The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts using python3 to display and write configuration state.- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external tool outputs and local files which could contain malicious instructions.
Ingestion points: orchestrator_capabilities tool output, ~/.config/pstack-t3/roles.json, and .pstack/t3-roles.json.
Boundary markers: None explicitly mentioned in the instructions for separating data from instructions.
Capability inventory: The skill can write files to the home directory, execute subprocesses, and delegate tasks to other models.
Sanitization: Instructions do not explicitly specify sanitization or escaping for the strings interpolated into commands.- [DYNAMIC_EXECUTION]: The skill dynamically constructs shell commands in Step 4 using user-provided role names and model options, creating a surface for command injection. Additionally, it offers to trigger the creation of new verification skills.

setup-pstack