The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes analyzed binaries and various sub-scripts to perform its tasks. Specifically, it executes the target binary with the --help flag to capture its CLI surface.
Evidence: scripts/binary/capture_cli_help.sh executes the variable $BINARY_PATH with the --help argument. scripts/reverse_engineer.py uses subprocess.run to orchestrate various analysis steps.
Mitigation: The skill enforces a mandatory --authorized flag for binary analysis, which serves as a consent gate to ensure the user has permission to analyze the target.
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch and analyze external content via git cloning and HTTP requests for sitemaps.
Evidence: scripts/reverse_engineer.py performs git clone using the provided --upstream-repo parameter. scripts/fetch_url.py uses urllib.request to download sitemap files from external URLs.
[DATA_EXFILTRATION]: The scripts/fetch_url.py utility script supports reading from the file:// protocol. This allows the skill to read local files and copy them to its temporary directory, which could be exploited if an attacker can control the sitemap URL provided to the agent.
Evidence: scripts/fetch_url.py explicitly handles file schemes by reading from the local filesystem path using Path.read_bytes().
[PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection because it processes untrusted repositories, binaries, and documentation that could contain malicious instructions.
Ingestion points: scripts/reverse_engineer.py (external repo content), scripts/fetch_url.py (sitemap data), and binary strings.
Boundary markers: Present in SKILL.md (e.g., "Do not provide steps to bypass protections", "Redact secrets", "do not output reconstructed proprietary source").
Capability inventory: Subprocess execution (binary analysis), network access (cloning), and file system writes.
Sanitization: The skill includes a dedicated scripts/security/scan_secrets.sh utility to detect and prevent leakages of common credentials (AWS keys, GitHub tokens, Slack keys) in its output reports.

reverse-engineer