Claude Code hooks that flag risky code on edit, review each turn's diff, and review commits for multi-file flaws.

How it works

Foe: Breach Spider, a security hole an attacker could use.

Foes

  • Breach Spider

Party · Context

  • RogueYour agent100/100
  1. Breach Spider

    The Breach Spider blocks the way.

    What really happensIt stands for a security hole an attacker could use.

    Rogue: Context 100/100.

  2. Exploit

    The spider strikes through a gap nobody guarded. Its bite lingers.

    What really happensAn attacker gets in through a hole nobody checked, and the harm spreads until it is patched.

    Rogue takes 11. Rogue: Context 89/100. Rogue: Poison.

  3. Anthropic Security Guidance Plugin

    A ward hums each time the hero writes to a file.

    What really happensOn Edit and Write, regex rules flag about 25 risky patterns such as yaml.load, pickle.load, raw innerHTML and hardcoded secrets.

    Breach Spider takes 15. Rogue: Ward.

  4. Anthropic Security Guidance Plugin

    When the turn ends, a sentry reads every change made.

    What really happensOn Stop, the plugin sends the diff to an LLM review and feeds high-severity findings back to Claude to fix.

    Breach Spider takes 24.

  5. Exploit

    The ward takes the blow.

    What really happensA test or check catches the problem before it lands.

    Rogue blocks the hit.

  6. Anthropic Security Guidance Plugin

    At the commit gate, a tracker follows data across files.

    What really happensOn git commit, an SDK reviewer reads related files to trace data flow and catch IDOR, auth bypass and cross-file SSRF.

    Commit reviewer: Context 40/40.

  7. Anthropic Security Guidance Plugin

    The summons work the problem, each in its own context.

    What really happensEach subagent does its own model and tool work, apart from the main conversation.

    Breach Spider takes 12.

  8. Anthropic Security Guidance Plugin

    Commit reviewer returns with a short report.

    What really happensOnly the subagent's final summary comes back to the main agent.

    Rogue blocks the hit. Rogue: Insight. Commit reviewer returns.

  9. Anthropic Security Guidance Plugin

    The spider finds its usual doors already barred.

    What really happensClaude fixes flagged code before you see the reply. Rules in claude-security-guidance.md add checks specific to your codebase.

    Breach Spider takes 49. Breach Spider falls.

  10. Victory

    The Breach Spider falls.

Learn this spell

Claude Code

/plugin install security-guidance@claude-plugins-official

Notes

Version 2.0.16. Needs Claude Code 2.1.144+ and Python 3.8+. Three layers: regex warnings for about 25 risky patterns, an LLM review of each turn's diff on Stop, and an agentic reviewer on git commit. The LLM reviews default to the newest Opus and cost tokens or plan usage. Diffs and related files go to your model endpoint. Add your own rules in .claude/claude-security-guidance.md. Turn layers off with env vars such as ENABLE_STOP_REVIEW=0.

This spell has no recorded sessions in the curator's logs. Tried it? Send a spell report.