The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as funnel analytics, user research, and market claims, creating a potential surface for indirect prompt injection. This risk is mitigated by explicit evidence policies in SKILL.md and references/skill-spec.json that require the classification of claims as fact, inference, or assumption, and the verification of external sources. Evidence Chain: (1) Ingestion points: Funnel analytics, user research, and optional external research sources. (2) Boundary markers: Mandatory evidence classification rules and structured output contracts. (3) Capability inventory: Internal routing scripts and optional web research tools. (4) Sanitization: Prohibitions against fabricating research and requirements for separating facts from unsupported assumptions.
[COMMAND_EXECUTION]: The skill utilizes internal framework scripts, including dynamic_router.py and neural_router.py, for decision routing and specialist agent selection. These are invoked via relative local paths (../../scripts/) and represent standard architectural components of the agent framework for coordinating task dependencies.

conversion-strategy