Harness step 1: executor defects and deletion
Orientation
Step 1 of mo-harness-in-mo. The findings are in
audit/evidence/2026-09-19/fable-overnight-review/README.md (E1, E2, E4, E5,
E6). The Python under toolchain/harness/executor/ is being cut down before
most of it moves to Mo, so fix only what stays or what the machine runs on, and
delete what is duplicate. Worker: a fresh Claude Opus session, bypass
permissions, own worktree. The lead owns the wiki, audit and acceptance.
Write scope
toolchain/harness/executor/** and toolchain/harness/provider/**/*.py only.
No Mo source, no compiler, no mo-wiki/, audit/, HANDOFF.md. Do not delete
or edit any existing evidence/, attempts/ or sources/ directory: they are
history. No machine (mo-executor-r01), Docker or /opt changes; the machine
run is the lead’s. No push. Never use tr.
Parts
- E1. A
docker rm -fthat times out must not power off the machine. Distinguish a cleanup timeout (report cleanup unconfirmed, leave evidence, retry within a stated bound) from a proven containment failure (the only case that may power off). Failing test first. - E2. A refusal before
claim()returns its structured refusal; noFileNotFoundError. The failing test issues the refusal as the first call to a fresh workspace. - E4, E6. A truncated or undecodable
state.jsonis quarantine, neverrefusal/completed.files.Refusalstops subclassingValueError, or the catch narrows, so programming errors surface. State writesfsyncbefore rename. Failing tests first. - E5. The machine side validates
nameagainst^mo-executor-[0-9a-f]{32}$before it reaches systemd. Failing test first. - Delete duplicates. The six guarded-runner copies
(
recovery/run_guarded.py,application/run_attempt.py,workspace_http/run_guarded.py,provider/run.py,provider/auth/run.py,provider/bridge/run.pyandexecute.py) become one helper of about 60 lines that records the child’s real return code. The per-run probe and inventory scripts (red_lifetime.py,review_baseline.py,ipc_deadline_probe.py,controller_deadline.py,live_frontend.py,frontend_control.py,live_owner.py,observe_regression.py,diagnose_runtime.py,inspect_machine.py,distribution_diff.py,final_inventory.py,final_receipt.py,smoke.py) become oneinventory.py, or go if nothing calls them. The four copies of the exec-bootstrap string become one. Check every caller before deleting; say what each deleted file was for in the commit message.
Do not restructure the live suites (live.py, local.py,
test_workspace_live.py, selftest.py, controls.py): that is step 2 and
another worker’s. Keep them importable and passing.
Numbers and done when
All local Python tests pass (84 before; report the new count) under
toolchain/bench/step36/guard.py, with the real summary line and exit code.
Report non-blank non-comment Python lines before and after, per directory,
with the counting command. Small commits as yourself with a Co-Authored-By
line naming your model. Final report to the lead: commits, decisions the brief
did not cover, anything you chose not to delete and why, what the lead must
re-run on the machine.
Result
Accepted 19 Sep 2026, 9:03 AM ET (decision-log). The worker’s report is committed with its work; the lead’s runs are in audit/evidence/2026-09-19/fable-lead-verification/.