The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content, including user ideas and large PRD documents (ingestion points). It uses 'gatekeeping markers' (boundary markers) and mandatory manual confirmation through the AskUserQuestion tool (sanitization) before performing operations like drafting requirements or clearing directories. The agent's capabilities are limited to writing to the docs/prd/ directory and delegating tasks to sub-agents (capability inventory).
[DATA_EXFILTRATION]: The skill uses git ls-files to perform automated project discovery. This metadata access is used to provide context for requirement analysis and is confined to local project scoping without external transmission.
[EXTERNAL_DOWNLOADS]: The skill integrates design context from well-known services such as MasterGo and Figma, and documentation sites like Storybook. These are identified as standard technology services and the interactions are limited to context retrieval.
[COMMAND_EXECUTION]: The skill performs file management within the docs/prd/ directory using Read, Write, and Glob tools. Operations that modify or delete files, such as the clear command, are protected by a mandatory dry-run and user approval workflow.

prd