Area
Other
What are you trying to accomplish?
Keep actionable CodeRabbit findings that refer outside the current diff visible to the PR gate, while still giving authors a legitimate way to resolve or acknowledge them without creating an empty commit.
What prevents this today?
The #1094 design counts outside-diff findings while a related unresolved bot thread exists. That avoids making immutable review-body text permanently sticky, but findings reported only in a review body can lose a durable resolution signal. Conversely, making the count unconditional can block a PR forever because the author cannot edit that review body.
#1175 improves revalidation when the status comment changes, but it does not by itself define the acknowledgement identity or who may clear it.
What should OpenCodex do?
Define a stable finding identity and a trusted resolution signal. The gate should:
- retain an outside-diff finding until its exact identity is resolved or acknowledged;
- accept a documented maintainer/bot acknowledgement path;
- re-evaluate from trusted base code;
- avoid running PR-head workflow code on review events;
- make the active and resolved finding counts explainable in the gate comment.
Example usage or interface
CodeRabbit reports finding F outside the patch
-> gate records F as active
-> author fixes the underlying file or a maintainer acknowledges F
-> trusted gate observes the resolution marker for F
-> F becomes resolved without an empty commit
The marker could be a bot-managed thread state or a maintainer command; the important part is that it is stable, scoped to F, and evaluated by trusted-base automation.
Alternatives or workarounds
- Always count review-body findings: durable but potentially impossible for authors to clear.
- Count only unresolved inline threads: clearable, but review-body-only findings may disappear.
- Re-run on review events: can violate the trusted-base workflow boundary.
Additional context
Related work: #1094 and #1175. Making the outside-diff count unconditional without an acknowledgement contract would make some findings impossible for an author to clear, so that implementation is intentionally not proposed.
Checks
Area
Other
What are you trying to accomplish?
Keep actionable CodeRabbit findings that refer outside the current diff visible to the PR gate, while still giving authors a legitimate way to resolve or acknowledge them without creating an empty commit.
What prevents this today?
The #1094 design counts outside-diff findings while a related unresolved bot thread exists. That avoids making immutable review-body text permanently sticky, but findings reported only in a review body can lose a durable resolution signal. Conversely, making the count unconditional can block a PR forever because the author cannot edit that review body.
#1175 improves revalidation when the status comment changes, but it does not by itself define the acknowledgement identity or who may clear it.
What should OpenCodex do?
Define a stable finding identity and a trusted resolution signal. The gate should:
Example usage or interface
The marker could be a bot-managed thread state or a maintainer command; the important part is that it is stable, scoped to F, and evaluated by trusted-base automation.
Alternatives or workarounds
Additional context
Related work: #1094 and #1175. Making the outside-diff count unconditional without an acknowledgement contract would make some findings impossible for an author to clear, so that implementation is intentionally not proposed.
Checks