The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The reference documentation (e.g., references/pwn-request.md and references/real-world-attacks.md) contains examples of remote code execution patterns, such as piping curl downloads directly into bash. These are documented as attack patterns for security auditors to identify in workflows.
[EXTERNAL_DOWNLOADS]: The skill documents URLs associated with malicious infrastructure (e.g., hackmoltrepeat.com) and generic attacker placeholders (attacker.com). These are used within the context of explaining how attackers deliver payloads in CI/CD environments.
[OBFUSCATION]: The skill provides examples of how attackers hide malicious commands using Base64 encoding and shell environment variable manipulation (e.g., using ${IFS} to bypass space filters) in branch names and filenames.
[COMMAND_EXECUTION]: The skill instructions allow the agent to use Bash and Task tools to perform repository audits, which is appropriate for a security review tool but involves executing commands in the local environment.
[DATA_EXFILTRATION]: Reference guides (e.g., references/runner-infrastructure.md) describe techniques for stealing credentials and sensitive configuration files from CI runners, such as accessing ~/.aws/credentials or ~/.docker/config.json and sending them to remote servers.
[INDIRECT_PROMPT_INJECTION]: The skill explicitly addresses the risk of AI agent configuration poisoning, where files like CLAUDE.md or .cursorrules are used to inject malicious instructions into AI coding agents running in CI.
[PERSISTENCE]: The documentation identifies methods for maintaining access on self-hosted runners, including the use of crontabs for periodic beacons and the installation of unauthorized SSH keys.
[PRIVILEGE_ESCALATION]: The skill analyzes how attackers escalate from PR-level access to repository ownership by stealing elevated credentials like Personal Access Tokens (PATs) and deploy keys.
[CREDENTIALS_UNSAFE]: The analysis identifies the risk of accessing sensitive paths (e.g., .env, .ssh/id_rsa, .aws/credentials) during workflow exploitation, documented as a threat vector in references/credential-escalation.md.

gha-security-review