fix: reduce review noise and clarify agent communication - #1657
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![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: 37bfe17102
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".
1ef0408 to
2f91f39
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2f91f391c1
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".
Require demonstrated consequences before retaining review findings and let document review apply proven corrections that fulfill existing decisions. Keep unresolved user choices and explicit edit restrictions protected, and make callers resolve supported technical work before presenting a handoff. Add behavioral fixtures and record cross-host evaluation results, including remaining Fable calibration limits. Full suite: 3,915 tests passed; release and plugin validation passed.
Preserve review calibration alongside upstream agent cleanup and blocked handoff handling. Keep both behavioral scenario sets. Move task visibility unchanged into the required scope reference to preserve the Codex prompt budget, with an always-loaded pointer. Full suite: 3,913 passed; the two prompt-budget failures were corrected and passed on targeted rerun. Release and strict plugin validation passed. Independent source review confirmed both parent contracts remain intact. Behavioral replay was not repeated for this mechanical reconciliation and verbatim relocation into an already-required reference.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: fc3000110e
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".
Use the main-reachable pre-change contract for the completed-review handoff scenario so fresh checkouts can run its pre arm after squash merge. Historical evaluation reports retain their original refs. Validation: catalog 16 passed. Full suite 3913 passed, 2 failed; both affected test files passed on targeted rerun (100 tests).
Merge origin/main (#1664 docs site, #1549 Compound Packs) into the review calibration branch. One textual conflict in skills/ce-code-review/references/scope.md: main extended the Stage 1b signals paragraph with `declared_packs`; this branch appended a Task Visibility section after it. Both are kept, main's paragraph first. The other nine shared files (guides, dispatch-reviewers, finish-review, doc-review subagent template, review contract test) auto-merged into disjoint hunks and were read for semantic fit: pack-rule contradictions enter as ordinary findings and pass the same admission and edit-authority gates this branch tightened, and nothing on main reintroduces the collection or residual gate language this branch removed. Release validation passed. Full suite: 4,008 passed; 13 timeouts were all in ce-work-unit-workspace-fallback, which neither side touched, and that file passes in isolation (15/15). Strict plugin validation is left to CI (no `claude` binary here). Co-authored-by: Kieran Klaassen <kieranklaassen@users.noreply.github.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 968c79007a
ℹ️ About Codex in GitHub
Codex has been enabled to automatically 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 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Keep upstream attributable-terminal collection and local review-noise filtering together. Preserve cleanup, incomplete-review protection, and separate host waiting and result delivery. Align collection assertions with the upstream contract while retaining admission checks. Validation: 4,022 tests passed; release metadata and strict plugin validation passed. Independent source comparison verified both parents' intent. No new behavioral replay for this mechanical reconciliation of existing collection contracts.
Remove the synthesis exception that reinserted preference-only settled conflicts into the primary report. Keep genuine defects actionable and preserve the boundary against reversing settled decisions during local apply. A bounded Fable baseline reproduced advisory resurrection; Fable and Astra post-change routing discarded the preference while retaining the ownership defect. All 4022 tests pass; release and plugin validation pass.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6d4a5b4218
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".
Replace preference-only FYI examples with verified advisory benefits and reconcile caller guides with the existing outcome-based completion gate. Preserve synthesis authority and caller behavior. Bounded Fable and Astra rendering checks retained only the useful FYI; the pre-change Fable run already did so. All 4022 tests and release/plugin validation pass.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6cf05fc0c7
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".
2 participants
Summary
Reviews should ask users about choices that need their judgment and explain those choices plainly. Agents now investigate findings, discard weak or settled concerns, and resolve authorized technical work before returning it to the user.
Document review can apply proven corrections needed to carry out concrete decisions already made in the document. It preserves explicit edit restrictions and unresolved product choices. Findings from another model need independent local support before silent application. Choosing a remedy and having permission to apply it remain separate checks.
Planning and implementation callers reassess returned findings instead of forwarding them unchanged.
ce-povsupports consequential questions selectively and returns missing context to its caller. Shared writing guidance removes internal jargon while preserving technical terms, evidence, and qualifications.Validation
ce-workcorrected the ownership guard and passed the relevant 403/404 tests on both models. Simplification-before-review ordering was not fully established by those runs.mainand the review fixes, all 4,022 tests passed. Release metadata and strict plugin validation passed.Detailed scenarios, iterations, and limitations include the earlier unsuccessful runs.
Security Disclosure
Expands default document-edit authority to proven corrections required by existing decisions. Explicit edit restrictions, protected user choices, and independent-review requirements remain. No new shell execution, credential handling, or dependencies.
Agent Disclosure