adversarial-review — Claude Code skill: what it does, how to install
adversarial-review subjects fresh builds to hostile inspection using isolated reviewer contexts and mandates empirical reproduction on the running system.
no license ★ 406 updated 2026-06-25
npx skills add syahiidkamil/Software-Engineer-AI-Agent-Atlas --skill adversarial-review -a claude-codeWhat adversarial-review does
Standard code inspections look for reasons to approve changes. This skill starts with the hostile premise that new code is broken and charges reviewers with proving it. While /code-review simply parses diffs statically without starting the software, this workflow executes targets directly to validate flaws.
Reviewers run inside isolated subagent contexts without the author's rationalizations. They attack specific angles: functional correctness, authorization lapses, unhandled empty-state scenarios, partial database updates, and network interruptions. A suspected issue remains unconfirmed until execution commands or test runs prove it exists.
Running code directly is mandatory here, because any issue that cannot be reproduced against the live target is discarded as unverified.
What it helps with
- uncover hidden logic flaws in freshly implemented features
- test application resilience against empty environments and missing configuration
- filter out false positives by recreating bugs with target runtime commands
- identify residual operational exposure before delivery
Example prompts for the agent
- Run an adversarial-review on our newly added checkout flow and verify each failure against the running app
- Attack this diff from the empty-world and data integrity angles, then report reproduced bugs
- Launch an adversarial review across the entire repository with isolated subagents
How to install adversarial-review
1Skills CLI (Vercel)
npx skills add syahiidkamil/Software-Engineer-AI-Agent-Atlas --skill adversarial-review -a claude-code2Manually: put the files in place
~/.claude/skills/adversarial-review/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/adversarial-review/
1Skills CLI (Vercel)
npx skills add syahiidkamil/Software-Engineer-AI-Agent-Atlas --skill adversarial-review -a codex2Manually: put the files in place
.agents/skills/adversarial-review/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add syahiidkamil/Software-Engineer-AI-Agent-Atlas --skill adversarial-review -a cursor2Manually: put the files in place
.cursor/skills/adversarial-review/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install syahiidkamil/Software-Engineer-AI-Agent-Atlas .claude/skills/adversarial-review2Manually: put the files in place
.github/skills/adversarial-review/SKILL.md1Skills CLI (Vercel)
npx skills add syahiidkamil/Software-Engineer-AI-Agent-Atlas --skill adversarial-review -a gemini-cli2Platform CLI
gemini skills install https://github.com/syahiidkamil/Software-Engineer-AI-Agent-Atlas --path .claude/skills/adversarial-reviewManually: put the files in place
.devin/skills/adversarial-review/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about adversarial-review
How to install adversarial-review in Claude Code?
Install it in your terminal by running npx skills add syahiidkamil/Software-Engineer-AI-Agent-Atlas --skill adversarial-review -a claude-code.
What separates this skill from /code-review?
The /code-review command inspects static diffs without launching software, while adversarial-review runs tests against the active app to verify every defect.
What sections appear in the final summary report?
It outlines confirmed fixes with verification, deliberate deferred gaps, disproven false alarms, and uncovered residual risks.