Borrowing it
Nothing to install: this file belongs to jakubsuplicki/codument. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/jakubsuplicki/codument/main/.agents/skills/review-work/SKILL.mdgit clone --depth 1 https://github.com/jakubsuplicki/codumentWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/jakubsuplicki/codument/review-work)<a href="https://agentmods.dev/skills/jakubsuplicki/codument/review-work"><img src="https://agentmods.dev/badge/skills/jakubsuplicki/codument/review-work/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/jakubsuplicki/codument/review-work"><img src="https://agentmods.dev/badge/skills/jakubsuplicki/codument/review-work.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00024 | $0.02093 |
| Opus 5 | $0.00012 | $0.01046 |
| Sonnet 5 | $0.00005 | $0.00419 |
| Haiku 4.5 | $0.00002 | $0.00209 |
Grade A, and why
review-work scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 91 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Review Work
Use this after a planned step has been implemented and before committing.
Adversarial pass (the independent reviewer)
AI must never be trusted to grade its own work. For a non-trivial diff, run an INDEPENDENT adversarial review before committing — a reviewer that assumes the change is wrong until a reproduction proves otherwise. codument review --require-review is the arbiter of "non-trivial": it requires the pass for more than one real change, any deletion, a config/data change, a risk touch, an ownership ambiguity, a module-level change, or anything beyond a single resolved symbol. A genuinely trivial diff skips this pass (the gate says so) and gets only the deterministic self-review below.
- Assemble the oracle.
codument review --bundle > .codument/review-bundle.json— the touched features' documented invariants, the tests that pin them, the diff, and the ownership/blast facts. It adds no new source of truth; it hands the adversary a contract to attack instead of an open-ended hunt. When a review of this same base is already recorded, the bundle scopes itself to what has moved since (scope: "delta") and carries the rest asalreadyReviewed+priorFindings;--fullforces the whole change set for a deliberate fresh attack. The bundle carries astamp— a digest of exactly what it handed over — and step 3 records it, so the attestation says what it was grounded in rather than only what it concluded. - Run the adversary against the bundle — independence by context, degraded gracefully.
- Subagent host (Claude): spawn a FRESH
adversarial-reviewersubagent fed ONLY the bundle — never your transcript or your reasoning, so it cannot re-anchor to the author's mental model. It readsgit diff <base>, attacks the invariants, writes a failing test for each real bug, and emits a findings JSON. - No-subagent host (Codex): run the same adversarial pass yourself against the bundle, with deliberately fresh eyes — apply the
adversarial-reviewermandate as if you had not written the code. Independence is weaker, but the deterministic confirm and the fingerprint-bound artifact are identical, so it is not theater.
- Subagent host (Claude): spawn a FRESH
- Record the verdict.
codument review --record <findings.json>writes the fingerprint-bound artifact. Copy the bundle'sstampinto the findings JSON asbundleStamp, so the record says which oracle it answered. Any later edit to a reviewed source — or to a finding's named test — auto-invalidates it, so you cannot review once and keep editing. A review that records no stamp is still accepted: it is reported on the verdict line, never refused, because refusing it would dead-end the first review of any diff and would be walked past by anyone willing to omit the field. - Enforce.
codument review --require-review. It RE-RUNS each finding's named test and blocks only on one that is genuinely red — never on a claimed status. Fix every confirmed (red-test) finding before committing; advisory findings are surfaced for your decision, never auto-blocking.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 91 lines · 24 tokens per session scan A ad7ad0eb0c49
review-work is a skill published in the GitHub repository jakubsuplicki/codument (47 stars, last pushed 18d ago), licensed Apache-2.0. It adds 24 tokens to every session and 2,093 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
review
Adversarial fresh-context review of an increment before it ships. Every finding cites path:line and is re-verified. Use when saying "review", "grill this", or "critique the implementation".
sw-review
Adversarial fresh-context review of an increment before it ships; every finding cites path:line and is re-verified. Use for "review this", "critique the implementation", "review before closing".
receiving-feedback
Use when receiving code review feedback or corrections - before implementing suggestions. Requires technical verification and reasoned evaluation, not blind agreement or performative responses.
atlas-review
Review code changes and capture learnings for CLAUDE.md by following the canonical review command doc.
rework-rate
Measure and interpret PR rework rate — the emerging 5th DORA metric.
review
Validate plans, execution, or PRs against wish criteria — returns SHIP / FIX-FIRST / BLOCKED with severity-tagged gaps.