Leaderboard/Security/skills
Last commit on September 1, 2026·Created on January 14, 2026

trailofbits/skills

Secures CI/CD pipelines against prompt injection and attacker-controlled input in AI agent integrations
Combined rank
#65
across all skills
In Security
#5
category rank
Stars
7.2k
+1.6% in last 7d
Forks
611
+1.2% in last 7d
Watchers
65
+4.8% in last 7d
Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
TL;DR

This skill audits GitHub Actions workflows to identify security vulnerabilities in AI agent integrations, including Claude Code Action and Gemini CLI. It helps security researchers detect attack vectors where untrusted input reaches AI agents through environment variables, expression injection, or permissive sandbox configurations.

The tool is designed for use during architecture reviews and threat modeling of CI/CD pipelines to ensure that agentic action configurations do not introduce critical security risks.

WHO IT'S FOR
Security researchers
performing vulnerability hunting and threat modeling
DevSecOps engineers
auditing CI/CD pipelines for AI prompt injection
Smart contract auditors
scanning Algorand and StarkNet contracts for bugs
Blockchain security engineers
validating access control and arithmetic in contracts
Repository contents

85 skill files

Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.

Use this skill collection

Follow the documented setup, then try a first task.

Claude Code: Audit Context Building

Installs: Installs the audit-context-building plugin only. · Claude Code

  1. Enter this command in a Claude Code agent session.

    /plugin install trailofbits/skills/plugins/audit-context-building

Give it something to do.

Suggested first task

Propose one strong property-based test

Uses property-based-testing

  • Replace <target-path> with the file, module, or small component to review.
  • Provide any relevant existing test directory or testing conventions if they are not discoverable from the project.

Check the repository for this skill’s setup.

Use the property-based-testing skill to review <target-path>. Identify one strong, non-tautological property supported by the code and sketch a focused test using the project's existing property-testing library. If the project has no such library, propose the specific property first and ask before adding a dependency.
Suggested first task

Search narrowly for variants of a known bug

Uses variant-analysis

  • Replace <codebase-path> with the repository or source-tree path.
  • Replace <known-issue-location> with the file and line, function, or other precise location of an already identified bug.
  • Provide the existing vulnerability or logic-bug claim and any known triggering conditions.

Check the repository for this skill’s setup.

Use the variant-analysis skill on <codebase-path> to search for variants of the known issue at <known-issue-location>. Start by stating its root cause, create an exact-match search that finds the known instance, then generalize one element at a time and triage the resulting candidates.