The Agent Skills Directory

[SAFE]: The skill gathers project context by reading README files, git history, and design documents. This data exposure is restricted to the local project environment and is used solely for informing the brainstorming session.
[SAFE]: The landscape awareness feature (Phase 2.75) includes a mandatory privacy gate, asking the user for explicit permission before performing external searches. It also uses generalized category terms to protect proprietary ideas.
[PROMPT_INJECTION]: While the skill ingests untrusted data from the codebase, it is protected by a 'HARD GATE' instruction that strictly prohibits the agent from writing code or taking implementation actions, which mitigates the impact of potential indirect prompt injection.
[SAFE]: The visual sketch feature generates self-contained HTML wireframes without external dependencies or CDN links, adhering to secure development practices for generated UI components.

office-hours