The crew starts every run through one agent-runner seam, with today's CLI as its first runner #12210

Every crew run now starts through one seam, src/crew/agent_runner.ts, with today's Claude CLI as its first runner. Two spawn-list tests fail, but the cause is already on main, not this pull.

What changed

• agent_runner.ts defines the runner: start a run with a brief, stream its steps, and report spend, pauses and end, as plain data in and out. The Claude runner is registered there.
• claude.ts implements it with no change in behaviour. The argv, env and outcomes are the same as before. This is a review claim: I did not run a before/after comparison of the argv and env.
• runner.ts, dispatch.ts, one_shot.ts and conversation.ts import the seam and none of them imports claude.ts. leftovers.ts, limits.ts, network.ts and run-cell.ts were repointed to the seam as well. These four are not on the issue's file list, and the diff for each is only two to four lines.
• test/agent-runner-seam.test.ts fails if any module under src/ other than the runner implementations imports claude.ts. It also checks that the four callers import the seam.
• This run I merged the latest main into the branch and pushed. No existing test was changed.

Checks

• scripts/gate.ts and tsc --noEmit are clean.
• I ran the seam, spawn-sites, spawn-pipes, crew-one-shot, crew-sandbox and crew-cancel-group tests: 46 pass, 2 fail.
• The two failures, in test/spawn-sites.test.ts and test/spawn-pipes.test.ts, come from src/data/query.ts. Main's commit cc205a800 added a process start there and did not list it in either test. It is outside this pull's files, so I left it for a separate fix.
• scripts/fast.ts found the diff too broad to name tests (1994 of 3035 files), so I picked the tests above myself. I did not run the full suite; the forge's check does that on push.

Not done

• No second runner and no provider selection. That belongs to the follow-ups under #12068.

Closes #12122

Every crew run now starts through one seam, src/crew/agent_runner.ts, with today's Claude CLI as its first runner. Two spawn-list tests fail, but the cause is already on main, not this pull.

What changed

• agent_runner.ts defines the runner: start a run with a brief, stream its steps, and report spend, pauses and end, as plain data in and out. The Claude runner is registered there.
• claude.ts implements it with no change in behaviour. The argv, env and outcomes are the same as before. This is a review claim: I did not run a before/after comparison of the argv and env.
• runner.ts, dispatch.ts, one_shot.ts and conversation.ts import the seam and none of them imports claude.ts. leftovers.ts, limits.ts, network.ts and run-cell.ts were repointed to the seam as well. These four are not on the issue's file list, and the diff for each is only two to four lines.
• test/agent-runner-seam.test.ts fails if any module under src/ other than the runner implementations imports claude.ts. It also checks that the four callers import the seam.
• This run I merged the latest main into the branch and pushed. No existing test was changed.

Checks

• scripts/gate.ts and tsc --noEmit are clean.
• I ran the seam, spawn-sites, spawn-pipes, crew-one-shot, crew-sandbox and crew-cancel-group tests: 46 pass, 2 fail.
• The two failures, in test/spawn-sites.test.ts and test/spawn-pipes.test.ts, come from src/data/query.ts. Main's commit cc205a800 added a process start there and did not list it in either test. It is outside this pull's files, so I left it for a separate fix.
• scripts/fast.ts found the diff too broad to name tests (1994 of 3035 files), so I picked the tests above myself. I did not run the full suite; the forge's check does that on push.

Not done

• No second runner and no provider selection. That belongs to the follow-ups under #12068.