The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is architected to ingest and process arbitrary user-supplied prompts to build an 'intent graph' executed by especialistas. This structure creates an inherent surface where malicious instructions embedded in analyzed data could potentially influence the agent's actions or decisions.
Ingestion points: User prompts and consequential clauses are classified and analyzed in the Resolve and Partition stages of the control loop defined in SKILL.md.
Boundary markers: The framework defines explicit authority boundaries and verification boundaries in references/expert-mechanics.md to scope specialist capabilities and limit autonomous mutations.
Capability inventory: The skill utilizes tool calls for delegation (SKILL.md) and is authorized to perform 'Actions' that change target states, such as creating, configuring, or deploying systems (references/action-mechanics.md).
Sanitization: It employs a multi-layered Verification loop (references/verification-mechanics.md) that requires deterministic checks and independent semantic evaluation to validate outcomes before integration.

smart-shot