The Agent Skills Directory

[SAFE]: The skill implements a structured workflow for requirements elicitation. It prioritizes reading existing project documentation (such as README.md, CLAUDE.md, and /docs) to ground its recommendations in established codebase conventions.
[PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection as it ingests untrusted user input (feature requests, Jira tickets). It mitigates this risk by requiring a mandatory sign-off gate where the agent must present the compiled requirements for user confirmation.
Ingestion points: Raw feature requests and external documentation parsed in SKILL.md Step 2.
Boundary markers: Uses a rigid template in references/requirements-template.md to structure output.
Capability inventory: Performs file reads and writes; utilizes external search and context tools.
Sanitization: Relies on the user interview process to clarify and validate interpretations before writing to the filesystem.
[EXTERNAL_DOWNLOADS]: The skill utilizes the tavily MCP tool for web searches. This is a recognized service used for gathering technical documentation on libraries or APIs, and its use is scoped to providing context for the requirements document.

gather-requirements