The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates by reading and modifying user-controlled source code files across a project. This design creates a surface for indirect prompt injection, as malicious instructions hidden within the processed code (such as in comments or string literals) could potentially influence the agent's behavior during the refactoring process. The instructions do not specify sanitization steps or clear boundary markers for the data being edited.
Ingestion Points: Found in SKILL.md workflows where the agent reads file content using Grep and modifies it using Edit.
Capability Inventory: The skill has the capability to write to the file system and execute shell-like commands through its refactoring tools.
Sanitization: No specific escaping or validation of external code content is defined.
[COMMAND_EXECUTION]: The skill relies on tools like Grep and Edit to interact with the file system. These tools perform operations based on user-supplied strings and patterns, which involves executing commands on the underlying host environment to find and replace data.
[DYNAMIC_EXECUTION]: For tasks involving 10 or more files, the skill transitions to an "execution mode" where it generates and runs a Python script via the code-execution tool. This script uses a transformation module to perform bulk updates, which constitutes dynamic execution of generated code at runtime.
Evidence: SKILL.md provides a Python snippet: from api.code_transform import rename_identifier; result = rename_identifier(...).

code-refactor