Real Behavior Proof

Behavior addressed

MCP processes are now reaped unconditionally on session reset, even if the agent run doesn't terminate within the grace window. Prevents stale MCP processes from holding file locks (SQLite/LadybirdDB) across session boundaries.

Real environment tested

• OS: Linux x64 (Node.js v24.13.1)
• OpenClaw: fix/gateway-mcp-stale-process-cleanup branch (commit c2c8788)
• Date: 2026-06-16

Evidence after fix

1. Automated test suite (session reset cleanup):
pnpm test src/gateway/server.sessions.reset-cleanup.test.ts
Test Files 2 passed (2)
Tests 24 passed (24)

2. Build verification: pnpm build succeeds.

3. Code change (verified diff):
The fix moves retireSessionMcpRuntime() from inside the if (ended) branch to BEFORE waitForEmbeddedAgentRunEnd(), ensuring MCP processes are always cleaned up:

+  // Always reap MCP processes — even if agent run doesn't terminate
+  await retireSessionMcpRuntime({
+    sessionId: params.sessionId,
+    reason: "gateway-session-cleanup",
+  });
+
   const ended = await waitForEmbeddedAgentRunEnd(params.sessionId, 15_000);
Observed result after fix
MCP runtime is retired unconditionally on session reset
Stale MCP processes cannot survive session boundaries
SQLite/LadybirdDB file locks released properly for next session
Build and all 24 session reset tests pass
What was not tested
Active MCP connections during reset timeout (requires live MCP server setup)
Long-running gateway with repeated session resets under load

</details>

Best possible solution:

Land the lease-aware lifecycle design after refreshing onto current main and confirming that reset-required retirement survives reuse while active leases remain protected, then let the merge close the canonical process-leak issue.

Do we have a high-confidence way to reproduce the issue?

Yes at high source confidence: the timeout path and linked reports establish the leak mechanism, while the real stdio child and SQLite-lock harness exercise the affected resource lifecycle. I did not independently run a failing current-main checkout reproduction in this read-only review.

Is this the best way to solve the issue?

Yes. Required deferred retirement is narrower and safer than killing the MCP transport immediately, because it deterministically reaps stale children while preserving in-flight work until the final aggregate lease releases.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 45f36640dc3e.

Label justifications:

• P1: The fix addresses leaked MCP children that can retain locks or ports and break subsequent session workflows until process or gateway recovery.
• merge-risk: 🚨 availability: Incorrect cleanup ordering or lease accounting could either strand MCP processes or terminate transports still serving active agent work.
• rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🦞 diamond lobster and patch quality is 🐚 platinum hermit.
• status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (linked_artifact): The PR provides exact-head linked evidence for a real stdio MCP child and real SQLite write lock, showing the child remains alive while leased and exits with lock release after the final lease; private environment details should remain redacted in any additional artifacts.
• proof: sufficient: Contributor real behavior proof is sufficient. The PR provides exact-head linked evidence for a real stdio MCP child and real SQLite write lock, showing the child remains alive while leased and exits with lock release after the final lease; private environment details should remain redacted in any additional artifacts.

PR surface:

Source +180, Tests +361. Total +541 across 20 files.

What I checked:

• Lifecycle entry point: Session reset/delete cleanup requests required MCP retirement before aborting or waiting, then keeps a watcher for runs that outlive the 15-second grace period. (src/gateway/session-reset-service.ts:335, 6aed7bf40780)
• Retirement ownership: The MCP manager distinguishes reusable deferred retirement from reset/delete retirement that remains armed across late runtime creation or reuse, while aggregate active leases prevent premature disposal. (src/agents/agent-bundle-mcp-manager.ts:302, 6aed7bf40780)
• Final-lease completion: Materialized runtime disposal releases its lease and asks the manager to complete pending retirement, allowing the child process to exit only after the last run, view, or request lease is gone. (src/agents/agent-bundle-mcp-materialize.ts:26, 6aed7bf40780)
• Replacement-aware waiting: The run waiter supports an unbounded lifecycle watch and rechecks embedded and reply-backed replacements under the same session ID rather than settling on the first handle transition. (src/agents/embedded-agent-runner/runs.ts:714, 6aed7bf40780)
• Real resource proof: The regression harness starts a real stdio MCP child, holds a real SQLite BEGIN IMMEDIATE lock, verifies the child remains alive while leased, and verifies final release terminates it and frees the lock. (src/agents/agent-bundle-mcp-runtime.test.ts:2049, 6aed7bf40780)
• Gateway race coverage: Focused tests cover reset/delete timeout behavior, required retirement ordering, overlapping cleanup, late runtime reuse, and reply-backed or replacement run draining. (src/gateway/server.sessions.reset-cleanup.test.ts:208, 6aed7bf40780)

Likely related people:

• steipete: Authored the branch's database-lock proof, replacement-run handling, overlap-race fixes, lifecycle typing, and lazy-loading refinements across the central gateway and MCP manager paths. (role: recent lifecycle contributor; confidence: high; commits: 67861c873a07, 83656fea4384, 664dafcf84cf; files: src/gateway/session-reset-service.ts, src/agents/agent-bundle-mcp-manager.ts, src/agents/embedded-agent-runner/runs.ts)
• allan0509: The bundled MCP session-runtime reuse behavior appears to originate in the merged work referenced as pull request 55090, making its reuse and rollover invariants directly relevant to this cleanup change. (role: feature introducer; confidence: medium; files: src/agents/agent-bundle-mcp-manager.ts, src/agents/agent-bundle-mcp-materialize.ts)

• 🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
• 🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
• 🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
• 🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
• 🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
• 🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
• 🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

• reviewed 2026-07-16T16:26:55.751Z sha 666ebeb :: needs maintainer review before merge. :: none
• reviewed 2026-07-17T06:55:16.578Z sha 2702c01 :: needs maintainer review before merge. :: none
• reviewed 2026-07-17T07:50:03.213Z sha ac125df :: needs maintainer review before merge. :: none
• reviewed 2026-07-17T08:18:39.586Z sha 050f9e3 :: needs maintainer review before merge. :: none
• reviewed 2026-07-17T08:27:16.055Z sha 757207b :: needs maintainer review before merge. :: none
• reviewed 2026-07-17T08:46:17.747Z sha d312a8b :: needs maintainer review before merge. :: none
• reviewed 2026-07-17T08:58:01.108Z sha 87c8e92 :: needs maintainer review before merge. :: none
• reviewed 2026-07-17T09:11:03.245Z sha 28fa82f :: needs maintainer review before merge. :: none