fix: prevent memory leak in KiloSessionPromptQueue.cancel for sessions without active tails

[IamCoder18]

Context

Fix a memory leak in KiloSessionPromptQueue.cancel() where calling cancel on a session with no queued prompts would leave permanent entries in the module-level versions, targets, latest, and activeSince maps. This occurred because cancel() always incremented and stored a version, but cleanup only happens in enqueue's release phase.

Closes #9211

Implementation

Modified cancel() to check if the session has an active tail before modifying state. If tails.has(sessionID) is false, the function now cleans up any existing entries in versions, targets, latest, and activeSince maps instead of creating new ones.

export function cancel(sessionID: SessionID) {
  return Effect.sync(() => {
    if (!tails.has(sessionID)) {
      versions.delete(sessionID)
      targets.delete(sessionID)
      latest.delete(sessionID)
      activeSince.delete(sessionID)
      return
    }
    versions.set(sessionID, version(sessionID) + 1)
  })
}

Get in Touch

Discord: @IamCoder18

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Resolved Issues

✅ Previously flagged (round 1): Missing test for the memory-leak scenario — resolved: dedicated test added in test/kilocode/session-prompt-queue.test.ts (line 494).

✅ Previously flagged (round 2): _hasInternalState exported from production namespace without an internal annotation — resolved: /** @internal - test-only helper */ JSDoc comment added above the function.

Logic Review

All three aspects of the fix are correct:

• cancel() logic: When !tails.has(sessionID), all four stale map entries are deleted and the function returns early. When a tail is active, the version is incremented so in-flight slots detect cancellation via slot.version !== version(sessionID).
• _hasInternalState helper: Checks the correct 4 maps (versions, targets, latest, activeSince), intentionally omitting tails since an active tail entry is valid live state. Annotated as @internal.
• New test: Verifies that cancel() on a never-enqueued session leaves no state and that a subsequent enqueue() executes the work branch normally.
• Changeset: .changeset/session-prompt-queue-memory-leak.md added with correct patch bump for @kilocode/cli and accurate user-facing description.

Files Reviewed (3 files)

• packages/opencode/src/kilocode/session/prompt-queue.ts
• packages/opencode/test/kilocode/session-prompt-queue.test.ts
• .changeset/session-prompt-queue-memory-leak.md

---

_{Reviewed by claude-sonnet-4.6 · 172,820 tokens}

[johnnyeric]

Hey @IamCoder18, could you provide details about how the issue was reproduced and fix tested? Also, please consider adding a unit test to prevent the issue from happening again.

[IamCoder18]

Hey @johnnyeric, added the test as requested.

The test verifies that calling cancel() on a session with no active tail doesn't leave stale entries in the internal maps (versions, targets, latest, activeSince). I added a debug helper _hasInternalState() to check map state.

The test:

1. Creates a fresh session ID (no prior enqueue)
2. Calls cancel() on it
3. Verifies no internal state remains (_hasInternalState() === false)
4. Verifies subsequent enqueue() works normally

Without the fix, step 3 fails (I've ran it and test this) because cancel() leaves a stale entry in the versions map.