The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation describes an automated 'Plan Phase Continuation' workflow that is triggered by and processes information from pull request bodies and markdown files. This architectural pattern introduces an indirect prompt injection surface.
Ingestion points: The workflow reads 'Plan:' and 'Phase:' markers from pull request descriptions and parses markdown files within the docs/plans/ directory to determine the next steps for automation.
Boundary markers: There are no specified delimiters or instructions in the documentation to treat the plan content as data or to ignore embedded instructions, which could allow malicious content in a plan to override agent behavior.
Capability inventory: The system uses this ingested data to 'delegate the next phase to Copilot,' suggesting the agent performs tasks or generates code based on the content of the plan files.
Sanitization: No explicit sanitization or validation of the markdown content is described in the provided workflow documentation before it is processed by the AI.
[EXTERNAL_DOWNLOADS]: The skill requires the use of external tools for environment setup and maintenance.
Evidence: Instructions in SKILL.md include running uv sync --all-groups for package management and uv run pre-commit install for setting up git hooks. These are standard development tools and practices for managing dependencies and repository integrity.

writing-plans