The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external specifications or requirements to create structured implementation plans, creating an opportunity for adversarial content to influence the agent's output.\n
Ingestion points: Ingests user-provided specifications or requirements as described in the skill's purpose and architecture templates.\n
Boundary markers: Lacks explicit delimiters or instructions for the agent to ignore potentially malicious embedded prompts within the provided specifications.\n
Capability inventory: The generated plans involve file system writes to docs/plans/, shell command execution for testing (pytest) and version control (git), and the use of sub-skills like superpowers:executing-plans.\n
Sanitization: No input validation, escaping, or filtering of external content is performed before it is interpolated into the plan document.

writing-plans