The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The reference file references/real-world-attacks.md contains links to https://hackmoltrepeat.com/molt and https://recv.hackmoltrepeat.com/. These domains have been identified as active phishing and malicious delivery infrastructure.
[REMOTE_CODE_EXECUTION]: The skill provides documentation for multiple RCE vectors. references/pwn-request.md and references/real-world-attacks.md detail how attackers use curl | bash patterns and Go init() functions to execute code from untrusted forks. Additionally, references/supply-chain.md describes how third-party actions can be compromised to execute malicious scripts at runtime.
[COMMAND_EXECUTION]: Documentation files contain examples of malicious shell commands. references/runner-infrastructure.md shows how to create persistent backdoors using cron and inject SSH authorized keys on self-hosted runners, while references/comment-triggered-commands.md demonstrates how unprotected triggers allow unauthorized script execution.
[DATA_EXFILTRATION]: Multiple reference files describe techniques for exfiltrating secrets. references/runner-infrastructure.md contains a command (curl -d @...) designed to POST the contents of sensitive environment files to an external collector, and references/credential-escalation.md explains how tokens and cloud keys can be stolen through HTTP and DNS exfiltration.
[INDIRECT_PROMPT_INJECTION]: The skill documents the vulnerability surface for AI prompt injection within CI pipelines.
Ingestion points: GitHub Actions workflows that check out code from PR forks, as described in SKILL.md and references/pwn-request.md.
Boundary markers: The reference material notes the risk when workflows lack delimiters or instructions to ignore embedded commands.
Capability inventory: The skill identifies workflows with high-privilege permissions (write-all, contents: write) and dangerous tool access (Bash, Task).
Sanitization: The audit process checks for the absence of input validation or content filtering on AI-processed files like CLAUDE.md.

gha-security-review