The Agent Skills Directory

[SAFE]: The skill implements a structured workflow with a "Plan gate" in SKILL.md that mandates explicit user approval before any specification files are authored, ensuring the agent does not operate autonomously on sensitive tasks without oversight.
[COMMAND_EXECUTION]: The skill utilizes Bash tools for Git, GitHub CLI, and Python. These operations are scoped to specific patterns (e.g., gh issue view) and are used for standard development workflows like reading issue content or executing the bundled validation script scripts/screen-states.py.
[EXTERNAL_DOWNLOADS]: The skill retrieves requirement data from external URLs or issue trackers. This is a primary function for gathering the context necessary to author the design specification.
[PROMPT_INJECTION]: The skill ingests untrusted data from external sources such as PRD files, issue tracker bodies, and web pages, which constitutes an indirect prompt injection surface.
Ingestion points: Data provided via $ARGUMENTS leads to reading local markdown files, fetching remote URLs via WebFetch, and retrieving issue content via gh or jira tools.
Boundary markers: The skill uses a comprehensive set of markdown templates (located in assets/templates/) to structure its output, and instructions emphasize a "source-of-truth hierarchy" to prevent the fabrication of requirements.
Capability inventory: The agent has access to Bash (git, gh, python, uv), Write, Edit, Task, and WebFetch tools.
Sanitization: The skill relies on the underlying model's safety filters and does not perform explicit character or script sanitization on the ingested text before interpolation into the design templates.
[SAFE]: The included Python script scripts/screen-states.py is a static linter that validates the presence of required fields and states in the generated specifications. It uses only Python standard library modules and performs no network operations.

product-design-spec