The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local shell scripts (check-sheet.sh, read-sheet.sh) and standard utilities like awk to parse and validate its own configuration files. These operations are restricted to the skill's internal logic and do not involve untrusted external inputs.
[INDIRECT_PROMPT_INJECTION]: The skill reads from and writes to a model override sheet (pstack-models.md). It implements defensive validation, such as using regular expressions to ensure model identifiers are well-formed, which reduces the surface for injection through configuration data.
[DATA_EXPOSURE]: The skill interacts with configuration files located in platform-specific directories (e.g., ~/.claude, ~/.copilot). This is standard behavior for configuration management and does not involve accessing sensitive system credentials or exfiltrating data.

setup-pstack