[mldangelo-oai]

Summary

Long error messages could prevent a scan from entering its failed state or lose the final cause when saved. Preserve complete diagnostics through scan failure persistence, Deep Scan validation retries, and Windows credential ACL subprocess errors.

Changes

• Remove the 2,400-character failure-message limit from the MCP schema and Python workbench, along with SDK and Deep Scan truncation before persistence.
• Preserve both the original scan failure and any result-publication failure, and include complete validator messages in discovery and reducer retries.
• Send Python workbench arguments through the existing stdin frame so long messages also survive operating-system command-line limits. Preserve full budget-completion warnings.
• Capture complete ACL subprocess stderr. Existing ACL validation, credential protections, terminal-state immutability, and saved errors on resume remain intact.

Testing

• Locked SDK and MCP dependencies verified with frozen offline installs.
• Python diagnostics: 9 new regressions, 10 existing focused failure/terminal-state tests, and 32 budget-completion cases passed.
• MCP coordinator, executor, smoke, and discovery/reducer diagnostic retry regressions passed.
• SDK saved-failure regressions passed (4 tests, including messages over 128 KiB), as did 5 workbench argument/payload cases and all 45 import/archive registration tests. ACL regressions passed (10 tests; 8 native Windows tests skipped on Linux).
• Required portable plugin checks passed: Ruff check/format, SDK build:ci, source compatibility, and all 9 source-check tests.
• SDK types and format passed.
• Full SDK suites passed with seeds 12345 and 3184034310: each run had 6,775 passing tests, 83 skips, and no failures across 230 files.
• Three independent native code reviews and verification passed with no remaining findings.

Risk and rollout

Longer messages now remain available in saved scan history and retry prompts. Existing truncated records are preserved on resume; immutable terminal records still reject conflicting updates. No dependency, CLI flag, duration, or archive-limit changes. Credential and path policies are unchanged. The ACL regression exercises the subprocess helper on Linux; native Windows behavior was not run locally.

Public disclosure review

• No customer, partner, prospect, or user identities, data, or identifying details are included.
• No credentials, personal data, private source, scan findings, or nonpublic links or tickets are included.
• I reviewed the branch name, title, description, commits, changes, comments, logs, screenshots, attachments, and links for public disclosure.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-23T10:02:47.704870Z	db4e987	PR opened
🔒 Security Review	✅ Completed 2026-10-23T10:04:05.915554Z	db4e987	PR opened

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[alandelong-oai]

Thanks for preserving the complete failure diagnostics through persistence and retries. No material findings at db4e9870f07f0817e71d1a82c80902a35be281da. Verified long-error retention, original/publication error composition, retry prompts, and large argument/payload framing with archive cancellation. Builds, types and formatting passed; 47 relevant Python cases passed in the exact-head hosted run. Local broader Python/MCP checks encountered scan-directory ownership restrictions, and native Windows behavior was not run locally. CI remains in progress; this is not a full-suite pass claim.