fix(recovery): make user-closed panes terminal#323
Conversation
|
@coderabbitai review |
|
@codex review |
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_5060debd-7c9b-4a94-9867-775f14553adb) |
|
You need to increase your spend limit or enable usage-based billing to run background agents. Go to Cursor |
✅ Action performedReview finished.
|
|
Warning Review limit reached
Next review available in: 15 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Organization UI Review profile: ASSERTIVE Plan: Pro Run ID: 📒 Files selected for processing (4)
📝 WalkthroughWalkthroughClose-forensics sweeps now return emitted events and are awaited before AgentEngine reconciliation, allowing matching agents to become terminal ChangesSurface close attribution
Sequence Diagram(s)sequenceDiagram
participant AgentEngine
participant closeForensicsRunner
participant AgentState
AgentEngine->>closeForensicsRunner: Run close-forensics sweep
closeForensicsRunner-->>AgentEngine: Return tab_close events
AgentEngine->>AgentState: Mark matching agents user_killed
AgentEngine->>AgentState: Reconcile absence and crash recovery
Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
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. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/server.ts`:
- Around line 6819-6830: In the close_surface state sweep, wrap the
stateMgr.updateRecord call for each matching record in try/catch. Ignore the
expected “Agent not found” error caused by concurrent lifecycle eviction so
surface closure still succeeds, while rethrowing unexpected errors; keep
lifecycleRegistry.set for successfully updated records.
🪄 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: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: a8d4adaa-5d80-4477-94bd-b03855b9ae2f
📒 Files selected for processing (5)
src/agent-engine.tssrc/close-forensics.tssrc/server.tstests/agent-engine.test.tstests/server.test.ts
📜 Review details
🧰 Additional context used
📓 Path-based instructions (6)
tests/**/*.test.ts
📄 CodeRabbit inference engine (CLAUDE.md)
tests/**/*.test.ts: Mirror source layout in tests (src/foo.ts->tests/foo.test.ts).
Do not add integration tests that require a running cmux instance; tests should be fully mocked.
Files:
tests/server.test.tstests/agent-engine.test.ts
tests/server.test.ts
📄 CodeRabbit inference engine (CLAUDE.md)
Server tests should mock the cmux client via
createServer({ exec, skipAgentLifecycle }).
Files:
tests/server.test.ts
src/**/*.ts
📄 CodeRabbit inference engine (CLAUDE.md)
src/**/*.ts: Build the project with TypeScript (tsc) and keep source code compatible with Node 20+ and Zod-based typing.
Use theok(data)anderr(error)helpers for consistent MCP tool responses.
All MCP tool handlers must return{ content: TextContent[], structuredContent?, isError? }.
Files:
src/server.tssrc/close-forensics.tssrc/agent-engine.ts
src/server.ts
📄 CodeRabbit inference engine (CLAUDE.md)
Register all MCP tools in
server.ts, including the 33 tool handlers, and conditionally skip agent-lifecycle tools whenskipAgentLifecycle: true.
Files:
src/server.ts
tests/**/*agent-engine*.test.ts
📄 CodeRabbit inference engine (CLAUDE.md)
Agent engine tests should use 1-second timeouts for state-change detection.
Files:
tests/agent-engine.test.ts
src/agent-engine.ts
📄 CodeRabbit inference engine (CLAUDE.md)
Implement agent lifecycle behavior in
agent-engine.ts, including spawning, monitoring, and quality tracking.
Files:
src/agent-engine.ts
🧠 Learnings (1)
📚 Learning: 2026-03-15T10:42:35.917Z
Learnt from: EtanHey
Repo: EtanHey/cmuxlayer PR: 1
File: tests/quality-tracking.test.ts:171-200
Timestamp: 2026-03-15T10:42:35.917Z
Learning: In tests/quality-tracking.test.ts for the cmuxlayer project, ensure that at or above 80% context quality degradation, behavior depends on depth: depth-0 agents receive a /compact command; depth > 0 agents are killed and logged (kill + log). Respawn of non-root agents is out of scope for v1. Treat the design doc quality tracking section as the authoritative source for this behavior, and align test expectations accordingly.
Applied to files:
tests/server.test.tstests/agent-engine.test.ts
🔇 Additional comments (12)
src/close-forensics.ts (4)
392-405: LGTM!
447-447: LGTM!
505-507: LGTM!
688-688: LGTM!src/agent-engine.ts (5)
47-51: LGTM!
318-326: LGTM!
770-772: LGTM!
3629-3629: LGTM!
3647-3703: LGTM!tests/agent-engine.test.ts (2)
39-39: LGTM!
3421-3502: LGTM!tests/server.test.ts (1)
9735-9792: LGTM!
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 55b6777f51
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".
| record.surface_id !== args.surface && | ||
| record.surface_uuid !== args.surface |
There was a problem hiding this comment.
When close_surface is called on a mutable ref that has been recycled, this condition marks any persisted agent whose stale surface_id equals args.surface even if that agent has a different surface_uuid. The registry code elsewhere treats UUID-bearing records as authoritative specifically to avoid adopting recycled refs; marking the stale record user_killed here suppresses crash recovery for the original agent after its UUID disappears. Only fall back to surface_id matching for UUID-less records, or verify the live UUID before setting this flag.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in aa08dd3. close_surface now resolves the current ref to its observed stable UUID before closing; UUID-backed records match only the requested/observed UUID, while mutable-ref fallback is limited to UUID-less records. The new recycled-ref regression proves a stale UUID owner remains recoverable, and the normal stable-owner close remains terminal.
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_2b83a2e1-d216-4e24-a541-7de7cc01cfdb) |
|
@coderabbitai review |
|
@codex review |
✅ Action performedReview finished.
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: aa08dd37bb
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".
| const matchesClosedSurface = record.surface_uuid | ||
| ? record.surface_uuid.toLowerCase() === requestedSurfaceKey || | ||
| record.surface_uuid.toLowerCase() === observedSurfaceUuid | ||
| : record.surface_id === args.surface; |
There was a problem hiding this comment.
When findSurfaceByRef cannot return a UUID (for example on older/ref-only surface enumeration or a transient listing failure), records that already have surface_uuid take this branch and are compared only against args.surface and undefined. In that case a forced close of the exact managed, crash-recoverable surface leaves user_killed false, so the later absence reconciliation can classify it as a recoverable crash and restart a pane the user explicitly closed. Please fall back to record.surface_id === args.surface when no observed UUID is available.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in d0e1109. close_surface now uses the observed stable UUID whenever enumeration succeeds, so a recycled ref cannot terminalize its stale owner; only when enumeration is unavailable/ref-only does an explicit close fall back to the requested mutable ref. The regression covers both the recycled-ref mismatch and the degraded UUID-unavailable path.
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_ccb1e207-de95-4451-8723-9d500fe0abd8) |
|
@coderabbitai review @codex review Final head is d0e1109. Please re-review the closed-pane recovery fix, including stable UUID matching, recycled-ref protection, concurrent state eviction, and the degraded UUID-unavailable fallback. |
🧩 Analysis chain🏁 Script executed: #!/bin/bash
set -e
rg -n "surface_uuid|recycled|eviction|user_killed|closeForensics" src/agent-engine.ts | head -80
echo "---close-forensics---"
rg -n "surface_uuid|recycled|eviction|uuid" src/close-forensics.ts | head -80Length of output: 3422 🏁 Script executed: #!/bin/bash
sed -n '3640,3710' src/agent-engine.ts
echo "=== uuid fallback ==="
sed -n '4390,4475' src/agent-engine.ts
echo "=== concurrent eviction / user_killed marking ==="
sed -n '4690,4900' src/agent-engine.ts
echo "=== close-forensics.ts full ==="
cat -n src/close-forensics.ts | sed -n '1,200p'Length of output: 9096 🏁 Script executed: #!/bin/bash
echo "=== forensics ingest wiring (3640-3710) ==="
sed -n '3640,3710p' src/agent-engine.ts
echo "=== uuid fallback (4390-4475) ==="
sed -n '4390,4475p' src/agent-engine.ts
echo "=== concurrent eviction / user_killed marking (4690-4900) ==="
sed -n '4690,4900p' src/agent-engine.tsLength of output: 12002
I'll confirm these hold up under the full incremental diff review and flag anything that doesn't line up with this reasoning. ✅ Action performedReview finished.
|
|
Final verification receipt for d0e1109:
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d0e11097be
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".
| record.surface_uuid.toLowerCase() === observedSurfaceUuid || | ||
| (observedSurfaceUuid === undefined && | ||
| record.surface_id === args.surface) | ||
| : record.surface_id === args.surface; |
There was a problem hiding this comment.
When cmux does return observedSurfaceUuid for the target, UUID-less records still match solely by the mutable surface_id here. If an older UUID-less state row still points at a ref that has since been recycled, forcing close_surface on the current UUID-backed surface will mark that stale row user_killed, suppressing crash recovery and allowing it to be purged even though stable identity was available to avoid the recycled-ref match. The ref fallback should only apply when the observation is UUID-unavailable/ref-only.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 568b40b. All mutable-ref fallback is now gated on observedSurfaceUuid being unavailable, including UUID-less legacy records. A RED regression added a UUID-less stale owner on a recycled ref and observed it incorrectly become user_killed:true; after the one-line guard, that stale row remains recoverable while the existing degraded UUID-unavailable explicit-close case remains terminal. Full 2131-test suite and live contract pass.
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_dd963762-eb49-4373-8d5e-6268eed3dac0) |
|
@coderabbitai review @codex review Final head is now 568b40b. Please verify the last UUID-less legacy-row recycled-ref guard; all mutable-ref fallback is restricted to UUID-unavailable observation, while the degraded explicit-close regression remains terminal. |
🧩 Analysis chain🏁 Script executed: #!/bin/bash
set -e
git log --oneline -5
echo "=== resolveAgentIoRoute region ==="
grep -n "resolveAgentIoRoute\|provenance\|UUID-unavailable\|uuid_unavailable\|legacy" src/agent-engine.ts | head -100Length of output: 2580 🏁 Script executed: #!/bin/bash
sed -n '2960,3080p' src/agent-engine.ts
echo "=== 3340,3470 ==="
sed -n '3340,3470p' src/agent-engine.ts
echo "=== 4390,4470 ==="
sed -n '4390,4470p' src/agent-engine.tsLength of output: 12659 🏁 Script executed: #!/bin/bash
git show 568b40b --stat
echo "=== diff ==="
git diff d0e1109 568b40b -- src/agent-engine.ts | head -200Length of output: 575 🏁 Script executed: #!/bin/bash
git diff d0e1109 568b40b -- src/server.ts tests/server.test.tsLength of output: 1856 🏁 Script executed: #!/bin/bash
sed -n '6760,6850p' src/server.tsLength of output: 3563 🏁 Script executed: #!/bin/bash
sed -n '9790,9910p' tests/server.test.tsLength of output: 3984 🏁 Script executed: #!/bin/bash
sed -n '9910,9990p' tests/server.test.tsLength of output: 2793 This confirms both properties EtanHey asked me to verify at 568b40b:
No further issues found in this narrow diff ( ✏️ Learnings added
✅ Action performedReview finished.
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 568b40b939
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".
| events | ||
| .filter( | ||
| (event) => | ||
| event.origin === "tab_close" && |
There was a problem hiding this comment.
When cmux emits surface.closed with origin: "workspace_teardown", this filter drops it even though it is the same app-level/user-close forensics path as a tab close. In that scenario a crash-recoverable agent whose workspace was closed by the operator is not marked user_killed before absence reconciliation, so it can be classified as an unintentional disappearance and respawned on a later sweep. Include workspace teardown in the terminal-intent set so closing a workspace does not resurrect its agents.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in cd1ac43. The pre-reconciliation intentional-close set now includes both app-level origins documented by close forensics: tab_close and workspace_teardown. The lifecycle regression is parameterized over both; workspace_teardown was RED with user_killed:false and is now GREEN, while the paired genuine PTY-crash resume test remains intact. Full 2132-test suite, typecheck/build, and live isolated runtime contract pass.
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_3a76b9e9-a8b0-4837-a7c0-750fc37c14c2) |
|
@coderabbitai review @codex review Final head is cd1ac43. Please re-review the intentional close-origin set: both |
🧩 Analysis chain🏁 Script executed: #!/bin/bash
set -e
git log --oneline -8
echo "=== diff 568b40b..cd1ac43 ==="
git diff 568b40b cd1ac43 --stat
echo "=== close-forensics diff ==="
git diff 568b40b cd1ac43 -- src/close-forensics.tsLength of output: 1011 🏁 Script executed: #!/bin/bash
set -e
git diff 568b40b cd1ac43 -- src/agent-engine.ts
echo "=== context around change ==="
rg -n "workspace_teardown|tab_close|origin" src/agent-engine.ts | head -50Length of output: 2698 🏁 Script executed: #!/bin/bash
set -e
rg -n "origin" src/close-forensics.ts
echo "=== test diff snippet ==="
git diff 568b40b cd1ac43 -- tests/agent-engine.test.ts | head -200Length of output: 6877
No issues found in this diff — the terminal-origin set is correctly widened and both new/old behaviors are covered by tests. ✏️ Learnings added
✅ Action performedReview finished.
|
…3vbw6 # Conflicts: # src/agent-engine.ts
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_15c9fd19-c70b-4f46-8ec7-2c94a500361a) |
|
@coderabbitai review @codex review Final conflict-free head is 07a5ce0 after merging current main. The only conflict was the |
🧩 Analysis chain🏁 Script executed: #!/bin/bash
set -e
git log --oneline -5
echo "=== merge commit stat ==="
git show 07a5ce0 --stat | head -30
echo "=== import block in agent-engine.ts ==="
sed -n '1,40p' src/agent-engine.ts
echo "=== check cleanScreenText usage ==="
rg -n "cleanScreenText" src/agent-engine.ts
echo "=== check CloseForensicsSweepResult usage ==="
rg -n "CloseForensicsSweepResult" src/agent-engine.tsLength of output: 3527 🏁 Script executed: #!/bin/bash
set -e
echo "=== conflict markers check ==="
rg -n "^(<<<<<<<|=======|>>>>>>>)" src/agent-engine.ts src/fleet-sidebar.ts src/fleet-sidebar-cli.ts src/screen-parser.ts || echo "none found"
echo "=== close-forensics origin logic intact ==="
rg -n "tab_close|workspace_teardown" src/agent-engine.ts
echo "=== unrelated changes touching close/user_killed area ==="
git diff cd1ac43 07a5ce0 -- src/agent-engine.ts | head -150Length of output: 6329
No issues found in this merge — the conflict resolution is correct and the PR's core fix remains intact alongside main's unrelated fleet-sidebar work. ✅ Action performedReview finished.
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 07a5ce0f72
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".
|
|
||
| private async runSweepOnce(): Promise<void> { | ||
| this.currentSweepScreenSignatures = new Map(); | ||
| await this.runCloseForensicsBestEffort(); |
There was a problem hiding this comment.
When another cmuxlayer process writes an agent state after this engine's previous sweep, that record is not present in this.registry.list() until registry.reconcile() runs. Because this line now consumes and advances close-forensics before that import, a tab_close for the newly-on-disk agent is skipped by markIntentionalSurfaceCloses; the same sweep then imports the record and later can crash-recover it because the close event will not be replayed. Consider reconciling/importing disk state before consuming forensics, or matching the events against stateMgr.listStates() as well.
Useful? React with 👍 / 👎.
| : observedSurfaceUuid === undefined && | ||
| record.surface_id === args.surface; |
There was a problem hiding this comment.
For an existing managed agent whose state record has no surface_uuid (for example an upgraded/ref-only record), a successful enumeration of the target surface sets observedSurfaceUuid, so this fallback refuses to match the record by its current surface_id. A forced close_surface then closes the pane but leaves user_killed=false, making the intentional close eligible for crash recovery instead of terminal handling. The recycled-ref guard still needs to avoid stale UUID owners, but the current branch also drops the real UUID-less owner of the requested ref.
Useful? React with 👍 / 👎.
1 participant
Summary
tab_closesetsuser_killedbefore crash recovery is evaluatedclose_surfacecalls terminal for matching managed agentsRoot cause and reproduction
The lifecycle sweep reconciled/evicted missing surfaces and ran crash recovery before the asynchronous close-forensics tail. A user-closing a managed tab therefore looked identical to a disappeared PTY:
crash_recover=true, a captured CLI session, and nouser_killedmarker.The live event stream reproduced the race twice on 2026-07-14: a
surface.closedevent withorigin=tab_closeat 16:23:48.852Z was followed by a replacement surface and a crash-recovery respawn; closing the replacement at 16:24:22.664Z produced the same sequence again.RED / counterfactual evidence
Before the fix, the two new regressions failed:
tab_close:newSplitwas called once, resurrecting the agentclose_surface:user_killedremained falseAfter implementation, temporarily disabling only the two new terminal-intent writes made both tests fail again with
user_killed=false. Restoring them returned the paired suite to green, including the existing genuine PTY-death respawn test.Verification
env -u CMUX_SOCKET_PATH -u CMUX_DAEMON_SOCKET bunx vitest run tests/agent-engine.test.ts tests/close-forensics.test.ts tests/server.test.ts: 460/460 passedenv -u CMUX_SOCKET_PATH -u CMUX_DAEMON_SOCKET bun run test: 104/104 files, 2,129/2,129 tests passedenv -u CMUX_SOCKET_PATH -u CMUX_DAEMON_SOCKET bun run typecheck: passedenv -u CMUX_SOCKET_PATH -u CMUX_DAEMON_SOCKET bun run build: passedgit diff --check: passedPre-commit CodeRabbit CLI review was attempted with the required three-minute bound; the service produced no findings before timeout. Manual diff review found no blocker.
Note
Medium Risk
Changes core lifecycle sweep ordering and crash-recovery eligibility via
user_killed; incorrect surface matching could suppress legitimate recovery, though tests cover recycled refs and UUID vs ref fallbacks.Overview
Fixes a sweep ordering race where missing surfaces were reconciled and crash recovery ran before cmux close-forensics could record operator intent, so UI tab closes looked like PTY deaths and agents with
crash_recover=truewere respawned.Close forensics now runs at the start of each sweep (awaited, not fire-and-forget at the tail). The sweep returns attributed
CloseForensicsEvent[]in addition to the emit count. NewmarkIntentionalSurfaceClosessetsuser_killed=trueon registry agents whosesurface_uuidmatches forensics with origintab_closeorworkspace_teardown, before reconcile/evict andrecoverCrashedAgents.close_surfacemarks matching managed agents terminal immediately after a successful close, using stable UUID when cmux exposes it and falling back to ref when observation is degraded—without marking stale UUID owners when a ref was recycled.Tests cover UI close vs respawn, forced close terminal intent, recycled-ref safety, and concurrent eviction handling.
Reviewed by Cursor Bugbot for commit 07a5ce0. Bugbot is set up for automated code reviews on this repo. Configure here.
Note
Mark user-closed panes as terminal before crash recovery in the agent engine
user_killed=truebefore reconciliation, preventing them from being treated as crash candidates.AgentEngine.runSweepOnce, replacing the previous non-awaited tail call.markIntentionalSurfaceClosesmethod filters forensics events by origin and updates theStateManagerand lifecycle registry for matching agents.close_surfacetool handler in server.ts also marks matching agentsuser_killed=trueon surface close, using stable UUID matching with a ref fallback.runCloseForensicsSweepin close-forensics.ts now returns the fullCloseForensicsEvent[]array alongside the emitted count so callers can act on individual events.Macroscope summarized 07a5ce0.
Summary by CodeRabbit
Bug Fixes
Tests