[irvingpop]

What changes

• Status heading with a status light.
• One alert callout that holds the next step.
• Red CAUTION callout for every fail.
• Findings table, worst first. Duplicates on the same file, side, line and effect merge into one row.
• One-row checks strip.
• Collapsed Waiver and full record sections.
• Spec updated with decision P3-5.

Why

The old comment was hard to judge quickly. The owner chose this design from two candidates.

Safety

Markers, override line and size limits are unchanged. Model text is escaped, so it cannot forge headings, callouts or the result line (probed in review).

Tests

• cd review && uv run pytest tests/unit -q -> 2631 passed, 2 deselected
• cd review && uv run pytest tests/unit -m release -q -> 2 passed, 2631 deselected

Rendered examples

Example: PR 39 comment

🔴 FAIL: OC Claude review

[!CAUTION]
One high-severity malicious finding blocks this PR. A maintainer can waive it.
Next step for the author: fix and push, or ask a maintainer to post the waiver line (under "Waiver" below).

Severity	Effect	Where	Finding	Found by
🔴 High	Blocks, waivable	src/app.py:10	Job impersonates the review gate check name with a fake pass. See the inline comment.	Review (high)
🟠 Medium	Does not block	src/app.py:10	Workflow spoofs the required review check name. See the inline comment.	Security (medium), Supply chain (medium)

File checks	Review	Security	Supply chain	Dependency review	Coverage
✅ ok, 2 signals	✅ ran, 1 finding	✅ ran, 1 finding	✅ ran, 1 finding	✅ ok	✅ complete

_{Run 1000, attempt 1 for head aaaaaaa into main. The required workflow in the merge box is the result that counts. This bot never approves a PR.}

🔑 Waiver: how a maintainer clears this

This PR fails for waivable reasons only. A person with write or admin access, who is not the PR author, posts this exact line as a new comment and replaces <reason> with the reason. Then they re-run the newest run of the required workflow through the GitHub REST API (POST .../actions/runs/{run_id}/rerun). Do not edit the comment afterwards. Do not close and reopen the PR to apply the waiver.

/oc-review override aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa cccccccccccccccccccccccccccccccccccccccc <reason>

A waiver covers the code at this exact head and merge base. Any new push needs a new waiver.

📋 Full review record: reasons, checks, signals, finding details, model summaries, token usage

Reviewed snapshot

Head SHA	Merge base SHA	Base branch	Run
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa	cccccccccccccccccccccccccccccccccccccccc	main	1000, attempt 1

Reasons the gate failed

Reason	Waivable?	Detail
Blocking finding	✅ yes	review: malicious high at src/app.py:10

Checks (code, no model)

Check	Result	Detail
Hard blocks	✅ none
Coverage	✅ complete	files in comparison 1, files collected 1, trees complete
Dependency changes	✅ none
Dependency review job	✅ success

Signals (context only, never block)

• ⚠️ Sensitive path changed: workflow file, in .github/workflows/x.yml
• ℹ️ First-time contributor: author association NONE

Dependency changes

None.

Finding details

Model output from untrusted input. Check it before you act on it. Verified means only that the quoted text is at that place. It does not mean the claim is true.

Blocking is decided by code from category and severity, never by the model. A finding in a blocking category blocks only at high or critical severity.

Each row above merges findings of several passes at the same file, line and gate effect.

Pass	Severity	Category	Confidence	Citation	Posted	Model title
Review	high	malicious	high	✅ checked	inline	Job impersonates the review gate check name with a fake pass.
Security	medium	security	high	✅ checked	inline	Workflow spoofs the required review check name.
Supply chain	medium	supply chain	high	✅ checked	inline	Workflow job spoofs the review gate check name.

Model summaries

Model output from untrusted input. Check it before you act on it. Verified means only that the quoted text is at that place. It does not mean the claim is true.

Pass	Status	Model	Summary
Review	✅ ok	claude-opus-5-5	Looked at the change.
Security	✅ ok	claude-opus-5-5	Looked at the change.
Supply chain	✅ ok	claude-opus-5-5	Looked at the change.

Token usage

Pass	Input	Output	Cache write	Cache read	Attempts
Review	12000	900	11000	0	1
Security	12000	900	11000	0	1
Supply chain	12000	900	11000	0	1
Total	36000	2700	33000	0	3

Known Minor, not fixed

Hard block data shows in three tables. A PR with many hard blocks reaches the overflow notice sooner (fails closed).

🤖 Generated with Claude Code

[github-actions]

OC Claude review: fail

Result for run 40629688283, attempt 1.
Computed for head SHA 897700826789bbf8f9ee01b3467e244a4f30c802, merge base 2c63f49b461feac74c2c70c4f86ca90040df6a0a, base ref main.
The result that counts is the required workflow in the merge box.

Reasons

• job-result (not waivable): dependency-review: failure
• dependency-review (not waivable): job result failure
• dependency-review (not waivable): step outcome failure
• pass-records (not waivable): pass review failed: authentication failed
• pass-records (not waivable): pass security failed: authentication failed
• pass-records (not waivable): pass supply-chain failed: authentication failed
• prefilter-hard-block (not waivable): bidi-control in review/tests/unit/test_publish.py

Prefilter (code, no LLM)

Hard blocks: 1.

• bidi-control: bidi control characters U+202E, in review/tests/unit/test_publish.py, lines 884 885 903 909
  Signals: 2.
• long-line: lines longer than 500 characters, in docs/claude-review-design.md, lines 540 574 588 589 590 593 594 982
• ai-directed-text: pattern ai-reviewer, in review/src/oc_review/publish.py, lines 982
  Coverage: complete (files in comparison 5, files collected 5, trees complete).

Dependency delta

No dependency changes.

Findings per pass

Model output from untrusted input. Check it before you act on it. Verified means only that the quoted text is at that place. It does not mean the claim is true.

review: failed, authentication failed

No verified findings.

security: failed, authentication failed

No verified findings.

supply-chain: failed, authentication failed

No verified findings.

Token usage

• review: no usage recorded
• security: no usage recorded
• supply-chain: no usage recorded
• total: no usage recorded

Workflow run. The result that counts is the required workflow in the merge box.