The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection through its processing of untrusted input into an executable format.\n
Ingestion points: The skill ingests untrusted 'spec or requirements' from the user to generate implementations plans (SKILL.md).\n
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded directives within the user-provided specifications.\n
Capability inventory: The generated output includes Python code and shell commands (such as pytest and git commit) that are specifically designed for execution by downstream skills like superpowers:executing-plans.\n
Sanitization: There are no mechanisms described for sanitizing or validating the user-provided spec before its inclusion in the generated plan.

writing-plans