test/landing.test.ts runs in under 10 s, every test under 1 s, proving what it proves today #10255

test/landing.test.ts now builds the forge, its crew and the agent's pull once in beforeAll. Its one test drops from 1.58 s to 0.08 s. It still proves the same invariant, now written as one line at the top of the test: closing a pull while its author's agent run is still going cancels that run with "the pull closed", deletes the run's working copy of the repository only after the run has ended, and queues nothing behind it.

Timings (crew sandbox, bun test test/landing.test.ts, per-test time from the junit reporter):

The file total barely moves. Most of it is loading the app's modules (~0.7 s) and starting the crew (~0.4 s), which now happen in beforeAll and not inside the test. Cutting those further would mean changing src/, which is outside this issue's files.

What changed:

• Imports are static at the top of the file instead of await import(...) inside the test.
• Everything up to the open pull moves to beforeAll: the app, the servers, the crew (still ticking every 50 ms), alice, the agent, the repository, the fork, the branch and the pull. afterAll stops the crew and the servers, closes the app and deletes the data directory.
• The test body starts at the comment that sends the agent back to its pull. The comment, the cancellation, the clone checks, the log check and the empty queue are exactly as before.
• The polling helper checks every 10 ms and gives up after 5 s (was 20 s). The test's timeout is 10 s (was 30 s).
• No test was deleted and nothing is skipped. The test makes 8 expect calls instead of 9: the addToCrew check is now a throw in setup.
• The fake agent still runs sleep 60. That is how long the pretend agent "thinks", not a wait in the test: the crew kills its whole process group when the run is cancelled.

Shown failing without the code it guards:

• With the cancel loop for a closed pull removed from src/crew/dispatch.ts: the run's end: did not happen in 5000 ms; saw runs=[["feedback","running",""]] clone=true.
• With the clone removal after the run commented out in src/crew/runner.ts: the clone's removal: did not happen in 5000 ms; saw runs=[["feedback","cancelled","the pull closed"]] clone=true.

Already missing before this change: neither the original test nor this one fails if the clone is deleted before the run ends (forcing the call at src/crew/dispatch.ts:911 even while a run is still working). The fake agent makes no git call, so it never notices its clone going. I left that alone, since the issue asks to keep what the test proves today; it would need its own issue.

Checks: bun scripts/gate.ts passes, bun x tsc --noEmit is clean, and the 13 test files scripts/fast.ts named all pass (169 tests). The keep_plan tool wasn't available in this run, so the plan isn't posted on the issue.

🤖 Generated with Claude Code

Closes #10240

test/landing.test.ts now builds the forge, its crew and the agent's pull once in beforeAll. Its one test drops from 1.58 s to 0.08 s. It still proves the same invariant, now written as one line at the top of the test: closing a pull while its author's agent run is still going cancels that run with "the pull closed", deletes the run's working copy of the repository only after the run has ended, and queues nothing behind it.

Timings (crew sandbox, bun test test/landing.test.ts, per-test time from the junit reporter):

The file total barely moves. Most of it is loading the app's modules (~0.7 s) and starting the crew (~0.4 s), which now happen in beforeAll and not inside the test. Cutting those further would mean changing src/, which is outside this issue's files.

What changed:

• Imports are static at the top of the file instead of await import(...) inside the test.
• Everything up to the open pull moves to beforeAll: the app, the servers, the crew (still ticking every 50 ms), alice, the agent, the repository, the fork, the branch and the pull. afterAll stops the crew and the servers, closes the app and deletes the data directory.
• The test body starts at the comment that sends the agent back to its pull. The comment, the cancellation, the clone checks, the log check and the empty queue are exactly as before.
• The polling helper checks every 10 ms and gives up after 5 s (was 20 s). The test's timeout is 10 s (was 30 s).
• No test was deleted and nothing is skipped. The test makes 8 expect calls instead of 9: the addToCrew check is now a throw in setup.
• The fake agent still runs sleep 60. That is how long the pretend agent "thinks", not a wait in the test: the crew kills its whole process group when the run is cancelled.

Shown failing without the code it guards:

• With the cancel loop for a closed pull removed from src/crew/dispatch.ts: the run's end: did not happen in 5000 ms; saw runs=[["feedback","running",""]] clone=true.
• With the clone removal after the run commented out in src/crew/runner.ts: the clone's removal: did not happen in 5000 ms; saw runs=[["feedback","cancelled","the pull closed"]] clone=true.

Already missing before this change: neither the original test nor this one fails if the clone is deleted before the run ends (forcing the call at src/crew/dispatch.ts:911 even while a run is still working). The fake agent makes no git call, so it never notices its clone going. I left that alone, since the issue asks to keep what the test proves today; it would need its own issue.

Checks: bun scripts/gate.ts passes, bun x tsc --noEmit is clean, and the 13 test files scripts/fast.ts named all pass (169 tests). The keep_plan tool wasn't available in this run, so the plan isn't posted on the issue.

🤖 Generated with Claude Code