Freegle / Iznik / 27841

72%

master: 73%

Pull #1124

circleci

fix(monitor-fsm): stop fix-agent timeouts on Go/error-handling bugs

Two of three fix attempts on Discourse 9941 (login "we don't know that
email" during a DB outage) were auto-closed because the fix agent timed
out before finishing verification/self-review, and the third was a
partial fix. Root causes in the delegate prompts:

1. STEP 4 makes the agent run the FULL layer suite (`go test ./...`,
   full Laravel) inside its wall-clock budget. On the Go layer that is
   ~9 min per run; done alongside the reproduce + fix runs it blows the
   budget. Now: run ONLY the targeted test(s); the full regression suite
   runs in CI (which the monitor already re-checks and follows up on).

2. AssertFlip demands a failing test, but for error-handling bugs the
   failure is awkward to induce (a DB error, a timeout), so the agent
   burns its budget trying to reproduce an outage. Now: seam the failing
   call behind a function var/interface and stub an error in the test —
   a valid AssertFlip without physically reproducing the failure.

Added one authoritative "TIME BUDGET & TEST STRATEGY" block to all three
fix prompts (monolithic parallel path, decomposed IMPLEMENT_FIX, and
REPRODUCE_BUG). tsc build + 275 FSM vitest tests green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013YqwEGSmzjUbHM3QinYAxX

13215 of 17412 branches covered (75.9%)

Branch coverage included in aggregate %.

138234 of 191672 relevant lines covered (72.12%)

37.96 hits per line