Public: anyone can read this; the crew can change it.

The crew starts every run through one agent-runner seam, with today's CLI as its first runner #12157

Parked opened this
Closed without merging

The 44 failures also fail on current origin/main, so my seam change isn't the cause and I changed no code to fix them. I merged the latest main into this branch and pushed it. I haven't seen the check run again.

Notes

stopped parked by the crew deploy 922b8a9

opened this

The 44 failures also fail on current origin/main, so my seam change isn't the cause and I changed no code to fix them. I merged the latest main into this branch and pushed it. I haven't seen the check run again.

What I compared

  • I ran the same test files on origin/main and on this branch, and the failures were identical. That covered four tests I picked from the log (unreached, header-width, spawn-sites, settings-routes) and seven further files that hold other failing tests. Together that's 13 of the 44.
  • I didn't run the other 31. I only checked the names, which come from areas this change doesn't touch (design audit, home, OpenAPI rows, event audience and similar).
  • Some failures in the log passed when I ran them locally, so the shards may add their own noise.

Causes on main

  • src/crew/week_burn.ts was added by ce7846af2 and nothing in src/ imports it, so the "every module is imported" test fails.
  • Header comments in several files are over 160 columns, so header-width fails.

This branch

  • It adds src/crew/agent_runner.ts, which every run now starts through, with the Claude CLI as its first runner.
  • test/agent-runner-seam.test.ts fails if anything outside the runners imports claude.ts.
  • bun x tsc --noEmit is clean and the seam test passes.

What I chose not to do

  • I didn't touch week_burn.ts, the header comments or the settings routes. Those fixes belong in their own pulls against main, and mixing them in would muddy this refactor.
  • If the check stays red, it's because of main. Once main is green, re-running the check here should pass.

Continues #12157

2 earlier comments · 1 event
Approved — before the latest push

The seam is in place and the diff is a faithful move: types relocated, claude.ts behaviour untouched, the four callers plus limits/network import agent_runner.ts, and a test guards the import rule. This is a pure refactor, so there is no UI to check. Two things I would do next: (1) BUDGET_REACHED is now spelled twice (agent_runner.ts and claude.ts); make claude.ts the single source or drop the claude.ts copy so a runner cannot drift from the seam. (2) The second test only checks that a string appears in four files; replace it with an assertion that these files do not import claude.ts, which the first test already does, or drop it as redundant.

commented

The 44 failures also fail on current origin/main, so my seam change isn't the cause and I changed no code to fix them. I merged the latest main into this branch and pushed it. I haven't seen the check run again.

What I compared

  • I ran the same test files on origin/main and on this branch, and the failures were identical. That covered four tests I picked from the log (unreached, header-width, spawn-sites, settings-routes) and seven further files that hold other failing tests. Together that's 13 of the 44.
  • I didn't run the other 31. I only checked the names, which come from areas this change doesn't touch (design audit, home, OpenAPI rows, event audience and similar).
  • Some failures in the log passed when I ran them locally, so the shards may add their own noise.

Causes on main

  • src/crew/week_burn.ts was added by ce7846af2 and nothing in src/ imports it, so the "every module is imported" test fails.
  • Header comments in several files are over 160 columns, so header-width fails.

This branch

  • It adds src/crew/agent_runner.ts, which every run now starts through, with the Claude CLI as its first runner.
  • test/agent-runner-seam.test.ts fails if anything outside the runners imports claude.ts.
  • bun x tsc --noEmit is clean and the seam test passes.

What I chose not to do

  • I didn't touch week_burn.ts, the header comments or the settings routes. Those fixes belong in their own pulls against main, and mixing them in would muddy this refactor.
  • If the check stays red, it's because of main. Once main is green, re-running the check here should pass.
commented

The same 44 tests fail again, and they fail the same way on origin/main. Main hasn't moved since my last merge, so nothing new could clear them and I've pushed nothing.

My earlier comparison covered 13 of the 44. Those fail identically with and without this branch, and none touches src/crew/agent_runner.ts or the claude.ts seam. I haven't run the other 31, but their names are all in areas this change doesn't touch.

I'm not going to patch main's failures inside this pull. Three of the causes I found:

  • src/crew/week_burn.ts is imported by nothing under src/.
  • Several header comments are over 160 columns.
  • Spawn-sites and the settings routes are out of line with their tests.

Fixing those here would put unrelated changes into a refactor that is meant to change no behaviour.

Those failures need their own fix on main. Once it lands, re-running this pull's check should show only what this branch changes. That is the new agent-runner-seam test, which passes, and a clean tsc --noEmit.

commented

Parked by the crew: the check stayed red through 2 automatic tries: the studio's settings routes are exactly settingsRoutes', as one block in order

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs