The Agent Skills Directory

[COMMAND_EXECUTION]: The engine in scripts/engine.py uses the ctypes module to dynamically load bundled shared libraries (.so files) from the parsers/ directory. This is used to interface with the tree-sitter parsing logic and is restricted to the local filesystem. Additionally, the test suite in tests/test_cli_cache.py uses subprocess.run to execute the skill's own CLI for verification.
[PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it reads and extracts content from arbitrary source files in a repository.
Ingestion points: scripts/engine.py reads file content from the local filesystem during scans and source retrieval.
Boundary markers: Absent. Extracted code and docstrings are returned directly to the agent context without explicit delimiters.
Capability inventory: Local file system read access and the ability to load local shared libraries.
Sanitization: None. The skill returns raw text extracted from the analyzed files.
[SAFE]: The skill implements a local caching mechanism that stores analysis results in a temporary directory or a user-defined path. This is used solely for performance optimization and includes security-conscious implementations such as atomic file writes and fingerprint validation.

tree-sitting