The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data from PRD files, design specifications, and user-provided URLs to establish review targets. This content is used to prompt both the primary agent and independent subagents, creating a surface for indirect prompt injection.
Ingestion points: Reads repository files such as PRD, DESIGN.md, and UI-target.md (SKILL.md, references/target-and-capture.md).
Boundary markers: Uses Markdown sections but lacks specific instructions to ignore malicious directives embedded within the ingested data.
Capability inventory: Has file system read/write access and executes git/environment commands (SKILL.md).
Sanitization: No explicit sanitization or escaping of external content is defined before it is passed to the subagent prompts.
[COMMAND_EXECUTION]: The skill interacts with the local shell to manage git state and control the UI testing environment.
Evidence: Executes git check-ignore and git status to verify and manage artifact visibility in the repository (SKILL.md).
Evidence: Instructions include starting, connecting to, and stopping local services to facilitate real-time UI capture (SKILL.md, references/target-and-capture.md).

review-ui