The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local CLI utilities such as pip show, pip index, and python -c to verify the existence, versions, and import paths of libraries. These operations are diagnostic in nature and are used as ground-truth verification against potential AI hallucinations.
[EXTERNAL_DOWNLOADS]: The skill performs metadata queries against HuggingFace's official API to verify AI model availability. HuggingFace is a recognized trusted organization, and these network operations do not involve sensitive data access.
[DATA_EXFILTRATION]: The process includes a mandatory 'Sanitize the Vision' phase that instructs the agent to strip tool names, vendor references, and phantom requirements from user-provided briefs. This serves as a security control to prevent data leakage and anchoring bias during the research phase.
[PROMPT_INJECTION]: The skill incorporates instructional guidelines such as 'SYNTHESIZE FIRST' and token-management rules to optimize agent performance. It also mandates an 'Adversarial Review' phase designed to challenge and verify all architectural claims, which acts as a quality gate against malicious or incorrect outputs.

deep-brainstorming