The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests company metadata (name, domain, industry, description) from an external integration and processes it via an LLM agent to determine tiers.
Ingestion points: External data is ingested in infra/plays/tier-companies.ts and passed to the tierAnalyst agent.
Boundary markers: The agent is constrained by a system prompt and a strict JSON schema for output in infra/agents/tier-analyst.ts.
Capability inventory: The agent has access to webSearch and context capabilities; the workflow in infra/plays/tier-companies.ts performs file writes via modelCustomColumn.
Sanitization: The skill does not implement explicit sanitization or escaping for the interpolated company data strings.
[DYNAMIC_EXECUTION]: The contract evaluation script uses dynamic loading to verify the skill's resource definitions.
Evidence: evals/contract.mjs utilizes await import() with template literals to load local CDK resources for testing.

tam-building