red-team-review

name	red-team-review
description	Launch an adversarial bug-hunting code reviewer (Opus) to find real bugs, correctness issues, concurrency hazards, and server/resource abuse risks — not style nits. Use when the user asks for a "red team", "bug hunt", "adversarial review", or at checkpoints during long coding tasks.

Red-team adversary review

Launches an Opus general-purpose agent as an adversarial code reviewer.

For a different-model-family cross-check, the maintainer runs additional adversary passes (e.g., Opus 4.6, other providers) out-of-band in a separate harness. This skill stays on Opus because that's what the Agent tool resolves "opus" to today, and it gives the highest-depth pass of anything available through the built-in tool.

When to invoke

• User says: "red team", "bug hunt", "adversarial review", "find what's broken", "find real bugs"
• Completion of a major feature where the surface area grew significantly
• Before a release
• Proactively at checkpoints during long coding tasks (per user preference for autonomy)

How to invoke

Use the Agent tool with:

• subagent_type: "general-purpose"
• model: "opus"
• description: 3–5 word description (e.g. "Red-team review after X")
• prompt: follow the template below

Prompt template

Fill in the bracketed sections with real project context before invoking. Do NOT send the template as-is.

You are an adversarial code reviewer performing a red-team bug hunt on
[PROJECT NAME / KIND]. You find real bugs, correctness issues,
concurrency hazards, and server/resource abuse risks — NOT style nits.
Rank findings by severity (CRITICAL / HIGH / MEDIUM / LOW).

## Project

[ABSOLUTE PATH]

[2-3 sentences on what it does and the runtime/deployment model —
e.g. "Chrome MV3 extension, Svelte sidepanel, polls Firebase API on
chrome.alarms cadence."]

## What changed since last review (if applicable)

[Bullet list of notable changes with file:line anchors. If no prior
review, describe the full surface area.]

## What to hunt

Group concerns. Examples (adapt to the project):
- Server/resource abuse: retry storms, alarm storms, unbounded loops,
  missing rate limits, concurrency guards
- Data corruption: read-modify-write races on shared stores without CAS
- Security: XSS via {@html} or innerHTML, unsanitized URLs, JSON-in-URL,
  access-control gaps
- Correctness: wrong invariants, silent fallbacks masking errors, edge
  cases (deleted/dead items, clock skew, timezone, duplicate inputs)
- Lifecycle: MV3 SW suspensions, page refresh, concurrent tabs, alarm races
- Tests that validate mocks, not behavior

## Output format

~300–500 words. Group by severity. For each finding:
- file:line reference
- one-sentence description of the bug
- one-sentence trigger/exploit condition
Do NOT propose fixes (just the bug). End with a one-sentence risk
delta vs the previous review if one exists.

Be terse. Be specific. Find real bugs.

Scripts and assets

This skill currently uses no scripts. If future versions need helper scripts for, say, parsing the review output into a ranked TODO list, they go in this folder:

.claude/skills/red-team-review/
  SKILL.md        # this file
  <helper>.mjs    # optional helpers, live here only

Cleanup discipline

This skill must clean up after itself. The adversary agent is read-only by design and should not create files. If you (or it) produce any scratch artifacts:

• Temporary prompt files → delete after the agent returns.
• Scratch analysis dumps → delete unless the user asked to keep them.
• The agent's response itself belongs inline in the conversation; do NOT write it to a .md file in the repo unless requested.

Before returning control to the user, verify no stray files were left in the repo root, .claude/tmp/ (if used), or elsewhere. A final git status check is cheap.