Repository navigation
Conversation
Keep per-role effort in the saved orchestration snapshot and dispatch Claude roles through the existing background runner. Preserve model defaults. Refuse incompatible upgrades before writes. Verify background cleanup with claude stop and session roster removal after live acceptance exposed survivors. Record independent review and automated/live validation in the history ledger. Quest/Co-Authored by Co-Authored-By: Codex <noreply@openai.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> in collaboration with KjellKod <kjell.hedstrom@gmail.com>
Preserve each role's last saved effort in the existing celebration payload so archived journals replay without consulting current defaults. Omit absent or malformed values and label settings as configured rather than effective. Quest/Co-Authored by Co-Authored-By: Codex <noreply@openai.com> in collaboration with KjellKod <kjell.hedstrom@gmail.com>
KjellKod
changed the title
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
There was a problem hiding this comment.
All reported issues were addressed across 42 files
Tip: cubic can generate docs of your entire codebase and keep them up to date. Try it here.
Re-trigger cubic
There was a problem hiding this comment.
0 issues found across 4 files (changes from recent commits).
Requires human review: Auto-approval blocked by 17 unresolved issues from previous reviews.
Re-trigger cubic
There was a problem hiding this comment.
0 issues found across 6 files (changes from recent commits).
Requires human review: Auto-approval blocked by 9 unresolved issues from previous reviews.
Re-trigger cubic
There was a problem hiding this comment.
All reported issues were addressed across 9 files (changes from recent commits).
Tip: Review your code locally with the cubic CLI to iterate faster.
Re-trigger cubic
There was a problem hiding this comment.
0 issues found across 6 files (changes from recent commits).
Requires human review: Adds per-role effort configuration with saved orchestration compatibility checks and Claude background dispatch under both orchestrators; end-to-end acceptance is pending and the upgrade/dispatch policies need human sign-off.
Re-trigger cubic
This branch was successfully deployed
1 participant
⭐ Why this matters
Configure reasoning effort separately for each Quest role, so planners, builders and reviewers can use the settings you choose for their work. Quest applies those settings during dispatch and preserves them in journals and celebrations for later review.
Summary
Add per-role effort configuration: set defaults in
.ai/allowlist.json, save the selected settings in each quest'sorchestration.json, and apply them whenever a role runs.The initial default is medium for Claude and Codex roles. Existing model assignments and review policy remain unchanged; medium is a starting policy, not a measured cost/quality optimum.
Warning
Existing allowlists need the new
effortmap. Incompatible saved quests require explicit reconfiguration or a new quest. The installer refuses incompatible upgrades before writes, including locally modified framework files that also changed upstream. See the upgrade guide.Changes
effort.<role>defaults and validation. Reviewer A/B settings remain independent. Resume uses the saved quest settings. Gemini effort pins are unsupported and reported explicitly; standalone/gptkeeps its existing scalar setting.claude stopand verifying session removal. Intentional human-input parking remains unchanged. Requires a Claude CLI supportingstop, tested with 2.1.286.Validation
Pending end-to-end acceptance
These two checks are still outstanding. They exercise the installed workflow, beyond the completed role-level tests below. Agents should drive them where runtime access permits; human involvement is limited to required approval gates or access. Use a separate disposable Git repository for each run, with a Python test command configured in its allowlist.
Install this PR branch from inside each disposable repository:
Record the installed revision and the orchestrator model/version. Use the normal supported orchestrator model, with authenticated Claude background execution and Codex access available. Set role effort defaults to planner
medium, builderhigh, and code reviewerslow; leave other valid role defaults in place. The differing values make accidental default reuse visible./quest "Add a greeting function that returns Hello, <name>! and a meaningful automated test. Keep the change minimal."Complete planning, required approvals, build, review and completion/archive. Confirm the saved role efforts and actual dispatch arguments match the configured values, the greeting test passes, canonical artifacts exist, and owned background sessions are cleaned up. Verify archivedorchestration.json, journal and celebration show the configured effort. Replay the generated journal with/celebrate <journal-path>and confirm the settings persist. Record evidence and any unavailable effective-effort metadata; do not substitute agent self-report for runtime evidence.$quest. Complete the same lifecycle and checks. Confirm Codex roles use native dispatch and Claude roles use the background runner, with explicit saved effort controls. After completion, change the repository's effort defaults and replay the generated journal with$celebrate <journal-path>; expected: historical configured effort remains unchanged. Confirm no missing/invalid pin silently falls back by testing the existing configuration validator against a disposable copy with one active role's effort removed; expected: a clear rejection before dispatch. Record the installed revision, dispatch evidence, test result, archive/replay result and cleanup result.A tiny greeting task is sufficient for this integration check. It does not establish an effort level's quality/cost benefit. Mark these boxes complete only after the full runs and evidence have been recorded.
Completed validation
Final commit
49efd16CI: 1,374 passed, 7 skipped. Configuration, security, formatting and Codex review checks passed.Optional-cache follow-up: 28/28 preflight cases passed, including successful probes with unwritable cache paths in both automatic and explicit background mode. Regression failed first; self-review completed before push.
October 5 review fixes: 1,381 tests passed before the final optional-cache follow-up, plus Black, generated-default, manifest and configuration checks. Every fix was self-reviewed before push; final independent cross-check found no actionable defects. Scoped checks cover saved transport recovery, invalid configuration/state, preflight cleanup, retained sessions, stale framework symlinks and missing-model effort display.
Completed by agents, no human repetition needed:
Journal follow-up: 167 focused tests passed, including seven new cases covering archive preservation, distinct reviewer settings, journal replay and missing/malformed effort. Black and whitespace checks passed; independent review found no actionable issues.
Unchanged-main baseline: 1,316 tests. Replacement suite before the journal follow-up: 1,356 passed.
Shell checks: orchestration 36/36, state 72/72, runtime 72/72, preflight 26/26, handoff contracts 3/3.
Black, generated-model defaults, configuration, manifest and whitespace checks.
Full installer acceptance: fresh install, incompatible-upgrade refusal without mutations, explicit recovery, custom configuration preservation and manifest backup.
Independent configuration, dispatch, installer and scope reviews; real Claude source review. Findings addressed.
Live Claude persona/tool/model/effort evidence, local and outside-in; Claude-led Claude builder at high effort; Claude-led Codex builder; controlled permission denial; canonical artifacts.
Fresh cleanup retest: session absent immediately and 44.1 seconds later without manual cleanup.
Boundaries: effective Codex model/effort was not independently reported. Native Codex selection is covered deterministically; the earlier native live smoke belongs to #179. No API bridge inference was used. Installer recovery did not launch a provider from the recovered installation.
Human attention: review the medium default and deliberate upgrade/resume policy. Full evidence and limitations: validation ledger. Design decisions and lessons: replacement plan.
▐▛███▜▌ ▝▜█████▛▘ ▘▘ ▝▝ Quest/Co-Authored by Co-Authored-By: Codex Co-Authored-By: Claude Opus 5.5 in collaboration with KjellKod