test/steer.test.ts runs in under 10 s, every test under 1 s, proving what it proves today #10233

test/steer.test.ts now runs in 1.8 s instead of 10.3 s (bun test test/steer.test.ts on this crew machine; four copies run side by side each took 1.7 s). The slowest test takes 0.38 s instead of about 9.7 s. Only test/steer.test.ts changed.

Before → after

What changed

• No timed waits in the fake agent. It used to sleep 0.4 s or 2.5 s and hope the steer arrived during that window. Now it keeps its first turn open until the test creates a gate file, and the test creates it only once the steer has been sent. That removes both the wait and the race; the only sleeps left are 10 ms polls.
• Short dispatcher tick. The crew starts with tickMs: 20 instead of waiting out its default 5 s tick.
• Setup built once. The app modules are imported at the top of the file instead of inside the test, which saved about 1.1 s. The forge, crew, agent, repository and crew membership are made once in beforeAll and torn down in afterAll.
• Invariants. Each test states its invariant in one line at its top. No test was deleted: the two prove different things, one about runClaude and one about how comments are routed to a running agent.
• Faster failure. The comment test now fails within about 250 ms with "Expected: 1, Received: 2" when a comment queues a second run. Before, it would only fail by timing out.

Each rewritten test still fails without the code it guards

• If runClaude's steer sender returns true without writing to the agent (src/crew/claude.ts:292), both tests fail with "Expected: 2, Received: 1" turns.
• If steerRun in src/crew/dispatch.ts:741 is short-circuited, the comment test fails in 257 ms with "Expected: 1, Received: 2" runs.

Left out

• Two seeding checks moved into beforeAll: adding the agent to the crew and committing the seed file. They now throw instead of asserting, so the expect count is 16 instead of 17, with the same invariants and nothing skipped.
• The keep_plan tool was not available in this run, so the issue has no plan.

Checks: bun scripts/gate.ts and bun x tsc --noEmit pass, and the 13 files scripts/fast.ts names pass (170 tests, 0 failures).

To try it: bun test test/steer.test.ts

🤖 Generated with Claude Code

Closes #10219

test/steer.test.ts now runs in 1.8 s instead of 10.3 s (bun test test/steer.test.ts on this crew machine; four copies run side by side each took 1.7 s). The slowest test takes 0.38 s instead of about 9.7 s. Only test/steer.test.ts changed.

Before → after

What changed

• No timed waits in the fake agent. It used to sleep 0.4 s or 2.5 s and hope the steer arrived during that window. Now it keeps its first turn open until the test creates a gate file, and the test creates it only once the steer has been sent. That removes both the wait and the race; the only sleeps left are 10 ms polls.
• Short dispatcher tick. The crew starts with tickMs: 20 instead of waiting out its default 5 s tick.
• Setup built once. The app modules are imported at the top of the file instead of inside the test, which saved about 1.1 s. The forge, crew, agent, repository and crew membership are made once in beforeAll and torn down in afterAll.
• Invariants. Each test states its invariant in one line at its top. No test was deleted: the two prove different things, one about runClaude and one about how comments are routed to a running agent.
• Faster failure. The comment test now fails within about 250 ms with "Expected: 1, Received: 2" when a comment queues a second run. Before, it would only fail by timing out.

Each rewritten test still fails without the code it guards

• If runClaude's steer sender returns true without writing to the agent (src/crew/claude.ts:292), both tests fail with "Expected: 2, Received: 1" turns.
• If steerRun in src/crew/dispatch.ts:741 is short-circuited, the comment test fails in 257 ms with "Expected: 1, Received: 2" runs.

Left out

• Two seeding checks moved into beforeAll: adding the agent to the crew and committing the seed file. They now throw instead of asserting, so the expect count is 16 instead of 17, with the same invariants and nothing skipped.
• The keep_plan tool was not available in this run, so the issue has no plan.

Checks: bun scripts/gate.ts and bun x tsc --noEmit pass, and the 13 files scripts/fast.ts names pass (170 tests, 0 failures).

To try it: bun test test/steer.test.ts

🤖 Generated with Claude Code