fix(claw-pilot): stopAccount before restart on config change (PILOT-188)

[matthew-pilot]

What failed

onAccountConfigChanged in channel-plugin-api.ts stopped the transport but left the account in PilotLifecycle.accounts, so the subsequent startAll → startAccount call threw "already running". The error was caught by startAll and routed to scheduleRetry, causing live config changes (e.g. adding a peer to the allowlist) to be silently lost until the gateway restarted.

Why this fix

Added PilotLifecycle.stopAccount(accountId) — a single idempotent method that stops the transport + pipeline + drain timer, then removes the account from the map. Replaced the manual acct.transport.stop() in onAccountConfigChanged with stopAccount() so startAll can re-create the account cleanly.

Verification

• npm run build (tsc): clean, no errors
• npm test (vitest): 230/230 tests pass, 28 test files, 4.37s
• 2 files changed, +14/−4 lines

Deferred

Bug 1 from the same ticket (handshakeTrustAutoApprove dead config — defined in schema but never consumed) is confirmed but not addressed here. Removing it touches 5 files (config.ts, openclaw.plugin.json, 3 test files) — the operator should split it to a follow-up or handle manually.

Closes PILOT-188

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[matthew-pilot]

🦾 Matthew PR Check — #2 PILOT-188

Status

• State: OPEN · MERGEABLE · CLEAN
• CI: 4/4 green — plugin (Node) ✅, iOS/ClawChat ✅, codecov/patch ✅, Snyk ✅
• Canary: N/A (claw-pilot is a plugin repo, no canary harness)
• Jira: PILOT-188 — QA/IN-REVIEW · assigned Teodor Calin · last updated 11:25 UTC

Summary

2 files (+14/−4) — adds PilotLifecycle.stopAccount() (lifecycle.ts) and replaces the manual transport stop in onAccountConfigChanged (channel-plugin-api.ts). All CI green. No canary required.

[matthew-pilot]

🔍 Matthew PR Explain — #2 PILOT-188

What changed and why

plugin/src/channel-plugin-api.ts:138 — the onAccountConfigChanged handler previously called acct.transport.stop() directly, which stopped the transport but left the account entry in PilotLifecycle.accounts. When startAll re-ran and hit startAccount for the same accountId, it threw "already running" — the error was caught and routed to scheduleRetry, so live config changes (e.g. adding a peer to the allowlist) were silently lost until gateway restart.

The fix replaces that two-line manual stop with a single call to lifecycle.stopAccount(accountId), which correctly removes the account from the live map.

plugin/src/lifecycle.ts:278-291 — new stopAccount(accountId) method. Idempotent (early-returns if the account isn't found). Clears the drain timer, stops the pipeline, awaits transport stop, and then — critically — deletes the account from this.accounts in a finally block. This ensures the map is always cleaned up even if one of the stop steps throws.

Diff summary: 2 files, +14/−4 lines. No new dependencies, no API surface change.