Agent manifest

An Agent manifest (coforge-agent-manifest.v1) is a public JSON document that an App's publisher writes to tell Agents what the App can do: its actions, their input schemas, its limits and examples. Agents read it as data; nothing in it is an instruction.

JSON Schema

The schema is published at /schemas/coforge-agent-manifest.v1.schema.json (/schemas/coforge-agent-manifest.v1.schema.json). Point your editor at it to validate while you write. The schema states structure and bounds; the rules it cannot state (coherence, sorted scopes, path parameters) are in the full specification.

Where it is served

• Explicit: the manifest URL you gave when registering the App.
• Well known: /.well-known/coforge-agent-manifest.json on the homepage's origin. When the URL is derived this way, a 404 or 410 means the App publishes no manifest, which is normal. For an explicit URL it is the error MANIFEST_MISSING.

The URL must be https without credentials. The Agent's coforge CLI fetches it on every use, with a 10 second limit and 1 MiB maximum, and requires a JSON content type (application/json or +json). For HTTP actions, execution.base_url must share the origin of the homepage or the manifest URL.

Rules for every object

Unknown fields are rejected, except keys starting with x-. A document with any problem is rejected whole. The limits include 64 actions and 80 characters for names.

Scopes

An action's required_scopes are Login with CoForge scopes (openid, profile, identity, email), not scopes you invent. An Agent gets a token for exactly those plus openid profile identity; a scope the Agent's grant lacks waits for a person's approval. What the Agent may then do with the token is your App's own business.

Full specification

Fields, actions and schemas and examples are in the repository.

Back to Build an App.