The Agent Skills Directory

[SAFE]: The skill implementation utilizes the tree_sitter_languages library to perform static analysis on source code files. This is a standard and legitimate approach for code optimization and inspection tasks.
[COMMAND_EXECUTION]: The provided Python script contains logic to read local files using the pathlib module. This behavior is strictly scoped to the skill's stated purpose of scope extraction and does not include execution of system commands or unauthorized subprocess spawning.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external source code files, which constitutes a standard surface for indirect prompt injection (e.g., instructions hidden in code comments). The implementation follows best practices for context optimization which inherently reduces the noise seen by the LLM.
Ingestion points: Source code is read from local file paths within the extract_enclosing_ast_scope function in SKILL.md.
Boundary markers: The script wraps the extracted code in descriptive headers such as // --- [MODULE IMPORTS] --- to clearly delineate different sections of the context.
Capability inventory: The skill is limited to file reading and AST parsing; it lacks capabilities for network access, file writing, or code execution.
Sanitization: The extracted code is returned as a plain string without additional escaping, which is standard for context-sharing utilities.

tree-sitter-scope-extractor