Agent resources are the bindings between an agent and the artifacts that shape its behavior — rules, MCP configs, skills, vaults, model selection, and integrations. Resources are how an agent gets its personality, its tools, and its secrets.

Each attachment is a small record (ccr_…) that links one agent (cci_…) to one entity (aar_…, amc_…, ask_…, vlt_…, or an integration). You can think of it as a many-to-many join: the same rule can be attached to multiple agents, and a single agent can have many rules.

How it works

Create the artifact via its own API (Rules, MCP configs, Skills, Vaults).
Attach it to an agent with POST /agents/{agent_id}/resources, passing the entity_type and entity_id.
Redeploy the agent (or call /deploy with update_config_only: true) for the change to take effect.
Detach with DELETE /agents/{agent_id}/resources/{resource_id} when no longer needed.

Supported entity types

Type	What it does
`agent_artifact_rule`	Adds a system-prompt rule (text instructions).
`agent_artifact_mcp_cfg`	Adds an MCP server config (external tool).
`agent_artifact_skill`	Adds a packaged skill (`.skill` / `.zip`).
`vault`	Injects vault secrets into the agent's environment.
`github_repository_integration`	Grants access to a GitHub repo.
`anthropic_integration`	Selects an Anthropic API account.
`llm_model`	Pins which LLM the agent uses (for example `opus-4.5`, `sonnet-4.6`).

When to use it

Compose an agent's behavior from existing artifacts. Attach rules, skills, MCP configs, vaults, GitHub repos, and a model — instead of redefining capabilities per agent, mix-and-match what you've already built.
Swap an agent's capabilities without recreating it. Detach an outdated skill, attach a new MCP config, switch the model — then /deploy with update_config_only: true to roll the change.

Overview

How it works

Supported entity types

When to use it

On this page