ci: give the deep review CodeRabbit-grade formatting, a human blurb, and 2x turns

[allxsmith]

What

Reworks the deep-review prompt in claude-review.yml so its output is structured, scannable, and human — closer to CodeRabbit — and doubles its turn budget.

Why

On #243 the deep review read as a jumbled paragraph, posted one terse summary line, and missed two real bugs CodeRabbit caught with deeper per-file analysis. Three fixes:

1. Structured inline findings. Every finding is now Markdown: bold title + severity (🔴 Critical / 🟠 Major / 🟡 Minor) + category, then What / Why it matters / Fix (a ```suggestion block or fenced diff), long traces folded into <details>, and a ```mermaid diagram when a flow picture beats prose.
2. A real summary review. A severity table + an Overall read + a short, human "California surfer" blurb — technical enough for the fix agent, readable enough for you. Built in a file and posted via gh api -F body=@file so the tables survive; the machine-parsed <!-- claude-deep-review --> marker stays line 1.
3. Double the budget — --max-turns 60 → 120. Richer, verified, well-formatted findings need room, and the extra depth should help it stop missing code-level bugs.

Not in this PR

The bounded 3-round reviewer↔bestaxbot rebuttal on refuted findings — that needs gate/state-machine changes (routing refuted deep-review threads to the reviewer, a rebuttal-round counter separate from the fix-iteration cap, and a clean contested→human exit). It's coming as a dedicated follow-up so a bug there can't destabilize the loop. The same formatting + surfer tone will be applied to the re-review/verify reviewer in that PR.

Scope

claude-review.yml only — prompt text + one turn number. No logic, permission, or model changes. YAML validated.

---

Generated by Claude Code

Summary by CodeRabbit

• Chores
  • Improved automated review output formatting for more consistent and structured feedback.
  • Expanded review processing limits, allowing deeper analysis before finalizing results.
  • Standardized the final review summary layout, including clearer findings, overall assessment, and a short recap.

[coderabbitai]

Walkthrough

This PR modifies the Claude deep review GitHub Actions workflow, increasing the --max-turns limit from 60 to 120 and rewriting the deep-review prompt to require a structured summary output written to a file, with a marker line, findings table, and narrative sections.

Changes

Claude review workflow configuration

Layer / File(s)	Summary
Max-turns and prompt output rewrite .github/workflows/claude-review.yml	--max-turns raised from 60 to 120; deep-review prompt rewritten to write summary to /tmp/deep-review.md starting with <!-- claude-deep-review --> marker, a findings table, Overall section, and surfer blurb, with defined zero-findings behavior, replacing the prior brief summary format.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Possibly related PRs

• allxsmith/bestax#223: Both PRs modify the same --max-turns setting in claude-review.yml's deep-review configuration.
• allxsmith/bestax#218: Both PRs increase --max-turns limits for Claude GitHub Actions workflows.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description explains the change well, but it omits most required template sections like packages, issue refs, type of change, and checklist.	Reformat the PR description to match the template and add the missing sections: affected package(s), related issues, type of change, checklist, screenshots/demos, and additional context.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately summarizes the workflow formatting change and doubled turn budget.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/deep-review-formatting

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Preview Deployment

Preview URL: https://8cb66b8e.bestax.pages.dev

[coderabbitai]

🧹 Nitpick comments (1)

.github/workflows/claude-review.yml (1)

140-168: 🎯 Functional Correctness | 🔵 Trivial | 💤 Low value

Template shows an instructional line above the marker, contradicting the "very first line" rule.

Line 166 says the literal marker must be the VERY FIRST line of the posted body, but the template at line 146 places (write this to /tmp/deep-review.md, filling in the values) above the marker on line 147. If the model treats the whole indented block literally, that parenthetical could end up as the actual first line of /tmp/deep-review.md. The downstream gate only checks contains(...) so this wouldn't break the loop, but it's worth tightening for clarity/robustness.

♻️ Suggested tweak

-               then post it:
+               then post it (the fenced block below is the exact file
+               content — do not include this parenthetical in the file):

-                 (write this to /tmp/deep-review.md, filling in the values)
                  <!-- claude-deep-review -->
                  ## Deep review — <N> finding(s)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In @.github/workflows/claude-review.yml around lines 140 - 168, The deep-review
template is inconsistent with the “marker must be the VERY FIRST line” rule
because the parenthetical instruction appears before the marker in the quoted
block. Update the template in claude-review.yml so the generated
/tmp/deep-review.md content starts immediately with the literal <!--
claude-deep-review --> marker, and move any instructional text outside the body
template or below the marker. Use the deep-review body template around the
summary section as the place to verify the marker placement.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In @.github/workflows/claude-review.yml:
- Around line 140-168: The deep-review template is inconsistent with the “marker
must be the VERY FIRST line” rule because the parenthetical instruction appears
before the marker in the quoted block. Update the template in claude-review.yml
so the generated /tmp/deep-review.md content starts immediately with the literal
<!-- claude-deep-review --> marker, and move any instructional text outside the
body template or below the marker. Use the deep-review body template around the
summary section as the place to verify the marker placement.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 13b91277-462d-470f-a4a3-dadb44f8ff21

📥 Commits

Reviewing files that changed from the base of the PR and between b1ff2dd and 3faf509.

📒 Files selected for processing (1)

• .github/workflows/claude-review.yml

[github-actions]

🎉 This PR is included in version 5.3.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀

[github-actions]

🎉 This PR is included in version 3.2.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀

[github-actions]

🎉 This PR is included in version 1.0.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀

[bestax-release-bot]

🎉 This PR is included in version 1.0.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀