The Agent Skills Directory

[SAFE]: The skill uses localized paths for its routing logic and specialist hooks (e.g., ../../scripts/neural_router.py, ../../hooks/evidence-gate.md). These references are consistent with a modular agent architecture and do not involve remote code execution from untrusted sources. \n- [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes external inputs such as 'product truth' and 'sales/support evidence' (Ingestion points in SKILL.md). However, it includes mandatory boundary markers that require the agent to classify all material claims as fact, inference, or assumption. The capability inventory is restricted to internal routing and narrative synthesis, and sanitization is enforced by invariants in references/skill-spec.json that mandate evidence verification and the rejection of unsupported recommendations.

product-marketing