The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external binaries and CLI tools to extract metadata and help documentation.
Evidence: scripts/binary/capture_cli_help.sh executes the target binary with the --help flag using run_with_timeout "$@" --help.
Evidence: scripts/reverse_engineer.py uses subprocess.run to execute various tools including git, python3, and bash scripts.
[DYNAMIC_EXECUTION]: The skill generates and executes Python scripts during its operation.
Evidence: scripts/reverse_engineer.py contains a hardcoded template in _write_wrapper_validate_feature_registry which it writes to validate-feature-registry.py and subsequently executes via subprocess.run.
Evidence: scripts/binary/analyze_binary.sh uses a heredoc to pass an embedded Python script to python3 to scan for ZIP signatures within binary files.
[EXTERNAL_DOWNLOADS]: The skill fetches content from external sources which may include untrusted code or documentation.
Evidence: scripts/fetch_url.py uses urllib.request.urlopen to download sitemaps from user-provided URLs.
Evidence: scripts/reverse_engineer.py performs git clone operations on upstream repositories provided in the command arguments.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface as it ingests and processes data from external repositories, sitemaps, and binary artifacts.
Ingestion points: Repository source files (e.g., package.json, go.mod), documentation files, sitemaps (sitemap.xml), and binary strings.
Boundary markers: The skill does not implement specific delimiters to separate untrusted repository content from the agent's instructions during analysis.
Capability inventory: The skill possesses the ability to execute shell commands, run binaries, write to the filesystem, and perform network requests via git and urllib.
Sanitization: Includes _ensure_real_directory and _assert_no_symlinks in scripts/reverse_engineer.py to prevent path traversal and symbolic link attacks in the output directory.

reverse-engineer