The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing untrusted external data, which could contain hidden instructions intended to override agent behavior.
Ingestion points: Untrusted data enters the agent context via the collection of interview transcripts, behavioral data, and customer reviews as defined in the references/skill-spec.json and references/decision-model.md files.
Boundary markers: The skill includes an "Evidence discipline" section in SKILL.md and instructions in references/skill-spec.json requiring the separation of facts, inferences, and assumptions, which acts as a logical boundary for processing external content.
Capability inventory: The skill is authorized to invoke local scripts for routing and task orchestration, specifically ../../scripts/dynamic_router.py and ../../scripts/neural_router.py.
Sanitization: The workflow incorporates several hooks designed to validate evidence, including ../../hooks/customer-language-check.md and ../../hooks/evidence-gate.md, though these are decision inputs rather than absolute filters.

customer-research