Skip to content
Back to search

review-skills

Two-model debate review of PRs and MRs, plus a babysitter that works the review rounds. Skills for any coding agent.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/amelnagdy/review-skills with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/amelnagdy/review-skills'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/amelnagdy--review-skills/badge.svg)](/listings/amelnagdy--review-skills)

HTML

<a href="/listings/amelnagdy--review-skills"><img src="/api/listings/amelnagdy--review-skills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 172 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # review-skills
  2. [![tests](https://img.shields.io/github/actions/workflow/status/amElnagdy/review-skills/tests.yml?branch=master&label=tests)](https://github.com/amElnagdy/review-skills/actions/workflows/tests.yml)
  3. [![skills.sh](https://www.skills.sh/b/amElnagdy/review-skills)](https://www.skills.sh/amElnagdy/review-skills)
  4. [![License](https://img.shields.io/github/license/amElnagdy/review-skills)](LICENSE)
  5. **Two models argue over a pull request before anything is posted. You keep the merge.**
  6. `debate-review` has a main reviewer read the PR, a second reviewer try to knock its findings down,
  7. and the main reviewer make the final call. One review with inline comments lands on the PR, posted
  8. from your own `gh`, `glab`, or `az` account. `babysit-pr` then works the rounds: verifies each finding, fixes
  9. the blockers, replies in-thread, resolves, and re-triggers the next review. Both skills run through
  10. whatever coding agent you already drive (Claude Code, Codex, Cursor, OpenCode, Grok, and others) and
  11. the model subscriptions you already pay for.
  12. ```bash
  13. npx skills add amElnagdy/review-skills
  14. ```
  15. Then ask your agent:
  16. ```text
  17. Use $debate-review on https://github.com/owner/repo/pull/123
  18. Use $debate-review --local on this repo before I open a PR.
  19. Use $babysit-pr on PR 123 until it is ready to merge.
  20. ```
  21. ```mermaid
  22. flowchart LR
  23. P["PR or MR"] --> M["Main reviewer<br/>lane review-main"]
  24. M -->|"findings"| D["Debate reviewer<br/>lane review-debate"]
  25. D -->|"confirm / refute / downgrade / add"| F["Main reviewer<br/>final call"]
  26. F --> R["One review<br/>inline P0 / P1 / P2"]
  27. R --> B["$babysit-pr<br/>verify, fix, reply, resolve"]
  28. B -->|"push, re-run"| M
  29. ```
  30. ## The skills
  31. | Skill | Job | Never does |
  32. | --- | --- | --- |
  33. | [`debate-review`](skills/debate-review/SKILL.md) | Reviews a GitHub PR, GitLab MR or Azure DevOps PR with two models in sequence and posts non-approval inline comments plus a summary. Findings that survive the debate are posted as agreed; findings the second model refuted but the main reviewer kept are posted as contested, with both sides' reasoning. | Edit code, approve, request changes, post twice for the same head sha. |
  34. | [`babysit-pr`](skills/babysit-pr/SKILL.md) | Harvests every reviewer thread on the PR (debate-review, Codex, Greptile, any bot), checks each finding against the code, fixes what is real, replies in-thread with evidence and attribution, resolves, and re-runs the review for the next round. Reports when the PR meets the merge gate. | Merge, resolve a thread it did not answer, reply as anyone other than "model on behalf of user". |
  35. ## Requirements
  36. - Node 18 or newer.
  37. - Git 2.31 or newer for Azure DevOps (`--config-env` keeps the access token out of command arguments).
  38. - `gh` (GitHub), `glab` (GitLab, including self-hosted) or `az` (Azure DevOps) logged in to an account
  39. that can comment on the PR. Reviews and replies post as that account. Azure DevOps needs no `az`
  40. extension: the script uses `az rest` against the REST API.
  41. - [delegate-skills](https://github.com/amElnagdy/delegate-skills), which dispatches the reviewer
  42. models through the `*-delegate` relays, read-only.
  43. - Two delegate lanes named `review-main` and `review-debate`. Create them once with `$delegate-setup`:
  44. ```text
  45. Use $delegate-setup to add two lanes: review-main on claude at high effort and review-debate on codex at high effort.
  46. ```
  47. Pick two different implementers. The debate is only worth something when the second model does not
  48. share the first one's blind spots. Keep these lanes for the reviewer; if you already use a `debate`
  49. lane for plan debates, leave it alone. Only implementers whose relay supports `--read-only` are
  50. accepted, so the reviewers cannot touch your tree.
  51. Reviewers run inside a disposable checkout of the PR head (or a `--local` snapshot). After the
  52. diff is computed, project agent config such as `.claude`, `.codex`, `.cursor`, `.opencode`,
  53. `opencode.json`, and `.mcp.json` is deleted from that checkout so a PR cannot execute hooks or
  54. start MCP servers on the reviewer machine. The commit diff still includes those files. `--local`
  55. never modifies your working tree; `--check` also leaves the real tree alone and warns if those
  56. paths are present.
  57. Then confirm both reviewers start on this machine, in about half a minute:
  58. ```bash
  59. node <skill-dir>/scripts/review-pr.mjs --check
  60. ```
  61. ## What a review looks like
  62. Each finding is one inline comment, anchored to the lines it is about, rendered as a forge alert so
  63. the colour reads before the words:
  64. | Level | Meaning | Rendered as |
  65. | --- | --- | --- |
  66. | P0 | blocking, security | `[!CAUTION]` |
  67. | P1 | blocking, anything else | `[!WARNING]` |
  68. | P2 | non-blocking | `[!NOTE]` |
  69. The review body carries a level count table, who reviewed, and a short summary. Each comment and the
  70. body start with an HTML marker (`<!-- debate-review ... -->`) that `babysit-pr` uses to recognise the
  71. threads, since they are posted from your account rather than a bot's. One review per head sha: a
  72. re-run on the same push exits with code 3 instead of posting again. Templates and the marker format
  73. are in [`comment-format.md`](skills/debate-review/references/comment-format.md); the JSON contract
  74. between the three passes is in [`schema.md`](skills/debate-review/references/schema.md).
  75. The reviewers are prompted for precision, not volume: blocking bugs with a concrete trigger and wrong
  76. result, spec and standards violations quoted against the rule, and little else. The second reviewer
  77. can only refute a finding when it can point at the code that makes it impossible. Findings below the
  78. confidence floor are dropped before anything is posted.
  79. ## GitHub, GitLab and Azure DevOps
  80. | | GitHub | GitLab | Azure DevOps |
  81. | --- | --- | --- | --- |
  82. | Review posting | `gh api`, PR review with inline comments | `glab api`, MR discussions with diff positions | `az rest`, one comment thread per finding plus a closed summary thread |
  83. | Target URL | `/pull/<n>` | `/-/merge_requests/<n>` | `/_git/<repo>/pullrequest/<n>`, on `dev.azure.com` or `*.visualstudio.com` |
  84. | Spec source (`#123`) | issue | issue | work item |
  85. | Alert colours | yes | 17.10+ | no, alerts fall back to plain quotes |
  86. | Thread harvest (babysit-pr) | GraphQL review threads + REST reviews | discussions + notes | not implemented yet |
  87. | Reply and resolve (babysit-pr) | verified live | implemented, not yet verified on a live instance | not implemented yet |
  88. | Bot author detection | reliable (`Bot` type) | only when the instance exposes `author.bot`; debate-review threads are found by marker either way | n/a |
  89. Azure DevOps has no single review object, so one debate-review is N inline threads plus one closed summary
  90. thread carrying the marker. `--force` and the "already reviewed" check read the same marker back off
  91. the PR's threads, so a re-run on an unchanged head still exits 3.
  92. ## Local preview
  93. `--local` reviews the files on disk (committed, uncommitted, and untracked, honoring `.gitignore`)
  94. against a base branch. It never calls a forge CLI. Reviewers see a throwaway snapshot, not your
  95. working tree; agent config is stripped from that snapshot the same way as for a PR worktree.
  96. `--dry-run` still needs a live PR; it only skips the post. The two flags do not combine. Local
  97. snapshots reject non-UTF-8 Git paths instead of silently changing their bytes.
  98. | Invocation | Source | Forge | Post |
  99. | --- | --- | --- | --- |
  100. | `--local` | Working tree snapshot | No | No |
  101. | `<pr> --dry-run` | Live PR | Yes | No |
  102. | `<pr>` | Live PR | Yes | Yes |
  103. ## Run it by hand
  104. Your agent normally runs these for you. They are here for testing, CI, or when there is no agent in
  105. the loop. `<skill-dir>` is the directory containing the skill's `SKILL.md`.
  106. ```bash
  107. node "<skill-dir>/scripts/review-pr.mjs" --local # working tree; print, no forge
  108. node "<skill-dir>/scripts/review-pr.mjs" <pr-url | number> --dry-run # print, do not post
  109. node "<skill-dir>/scripts/review-pr.mjs" <pr-url | number> # post
  110. "<babysit-skill-dir>/scripts/threads.sh" <number> # harvest one round as JSON
  111. ```
  112. `review-pr.mjs --help` lists the flags: `--local` to review the working tree with no forge, `--main` /
  113. `--debate` to override the lanes for one run, `--contested post|drop`, `--min-confidence`, `--timeout`
  114. (default 30 minutes per reviewer), `--force` to post again on the same head, `--keep` to leave the
  115. temporary worktree or snapshot clone. Every run leaves its briefs, raw model output, and the three JSON
  116. documents under `~/.cache/debate-review/<owner>__<repo>/<N>/<head>/` so a surprising review can be
  117. traced back to the pass that produced it.
  118. ## How this relates to the sibling repos
  119. - [delegate-skills](https://github.com/amElnagdy/delegate-skills) is the transport. `review-pr.mjs`
  120. reads your lanes and calls the matching relay; this repo has no model code of its own.
  121. - [guard-skills](https://github.com/amElnagdy/guard-skills) are review lenses for code an agent just
  122. wrote. The reviewers here cannot load them (the relays run the reviewer with skills disabled, on
  123. purpose), and they are tuned for a different job than Clean Code checks. Where a guard fits is the
  124. babysitter's own fixes: `babysit-pr` runs the matching guard on a fix before pushing it, when one is
  125. installed.
  126. ## Development
  127. ```bash
  128. node --test test/*.test.mjs
  129. ```
  130. The babysit tests run `threads.sh` end to end against fake `gh` and `glab` binaries in
  131. `test/fixtures/babysit/`. The debate-review tests drive the Azure forge functions against a fake
  132. `az` in `test/fixtures/azure/`. Both suites run without network.
  133. ## License
  134. [MIT](LICENSE)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.