The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest untrusted data (user task descriptions and project files) to generate and save plan files to the project root.
Ingestion points: User-provided task descriptions and project source files accessed via Read, Glob, and Grep tools.
Boundary markers: The instructions do not define specific delimiters or guidelines for the agent to ignore potentially malicious instructions embedded within the ingested task data.
Capability inventory: The skill allows the agent to read project files, search via grep/glob, and write markdown files ({task-slug}.md) directly to the project root.
Sanitization: There are no defined mechanisms for sanitizing or validating external input before it is interpolated into the generated plan files.

plan-writing