Definition files

Everything that describes behaviour is YAML under config/agents/. Each kind below has its fields (generated from the JSON Schema the Rust structs export), the types they use, and the file this repository ships, in full. For how the files combine and what boot checks, see Configuration files.

Kind	Files	What it is
Workflows	config/agents/workflows/*.yaml	A graph of task-kind nodes and the edges between them, compiled at boot. The agent list, the approval card and the run itself all come from this file. Each task kind's with parameters are listed under Task kinds.
Models	config/agents/models.yaml	The model registry: which provider serves each model, its pinned revision and its license, checked against the allow-list in policies.yaml.
Prompts	config/agents/prompts/*.yaml	A versioned prompt for llm.extract: the system and user messages, and the limits on the untrusted document text placed in them. The llm.extract node names the model and the extraction schema.
Connectors	config/agents/connectors/*.yaml	An external system the platform calls: its kind, credentials, options and the actions that may be invoked, with whether each needs approval and is idempotent.
Tools	config/agents/tools/*.yaml	A tool exposed to a model, bound to a connector action. Declared and validated; not yet wired into the model (see Project status).
Roles (rbac.yaml)	config/agents/rbac.yaml	Roles, what each can do, and what it inherits.
Policies	config/agents/policies.yaml	Platform-wide policy: allowed model licenses, retention, PII handling and rate limits.
Tenant overrides	config/agents/tenants/*.yaml	One customer's differences, deep-merged over the shared definitions: thresholds, connectors, auto-approval.
Extraction schemas	config/agents/schemas/*.yaml	The JSON Schema llm.extract output must satisfy.