The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill relies on reading and updating project-specific files such as task.md, changelog.md, files.md, and the prds/ directory, which creates a potential surface for indirect prompt injection.
Ingestion points: The agent is instructed in SKILL.md to read the prds/ folder, task.md, changelog.md, and files.md to synchronize documentation and track task status.
Capability inventory: The skill authorizes the agent to write files to the repository and execute verification commands (e.g., npx convex dev, npm run typecheck) as part of the PRD workflow.
Boundary markers: The instructions do not define boundary markers or explicit instructions to treat content from these project files as untrusted data rather than instructions.
Sanitization: There are no requirements for sanitizing or validating the content read from these markdown files before the agent incorporates them into its execution logic.

project-workflow