OMNI_INSTANCE is set but OMNI_CHAT is missing (instanceId/chatId variables),
which silently allows fallback; change this to fail closed by returning the same
explicit deny used in the natsPublish branch (mirror its behavior) so the caller
receives an explicit deny response instead of {}; update the logic in the Claude
executor (where instanceId, chatId, agent are read) to check if instanceId is
present and chatId is missing and then return the explicit deny object used
elsewhere in this module (use the same shape/value as the natsPublish-deny) to
prevent silent fallback to SendMessage.

In `@src/services/omni-bridge.ts`:
- Around line 667-675: The session entry remains routable while awaiting
executor.shutdown, allowing routeMessage to re-resolve it; before awaiting
shutdown in the block handling reset (around the console.log for
sessionKey/actionTag and this.executor.shutdown(entry.session)), mark the
session as non-routable (e.g., set a flag like entry.routable = false or remove
the entry from this.sessions) immediately after
this.turnTracker.close(sessionKey, 'reset'), then await
this.executor.shutdown(entry.session); after shutdown complete call
this.removeSession(sessionKey) and await this.drainQueue(); ensure you update
any callers that check routability (routeMessage) to respect the chosen flag.
- Around line 659-668: The code emits a close with action 'reset' but the
TurnTracker/closedAction type currently permits only
'message'|'react'|'skip'|'timeout', so update the type definitions and
signatures to include 'reset' to keep typing exhaustive and metrics accurate:
add 'reset' to the closedAction union (and any alias types/interfaces) used by
TurnTracker.close and related files (e.g., the TurnTracker class/method
signature and the closedAction type in omni-turn.ts), remove any casts that
force the string, and ensure all handlers/metrics switch statements are updated
to handle the new 'reset' case consistently where TurnTracker.close(sessionKey,
'reset') is used (including the emit in omni-bridge.ts).
- Around line 642-667: Replace the new console.log calls with the module's
logger (use this.logger.debug or appropriate logger method) so core source
doesn't use console.log; e.g., in the reset-path around sessionKey,
entry.spawning and the subsequent branches (the log lines for "Session reset for
cold chat ... — no-op", "Session reset for spawning ... marking cancelled", and
"Session reset for ... evicting") should call this.logger.debug(`[omni-bridge]
...`) (or this.logger.info if you prefer a higher level) instead of console.log,
preserving the same message and actionTag interpolation; ensure the class has a
logger property or use the existing logger instance used elsewhere and remove
the console.log imports/usages afterwards.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 430910b0-df01-49cb-be02-d9205afe2504

📥 Commits

Reviewing files that changed from the base of the PR and between dc8d640 and 7cb9692.

📒 Files selected for processing (11)

• .claude-plugin/marketplace.json
• .genie/agents/metrics-updater/runs.jsonl
• .genie/agents/metrics-updater/state.json
• package.json
• plugins/genie/.claude-plugin/plugin.json
• plugins/genie/package.json
• src/services/__tests__/omni-bridge.test.ts
• src/services/executors/__tests__/claude-sdk.test.ts
• src/services/executors/claude-sdk.ts
• src/services/executors/turn-based-prompt.ts
• src/services/omni-bridge.ts

[coderabbitai]

⚠️ Potential issue | 🟠 Major

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -euo pipefail

echo "== Latest persisted state =="
cat .genie/agents/metrics-updater/state.json

echo
echo "== Last 3 run records =="
tail -n 3 .genie/agents/metrics-updater/runs.jsonl

echo
echo "== Verify fallback gate and final logging logic =="
fd run-metrics.sh .genie | while read -r f; do
  echo "--- $f"
  rg -n -C3 'LAST_METRICS|FETCH_OK|fallback|RUN_STATUS|RUNS_FILE' "$f"
done

Repository: automagik-dev/genie

Length of output: 4877

---

🏁 Script executed:

#!/bin/bash
# Check if the Python state load in run-metrics.sh has error handling
# and verify the exact logic around lines 79-88

sed -n '75,90p' .genie/agents/metrics-updater/tools/run-metrics.sh

echo "---"
echo "Looking for error handling in Python state load:"
sed -n '78,88p' .genie/agents/metrics-updater/tools/run-metrics.sh | cat -n

Repository: automagik-dev/genie

Length of output: 774

---

State file loading silently fails, disabling fallback when fetch errors occur.

The 2>/dev/null || true on line 86 of .genie/agents/metrics-updater/tools/run-metrics.sh suppresses all stderr from the Python state load, causing LAST_METRICS to remain empty if the state file is corrupted, unparseable, or missing. When line 22 ran and the fetch failed, this silent error prevented fallback from activating—even though cached metrics likely existed. The run was correctly recorded as fallback:false (since no metrics were available), but the root cause (state load failure) is invisible.

Replace 2>/dev/null || true with explicit error handling that logs state load failures so they can be diagnosed and recovered.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In @.genie/agents/metrics-updater/runs.jsonl at line 22, The state-load step for
LAST_METRICS is silently discarding Python stderr via `2>/dev/null || true`,
which hides parse/load failures and prevents fallback; update the state-load
invocation in run-metrics.sh (the command that sets LAST_METRICS) to remove the
blind redirection and instead capture the Python command's stderr and exit
status, log a clear error message to stderr or the script logger including the
stderr output and the failing state file name, and on non-zero exit set a
fallback flag or ensure LAST_METRICS is left unset/empty but record
`fallback=true` so cached metrics are used; locate the block around the
LAST_METRICS assignment and replace the `2>/dev/null || true` pattern with
explicit error handling that logs the Python error and handles the exit code.

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Replace the fixed sleep with a deterministic barrier.

Line 926 makes this reset-during-spawn test timing-sensitive. On a busy runner the placeholder may not exist yet, and on a fast runner the sleep is just dead time. Signal spawnFn when the spawn has reached the blocked state and await that promise instead.

Suggested fix

     // Hold the spawn promise open so we can fire reset mid-spawn.
     let releaseSpawn!: (s: OmniSession) => void;
+    let markSpawnStarted!: () => void;
     const spawnGate = new Promise<OmniSession>((resolve) => {
       releaseSpawn = resolve;
     });
+    const spawnStarted = new Promise<void>((resolve) => {
+      markSpawnStarted = resolve;
+    });

     const { executor, calls, makeSession } = makeMockExecutor({
       spawnFn: async (agentName, chatId) => {
+        markSpawnStarted();
         // Block until the test releases the spawn.
         const session = await spawnGate;
         return session ?? makeSession(agentName, chatId);
       },
     });
@@
-      // Yield so spawnSession installs the placeholder before we reset.
-      await new Promise((r) => setTimeout(r, 5));
+      // Wait until the spawn path is definitely blocked.
+      await spawnStarted;
       expect((bridge as any).sessions.has('test-agent:chat-1')).toBe(true);
       expect((bridge as any).sessions.get('test-agent:chat-1').spawning).toBe(true);

As per coding guidelines, **/*.test.ts: Use Promise.allSettled() pattern for concurrency tests.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/services/__tests__/omni-bridge.test.ts` around lines 899 - 927, The test
uses a fixed sleep to wait for the placeholder session to be installed which is
timing-sensitive; change the test to use a deterministic barrier by having the
provided spawnFn signal when it has reached the blocked state (e.g., resolve a
"spawnStarted" promise inside spawnFn) and have the test await that promise
before calling bridge.reset, then release the original spawnGate via
releaseSpawn to continue; update the concurrency assertions to use
Promise.allSettled() for any parallel operations (references: spawnFn,
spawnGate/releaseSpawn, routeMessage on bridge, and the OmniBridge sessions map)
so the test no longer relies on setTimeout timing.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

This regression test is already red and it hides the real failure.

The quality gate is failing at Line 710 with 0 queryMock calls. Because waitForDeliveries() only waits on the caught queue promise, any _processDelivery() rejection is swallowed and this test degrades into a misleading "not called" assertion. Reset the shared mocks in this describe and assert through a delivery-specific signal so the underlying setup error surfaces.

🧰 Tools

🪛 GitHub Check: Quality Gate (typecheck + lint + test)

[failure] 710-710: error: expect(received).toHaveBeenCalled()
Expected number of calls: >= 1
Received number of calls: 0

  at <anonymous> (/home/runner/_work/genie/genie/src/services/executors/__tests__/claude-sdk.test.ts:710:25)

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/services/executors/__tests__/claude-sdk.test.ts` around lines 694 - 719,
The test hides the real failure because waitForDeliveries() only waits the
caught queue promise and swallows _processDelivery() rejections; update the test
to reset shared mocks (resetAllMocks or clearMocks) at the start of this
describe and replace the broad waitForDeliveries() usage with an explicit
delivery-specific signal: after calling executor.deliver(...) await a promise
that resolves when that delivery's processing completes (for example by awaiting
a delivery-specific event/Promise returned by executor.deliver or by spying on
executor._processDelivery and awaiting its resolved/rejected call), then assert
queryMock was called; ensure you reference executor.spawn, executor.deliver,
executor._processDelivery and queryMock when locating the code to change.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Return an explicit deny when bridge context is incomplete.

Once OMNI_INSTANCE is set, this executor is already in turn-based/Omni mode. Returning {} here when OMNI_CHAT is empty silently falls back to native SendMessage, so the model believes it replied even though nothing can be published to omni.reply.*. Mirror the !natsPublish branch and fail closed instead.

Suggested fix

-    if (!instanceId || !chatId) return {};
+    if (!instanceId || !chatId) {
+      return {
+        hookSpecificOutput: {
+          hookEventName: 'PreToolUse',
+          permissionDecision: 'deny',
+          permissionDecisionReason: 'Omni bridge unavailable — missing OMNI_INSTANCE/OMNI_CHAT context.',
+        },
+      };
+    }

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	const instanceId = env.OMNI_INSTANCE ?? '';
	const chatId = env.OMNI_CHAT ?? '';
	const agent = env.OMNI_AGENT ?? '';
	if (!instanceId || !chatId) return {};
	const instanceId = env.OMNI_INSTANCE ?? '';
	const chatId = env.OMNI_CHAT ?? '';
	const agent = env.OMNI_AGENT ?? '';
	if (!instanceId || !chatId) {
	return {
	hookSpecificOutput: {
	hookEventName: 'PreToolUse',
	permissionDecision: 'deny',
	permissionDecisionReason: 'Omni bridge unavailable — missing OMNI_INSTANCE/OMNI_CHAT context.',
	},
	};
	}

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/services/executors/claude-sdk.ts` around lines 208 - 212, The executor
currently returns an empty object when OMNI_INSTANCE is set but OMNI_CHAT is
missing (instanceId/chatId variables), which silently allows fallback; change
this to fail closed by returning the same explicit deny used in the natsPublish
branch (mirror its behavior) so the caller receives an explicit deny response
instead of {}; update the logic in the Claude executor (where instanceId,
chatId, agent are read) to check if instanceId is present and chatId is missing
and then return the explicit deny object used elsewhere in this module (use the
same shape/value as the natsPublish-deny) to prevent silent fallback to
SendMessage.

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Don't add new console.log calls in src/.

These reset-path logs introduce fresh console.log usage in core source. Route them through the existing logger path or a permitted level instead.

As per coding guidelines, src/**/*.ts: No console.log in source code (enforced by biome, relaxed in tests).

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/services/omni-bridge.ts` around lines 642 - 667, Replace the new
console.log calls with the module's logger (use this.logger.debug or appropriate
logger method) so core source doesn't use console.log; e.g., in the reset-path
around sessionKey, entry.spawning and the subsequent branches (the log lines for
"Session reset for cold chat ... — no-op", "Session reset for spawning ...
marking cancelled", and "Session reset for ... evicting") should call
this.logger.debug(`[omni-bridge] ...`) (or this.logger.info if you prefer a
higher level) instead of console.log, preserving the same message and actionTag
interpolation; ensure the class has a logger property or use the existing logger
instance used elsewhere and remove the console.log imports/usages afterwards.

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

reset is outside the current turn-close contract.

TurnTracker.close() currently stores closedAction as 'message' | 'react' | 'skip' | 'timeout'. Emitting 'reset' here only works because src/services/omni-turn.ts casts the string, so downstream exhaustive handling and metrics still don't know reset closes exist. Extend that union before writing this value.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/services/omni-bridge.ts` around lines 659 - 668, The code emits a close
with action 'reset' but the TurnTracker/closedAction type currently permits only
'message'|'react'|'skip'|'timeout', so update the type definitions and
signatures to include 'reset' to keep typing exhaustive and metrics accurate:
add 'reset' to the closedAction union (and any alias types/interfaces) used by
TurnTracker.close and related files (e.g., the TurnTracker class/method
signature and the closedAction type in omni-turn.ts), remove any casts that
force the string, and ensure all handlers/metrics switch statements are updated
to handle the new 'reset' case consistently where TurnTracker.close(sessionKey,
'reset') is used (including the emit in omni-bridge.ts).

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Make the session non-routable before awaiting shutdown.

The first await in this branch is executor.shutdown(entry.session), but the entry stays in this.sessions until that completes. If another omni.message.* arrives for the same chat in that window, routeMessage() can still resolve this entry and deliver into the session the user just reset. Remove or mark it non-routable before the await.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/services/omni-bridge.ts` around lines 667 - 675, The session entry
remains routable while awaiting executor.shutdown, allowing routeMessage to
re-resolve it; before awaiting shutdown in the block handling reset (around the
console.log for sessionKey/actionTag and this.executor.shutdown(entry.session)),
mark the session as non-routable (e.g., set a flag like entry.routable = false
or remove the entry from this.sessions) immediately after
this.turnTracker.close(sessionKey, 'reset'), then await
this.executor.shutdown(entry.session); after shutdown complete call
this.removeSession(sessionKey) and await this.drainQueue(); ensure you update
any callers that check routability (routeMessage) to respect the chosen flag.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2ae5f390e3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Guard cancelled-spawn cleanup against newer same-key sessions

When a reset arrives during spawnSession, the placeholder is removed immediately, so a new spawn for the same {agent}:{chat} key can start before the original executor.spawn resolves. In that case, this cancelled-path executor.shutdown(session) can target the newer live session because executors key state by session.id (same {agent}:{chat}), which can terminate the replacement session and drop subsequent messages. Add a generation/token check (or verify the current map entry is still this placeholder) before shutting down.

Useful? React with 👍 / 👎.