The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands to execute a Python-based search utility located within the workspace environment.
Evidence: The instructions specify the execution of python .codestable/tools/search-yaml.py with various filters and query parameters.
Risk: The agent executes local scripts provided in the .codestable/tools/ directory, which relies on the integrity of the files within the user's project workspace.
[PROMPT_INJECTION]: The skill operates as a processing layer for local files, creating a surface for indirect prompt injection from data in the workspace.
Ingestion points: The skill reads project-specific context from .codestable/attention.md and decision records from the .codestable/compound/ directory.
Boundary markers: There are no explicit instructions for the agent to use delimiters or ignore embedded commands when reading these files.
Capability inventory: The skill has the capability to write new files to the directory and execute local shell commands via Python.
Sanitization: No sanitization or validation logic is defined for the content extracted from the decision records before it is incorporated into the agent's context.

cs-decide