fix(mcp): clarify validation and assurance diagnostics - #602
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: 9cccc57b7e
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".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 75ec2ef2c2
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".
When an installed capability has a valid pattern longer than 1 KB and an input fails that pattern, the rendered expectation exceeded CapabilityDiagnostic.expected's 1024-character limit, causing an uncaught Pydantic ValidationError instead of returning a structured INVALID_REQUEST result. Truncate the rendered pattern value while retaining the validator identity.
Bound large schema constraint values (pattern, enum, const) in schema_violation_details to prevent multi-megabyte diagnostic responses. Strings are truncated to 1024 chars; non-string values are JSON-rendered and truncated if the rendered form exceeds 1024.
Co-authored-by: morluto <morluto@users.noreply.github.com>
3db2feb to
7943b27
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7943b27ff2
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".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 33da5f9503
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".
3 participants
Summary
VERIFIEDassurance from an input, premise, factorization, or related artifact to a model-derived final claimRoot cause
Focused Jacobian visibility evaluations exposed two interface problems. Out-of-range inputs reported only a generic schema mismatch, so agents could not see the advertised bound and retried indirectly. Separately, one trajectory treated independently verified supporting factorizations as verification of a derived GCD claim. Verification records are claim-bound, so that assurance transfer was unsound.
Overlap audit
Before publishing, I checked recent commits and all open/draft PRs for validation-diagnostic and assurance-guidance changes.
This PR therefore complements rather than duplicates those changes.
Evaluation
A focused 10-run evaluation (five server-only and five full-skill) completed without infrastructure failures. Totient, Möbius, radical, and abstention cases behaved correctly in both configurations. The exact-GCD verification case failed safely: because no installed
integer.gcd.verifycapability exists, the agent reportedCOMPUTEDrather than falsely claimingVERIFIED.Generated evaluation configs and transcripts are intentionally excluded from this PR.
Validation
git diff --check