The Agent Skills Directory

Prompt Injection (SAFE): No malicious overrides or safety bypass attempts. The use of role-play for sub-agents is core to the skill's intended brainstorming function and is constrained by a structured facilitation process.- Data Exposure & Exfiltration (SAFE): No access to sensitive files, environment variables, or hardcoded credentials. The skill explicitly instructs the agent not to read files or check git unless explicitly directed.- Remote Code Execution (SAFE): The skill does not download or execute external scripts or packages. It relies entirely on internal model logic.- Indirect Prompt Injection (SAFE): The skill ingests user input for ideation but lacks dangerous tools that could be exploited via malicious seed ideas. 1. Ingestion points: User-provided seed ideas and selection responses (SKILL.md). 2. Boundary markers: Explicit instructions to use for ideation only and not for implementation planning. 3. Capability inventory: Uses Task and AskUserQuestion tools. 4. Sanitization: None present, but risk is mitigated by the non-executable nature of the brainstorming output.

brainwriting