The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes user-provided requirements to generate agent instructions, which presents a surface for indirect prompt injection. This risk is effectively mitigated through a structured validation framework and quality gates.
Ingestion points: User requirements are ingested and analyzed in Stage 1 and Stage 2 as described in reference.md and examples.md.
Boundary markers: The skill uses 'Validation Gates' and 'TRUST 5' compliance notes to establish clear operational boundaries for generated agents.
Capability inventory: Generated agents are granted toolsets (e.g., Read, Write, Bash) based on a complexity scoring algorithm that minimizes unnecessary permissions.
Sanitization: The Validation Framework includes four distinct gates (YAML syntax, structure, content quality, and compliance) to ensure generated agents are safe and functional before deployment.
[COMMAND_EXECUTION]: The skill permits the use of the Bash tool for generated agents. Analysis of the examples and reference documentation confirms that this tool is intended for benign development tasks, such as running the 'Black' code formatter or other language-specific utilities.
[EXTERNAL_DOWNLOADS]: The Research Engine component utilizes WebFetch and the Context7 MCP server to retrieve official documentation. These operations are performed as part of a controlled research workflow to synthesize best practices from trusted framework documentation.

moai-core-agent-factory