FE-1125: Reconcile executor run driver

[kostandinang]

Summary

• Adds execute_orchestrate: drives a run end-to-end (created → run-local promotion_prepared) instead of calling each execute_* step by hand.
• src/executor/orchestrate.ts — a pure RunScheduler (linearScheduler) plus a generic drive() loop; run.json status is the loop state, and per-slice readiness comes from completion facts.
• Halts without advancing if a step cannot execute. Host promotion stays off the driven chain (separate accepted surface).
• ready() returns a set (length-1 today) so a future PetriScheduler drops in without reshaping the loop (D102-L).
• Clarifies the AgentRunnerPort fail-closed message so a launch without subagent deps is diagnosable, not mistaken for an unbuilt feature (surfaced during driver testing).

Verification

• npm run verify

[cursor]

PR Summary

Medium Risk
The new driver chains git worktree, agent, test, and run-local land steps; behavior is heavily tested but failures mid-run now depend on halt semantics instead of explicit per-step calls.

Overview
Adds execute_orchestrate, a single tool/agent surface that advances an executor run from created through run-local promotion_prepared instead of invoking each execute_* lifecycle step manually.

The core change is drive() in orchestrate.ts: a loop reads run.json, asks a pure RunScheduler (linearScheduler today) for the next ready step, and dispatches to the existing step functions via ExecutionPorts. Slice progression uses completion facts (completedSliceIds), not only the coarse status enum. The loop halts without advancing when a step returns unchanged status (failed tests, land failures, no changes). Host promotion/land stays off the driven path—the scheduler returns no steps at promotion_prepared. The scheduler’s set-returning ready() contract is shaped for a future Petri-style driver.

Wiring: Pi extension registration, executor allowed-tools policy, execute_status ported-tools list, and a large orchestrate.test.ts suite (end-to-end drive, manual-crank parity, halt cases).

Also clarifies AgentRunnerPort’s fail-closed message when subagent deps are missing (diagnosable vs “not implemented”). Minor test tweak for review-set render-honesty (nodes.2.body vs wildcard).

^{Reviewed by Cursor Bugbot for commit 70f9687. Bugbot is set up for automated code reviews on this repo. Configure here.}

[kostandinang]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• FE-1125: Reconcile executor run driver #285 👈 (View in Graphite)
• FE-1118: Reconcile host executor promotion #282 : 1 other dependent PR (#284 )
• FE-1112: Reconcile run-local executor promotion #279
• FE-1111: Reconcile executor agent runner #278
• FE-1109: Reconcile executor sandbox worktrees and test runner ports #275
• FE-1089: Reconcile executor with alpha branch #274
• next

This stack of pull requests is managed by Graphite. Learn more about stacking.

[kostandinang]

Merge activity

• Jul 3, 5:42 PM UTC: A user started a stack merge that includes this pull request via Graphite.
• Jul 3, 5:42 PM UTC: Graphite couldn't add this pull request to the GitHub merge queue because a downstack PR FE-1089: Reconcile executor with alpha branch #274 failed to merge.