The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input to generate implementation plans that contain shell commands and code.\n- Ingestion points: The skill is initiated using a "spec or requirements" as the primary data source (SKILL.md).\n- Boundary markers: The instructions do not prescribe the use of delimiters or boundary markers to isolate the untrusted specification content from the agent's logic.\n- Capability inventory: The skill produces plans containing executable tasks, including shell commands (git, pytest) and Python code, which are then passed to execution tools.\n- Sanitization: There are no instructions for validating or sanitizing the input specification to prevent malicious instructions from being incorporated into the plan.

writing-plans