The Agent Skills Directory

[SAFE]: The skill provides documentation and templates for integrating AI capabilities into CommandKit projects. No malicious instructions, obfuscation, or unauthorized access patterns were identified.
[DATA_EXFILTRATION]: No exfiltration or unauthorized data access patterns were found. Secret management in examples (e.g., in references/02-configure-ai-model.md) correctly demonstrates the use of environment variables for API keys.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of AI-driven command handlers, which process data originating from LLM outputs. This risk is addressed by explicit documentation in SKILL.md advising developers to treat AI output as untrusted and to maintain narrow tool descriptions.
Ingestion points: AI command parameters (ctx.ai.params) and tool parameters in references/03-ai-command-schema.md and references/04-custom-tools.md.
Boundary markers: None implemented in code snippets; instructions provided in the 'Guardrails' section of SKILL.md.
Capability inventory: Message replies and custom tool execution.
Sanitization: Recommended in the 'Guardrails' section of SKILL.md but not explicitly implemented in the example templates.

commandkit-ai