[AbirAbbas]

Summary

The PR gate's touched codeaf leg on workstreams/main hits the 15-minute package timeout (panic: test timed out after 15m0s), so every cmd/codeaf test after the cut goes unchecked. This is ENG-CI of the Workstreams plan.

Measured with a per-test JSON run of the whole package on the CI runner at dd749e133:

	time
whole package	1065 s (17.8 min), 3128 top-level tests
tests that predate the workstream work	1052.7 s; 60 tests of 5 s or more take 792 s (slowest: TestAppRelayOutageO1O2O3O4O5 93 s, TestAppRelayRestartP2UnknownUntilPeerReturnsThenActuallyOffline 66 s, TestC27NewChatByDefaultStaysQuiet 55 s, TestMemoryPausedJobLifecycle 39 s)
workstream tests	11.2 s, of which 10.04 s is one subtest: TestAppWorkstreamWireConformance/errors/busy waiting out SQLite's real 10 s busy timeout

The base branch's own run at 19f4df464 times out the same way, so the base alone is over the limit.

Changes

1. tests: the conformance busy case sets PRAGMA busy_timeout to 200 ms on the backend's own single store connection, reads it back and fails unless it took, then holds the same real writer on a second handle. The refusal still goes SQLITE_BUSY -> ErrBusy -> wire busy through the real store and dispatch; no synthetic error. Production keeps its 10 s. Measured after, same base: the busy subtest 10.04 s -> 0.22 s (passes), the workstream tests 11.2 s -> 1.8 s; with app: workstream agent prompt and tools (proposals, drafts, its own pages) #2289's agent tests also in, 42 workstream tests take 2.0 s.
2. ci: only the codeaf leg passes --timeout 25m to touched-verdict.sh (a single expression on the existing line, with a comment giving the measurement). tui3, session and rest pass no --timeout and keep the Makefile's 15m. A timeout is still red. Sharding cmd/codeaf failed in ci: touched packages runs in parallel legs and re-checks a failure before blaming the change #1734, so it is not sharded here. docs/rules/ci.md records the exception and that it goes once the suite is back under 15 minutes.

No t.Parallel() and no seeding changes: the rest of the workstream tests total about 1.2 s.

Validation contract

• (a) The busy case still gets the wire busy code from a real lock held by a second handle, through the backend's real store and error mapping. Test: TestAppWorkstreamWireConformance/errors/busy (assertions unchanged).
• (b) The backend's handle waits 200 ms, and the test fails if that is not in effect. Test: the read-back check in workstreamWireHoldWriter.
• (c) Production keeps the 10 s busy wait: no diff under internal/workspace/ or in any non-test Go file.
• (d) Only the codeaf leg gets 25 minutes: the one ci.yml line; the touched codeaf leg on this PR finishes without the timeout.
• (e) docs/rules/ci.md states the exception and why.

CI on this PR

touched codeaf ran the whole package in 1035 s (17.3 min) with --timeout 25m: no timeout. It is red only on failures also red on workstreams/main's own runs: TestAppChatRowsR2SavedTurnTime and TestAppChatStartupR3QuietRelease (attribution inconclusive, as on the base) and TestRecipeDoorWritesTheWorktreeWhenOnlyTheCheckoutIsHeld (already failing on the base). light gate is red only on the internal/session laws ratchet (TestNoRoadInTheTaskEngineHasMoreEndingsThanItsLedgerRow, TestNoRoadInTheMemoryEngineHasMoreEndingsThanTheCeiling), also red on the base. For comparison, the codeaf leg on workstreams/main at dd749e133 and 4dcd5150f timed out at 15m.

Left for the base's owners

The real fix is the base's long real-time waits (the 60 tests of 5 s or more, many of them exactly 10, 15, 20, 25 or 30 s). Once the suite is back under 15 minutes, drop the 25m line. Full check runs cmd/codeaf under the same 15m and is not changed here.

🤖 Generated with Claude Code