The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided project specifications and requirements, which constitutes an ingestion surface for untrusted content.
Ingestion points: Project specifications and requirements provided as input (SKILL.md).
Boundary markers: The workflow relies on task-based structure and a reviewer subagent rather than specific delimiters.
Capability inventory: The skill proposes shell command execution for version control (git) and testing (pytest), and writes implementation plans to the docs/ directory.
Sanitization: The process includes a mandatory review step where a subagent verifies the generated plan against the original specification using the plan-document-reviewer-prompt.md template.
[DYNAMIC_EXECUTION]: The skill's core functionality involves generating executable Python code and shell commands within markdown implementation plans.

writing-plans