Skip to content

docs(#832): prohibit unsubstantiated verification claims in review output - #835

Open
fullsend-ai-coder[bot] wants to merge 1 commit into
mainfrom
agent/832-prohibit-unsubstantiated-verification
Open

docs(#832): prohibit unsubstantiated verification claims in review output#835
fullsend-ai-coder[bot] wants to merge 1 commit into
mainfrom
agent/832-prohibit-unsubstantiated-verification

Conversation

@fullsend-ai-coder

Copy link
Copy Markdown
Contributor

Summary

  • Add a "No freeform verification sections" formatting rule in the pr-review SKILL.md output format section (step 7), prohibiting the review agent from claiming verified properties beyond what static diff analysis can confirm (e.g., "Verified: ✅", "zero X remain", "delivery chain verified"). Prior findings no longer present must use "not observed in current diff" rather than "verified resolved."
  • Add a "Challenge verification claims" step (step 5) to the challenger sub-agent procedure, instructing it to challenge and remove unsubstantiated verification text in the aggregated output — not just findings.

Context

During the 13+ review iterations on PR #211, the review bot produced freeform text asserting "zero references to deleted files remain" and "GH_TOKEN delivery chain verified" at commits where both were provably false. The prescribed output format does not include a "Verified" section, and the challenger sub-agent was not scoped to challenge such claims.

Test plan

  • Verify the SKILL.md formatting rules section now includes the prohibition
  • Verify the challenger.md procedure section now includes step 5
  • Confirm hack/lint-agent-docs-test.sh passes
  • Confirm make check-bundle passes (no script changes)
  • Monitor next 10 review runs for absence of blanket verification claims

Closes #832

Post-script verification

  • Branch is not main/master (agent/832-prohibit-unsubstantiated-verification)
  • Secret scan passed (gitleaks — 1a3086f5512b3700ab8e081c689cc7586c9af20f..HEAD)
  • PR body secret scan passed (gitleaks — no-git)

…tput

Add explicit prohibition in the pr-review SKILL.md output format
section against freeform verification sections that claim properties
verified beyond what static diff analysis can confirm. The review
agent cannot verify reference integrity, credential flows, or
runtime behavior — those require CI/runtime validation. Prior
findings no longer present should use "not observed in current diff"
rather than "verified resolved."

Expand the challenger sub-agent procedure to challenge verification
claims in aggregated output, not just findings. The challenger now
removes unsubstantiated verification text (e.g., "all references
verified", "delivery chain confirmed", "zero X remain").

Closes #832
@fullsend-ai-coder
fullsend-ai-coder Bot requested a review from a team as a code owner August 17, 2026 12:41
@fullsend-ai-review

fullsend-ai-review Bot commented Aug 17, 2026

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 12:42 PM UTC · Completed 12:56 PM UTC

Commit: 4761001 · View workflow run →

@fullsend-ai-review

Copy link
Copy Markdown

Review

Findings

Medium

  • [protected-path] skills/pr-review/SKILL.md, skills/pr-review/sub-agents/challenger.md — Both changed files are under skills/, a protected path. The PR links to issue Review agent should not produce unsubstantiated verification claims #832 which explicitly authorizes these modifications, and the description explains the rationale. Human approval is always required for protected-path changes, regardless of context.

Labels: PR modifies review agent skill files under skills/pr-review/

@fullsend-ai-review fullsend-ai-review Bot added requires-manual-review Review requires human judgment review-agent documentation Improvements or additions to documentation labels Aug 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation ready-for-review requires-manual-review Review requires human judgment review-agent

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Review agent should not produce unsubstantiated verification claims

0 participants