fix(orchestration): align remora with pilotfish-codex v1.6.0 - #21
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: aa37fd7451
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
aa37fd7 to
4771f3b
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4771f3b3c4
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
4771f3b to
f00c0de
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f00c0debdc
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
f00c0de to
ee7b356
Compare
Correct the stale pilotfish-codex v1.3.2 semantic basis shipped in v0.1.18 against the official v1.6.0 tag. The session policy now classifies execute, explore_then_plan, and co_discover work; records bounded route signals and discovery budgets; treats review_intent as an explicit turn-scoped preference; and adds decision-card and direction-checkpoint contracts. Require the actual main-session AskUserQuestion tool for material decisions whenever Claude Code exposes it, instead of allowing a plain-text imitation. Tool-unavailable sessions fail closed with PAUSED_NEEDS_USER, while pending authority gates stop execute work before local action or delegation. Preserve Remora's Claude-native transport, role models, tool allowlists, isolation, approval boundaries, and canonical verifier verdicts. Mandatory security and plan review gates outrank optional intent routing, unavailable named reviewers fail closed, and direction guidance stays advisory beneath CONFIRMED, REFUTED, or INCONCLUSIVE. Bump the package to 0.1.19, update pinned install references and architecture notes, and add focused policy regressions. Verified with 96 unit tests plus install, bootstrap, compile, shell-syntax, agents, dry-run, packaging, binding-parity, and archive checks.
ee7b356 to
f37f99f
Compare
1 participant
What changed
79852a749ed8594982ae8a0b529147a5586dd548)execute,explore_then_plan, andco_discoverrouting with bounded discovery budgets and decision cards that call the actualAskUserQuestiontool when exposedreview_intentwhile keeping mandatory security, approval, release, and external-action gates authoritativedirection_checkpointhandling beneath the existingCONFIRMED/REFUTED/INCONCLUSIVEverifier contractBoundaries preserved
plan-verifier; security-sensitive Plans requiresecurity-reviewerfirstVerification
python3 -m unittest tests.test_remora -v— 96 passedmake package— passed full unittest, install/bootstrap, compile, shell syntax, agents, dry-run, and packaging gates2facb02fb04a9c436f1a2c295171875a38ad0f02ba9738adb4c7e9ff41ea1319git diff --checkclean; only the intended 9 files changedUpstream evidence
pilotfish-codex v1.6.0004fa9b9d81930dfd48b65c58622828e381afb8379852a749ed8594982ae8a0b529147a5586dd548