The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from multiple project-local files which could potentially contain malicious instructions or unexpected content designed to influence agent behavior.
Ingestion points: The skill reads from design/gdd/, design/registry/entities.yaml, docs/engine-reference/, and .claude/docs/.
Boundary markers: The skill implements a strict 'Section Cycle' and 'Collaborative Protocol' requiring the agent to present drafts to the user and obtain explicit approval via AskUserQuestion before using the Edit or Write tools.
Capability inventory: The skill uses Read, Glob, Grep, Write, Edit, and Task tools. It possesses the capability to modify the project's design documents and registry based on its analysis.
Sanitization: The skill relies on the user's manual review of drafts as its primary sanitization mechanism rather than automated filters.
[NO_CODE]: The skill is composed entirely of markdown instructions, metadata, and tool configurations. It does not include or execute any standalone script files (e.g., .py, .js) or binary executables.

design-system