feat(fleet): implement F1-F7 dispatch-path fixes#196
Merged
NagyVikt merged 1 commit intoMay 18, 2026
Merged
Conversation
Direct implementation of all 7 findings (the fleet's own workers were
blocked by F5+F6, so this commit lands what the fleet would have).
- F1 (show-fleet.sh): dead_panes_report() emits JSON to stderr from
tmux #{pane_dead}; alerts at age >60s via firstseen markers.
- F2 (cap-probe.sh): CACHE_TTL_HEALTHY default 60s (was 300s),
CODEX_FLEET_CAP_CACHE_TTL env override, bringup-failure marker
zeroes the TTL for cold re-probe.
- F3 (full-bringup.sh): CODEX_FLEET_AUTO_WAKE=1 fires wake-prompt.sh
once at bringup tail before DONE banner.
- F4 (plan-watcher.sh): run_plan_validator() passes --allow-waves to
the validator (matching bringup); CODEX_FLEET_PLAN_VALIDATOR_FLAGS
env layers extra operator flags.
- F5 (force-claim.sh): dispatch() pre-checks pane_in_mode + Codex `›`
glyph + Working() heuristic; defers (does NOT consume the claim)
when pane not ready. FORCE_CLAIM_SKIP_READY_CHECK=1 escape hatch.
- F6 (test/codex-auto-submit-test.sh): integration smoke test that
spawns a 1-pane Codex worker, sends-keys a wake prompt, asserts
>=1 Colony claim within 90s. Currently fails (proves bug);
passes once the working submit-key sequence ships.
- F7 (full-bringup.sh + supervisor + test): CODEX_FLEET_AUTO_BYPASS=1
fires codex-first-launch-supervisor.sh once before auto-wake to
drain "Do you trust" / "External agent config" / "Press enter"
prompts. Smoke test test/first-launch-bypass-test.sh PASSES (live).
All scripts pass `bash -n`. Plan workspace + change tasks.md flipped
to completed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
1fa8839 to
b59dc3f
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Direct implementation of all 7 dispatch-path findings — follow-up to merged PR #189 (scaffold + supervisor + interval tune). The fleet's own workers were blocked by F5+F6 so this PR lands what the fleet would have produced.
Changes
scripts/codex-fleet/show-fleet.shdead_panes_report()emits JSON to stderr fromtmux #{pane_dead}; alerts at age >60s via/tmp/claude-viz/dead-pane-firstseen/markers.scripts/codex-fleet/cap-probe.shCACHE_TTL_HEALTHYdefault 60s (was 300s),CODEX_FLEET_CAP_CACHE_TTLenv override, bringup-failure marker zeroes the TTL.scripts/codex-fleet/full-bringup.shCODEX_FLEET_AUTO_WAKE=1default fireswake-prompt.shonce beforeDONE.banner.scripts/codex-fleet/plan-watcher.shrun_plan_validator()passes--allow-wavesto validator;CODEX_FLEET_PLAN_VALIDATOR_FLAGSenv override layers extra flags.scripts/codex-fleet/force-claim.shdispatch()pre-checks#{pane_in_mode}+ Codex›glyph +Working(…)heuristic before send-keys; defers (does NOT consume the Colony claim) when pane not ready.FORCE_CLAIM_SKIP_READY_CHECK=1escape hatch.scripts/codex-fleet/test/codex-auto-submit-test.sh(new)scripts/codex-fleet/full-bringup.sh+codex-first-launch-supervisor.sh(already in #189)CODEX_FLEET_AUTO_BYPASS=1default fires the supervisor once before auto-wake to drain "Do you trust" / "External agent config" / "Press enter" prompts.Plus
docs/fleet-telemetry-cases.mdandtest/first-launch-bypass-test.sh(PASSES live).Verification
bash -non all 8 modified/new scripts — passlib/plan-validator.sh openspec/plans/fleet-dispatch-fixes-2026-05-18/plan.json --allow-waves→ok:truetest/first-launch-bypass-test.sh→ PASS: all 3 prompt markers drained from live screentest/codex-auto-submit-test.sh→ currently FAILS (proves F6 bug); will pass once working submit-key sequence is identifiedfleet-dispatch-fixes-2026-05-18and assert >=4 Colony claims in 90s (operator-run; new fleet uses the fixed dispatch path)Env knobs introduced
CODEX_FLEET_AUTO_BYPASS1CODEX_FLEET_AUTO_WAKE1CODEX_FLEET_CAP_CACHE_TTL60CODEX_FLEET_PLAN_VALIDATOR_FLAGS""FORCE_CLAIM_SKIP_READY_CHECK0Telemetry attached
See
docs/fleet-telemetry-cases.mdfor live failure-mode evidence per finding.🤖 Generated with Claude Code