Create an agent
Create a new agent for the caller's team. Agents can be used as the execution principal for team-owned runs.
Request body
A name for the agent
Optional description of the agent
Optional base prompt for this agent
Optional default cloud environment ID for runs executed by this agent. The environment must be owned by the same team as the agent.
Optional default runner UID for runs executed by this agent. When set, it overrides the selected environment's default runner for runs that do not specify their own runner_id. The editor must have View permission on the referenced runner.
Optional list of skill specs to associate with the agent. Format: "{owner}/{repo}:{skill_path}" (e.g., "warpdotdev/warp-server:.claude/skills/deploy/SKILL.md"). Each spec is validated and normalized at attach time using the team's GitHub credentials; inaccessible or malformed specs are rejected.
Optional base model for runs executed by this agent.
Optional map of MCP server configurations by name to attach to runs executed by this agent. Run-level MCP config takes precedence over this agent-level default.
Optional default harness for runs executed by this agent. Deprecated - use harness instead. Kept for backward compatibility; when both are sent, harness is authoritative and a conflicting type is rejected with invalid_request.
Default credential strategy for runs executed by a named agent.
- EXECUTOR: runs authenticate with the named agent's own credentials (e.g. a GitHub App installation token for the agent's team).
- CREATOR: runs authenticate with the credentials of the principal that created the run. Unlike the factory default, an agent may leave this unset. The strategy applied to a run is resolved in this order: the run's config.credential_strategy, then the agent's default, then the factory's default for factory-seeded agents, and finally EXECUTOR. The inherited strategy is validated at run creation time (the required credential must be mintable), like an explicit run-level value.
Whether runs created with this agent's API key may use the on_behalf_of field to attribute runs to another team member. Defaults to false. Only team admins may set this field.
Optional default worker host for runs executed by this agent. Omission, null, or an empty value stores no Agent default, in which case the workspace default applies. A non-empty value is trimmed and stored; use "warp" to force Warp-hosted execution over a self-hosted workspace default. The precedence order for worker host resolution is:
- The host specified on the run itself
- The agent's default host
- The workspace default host
Response
Agent created successfully
Unique identifier for the agent
Name of the agent
Optional description of the agent
Optional base prompt for this agent
Default cloud environment ID for runs executed by this agent. The precedence order for environment resolution is:
- The environment specified on the run itself
- The agent's default environment
- An empty environment
Default runner UID for runs executed by this agent. When set, it overrides the selected environment's default runner for runs that do not specify their own runner_id. The precedence order for runner resolution is:
- The runner specified on the run itself
- The agent's default runner
- The selected environment's default runner
- The environment's legacy inline compute fields
- System defaults
Whether this agent is within the team's plan limit and can be used for runs
When the agent was created (RFC3339)
When the agent was last updated (RFC3339)
Ordered list of normalized skill specs associated with this agent. Always present; empty when no skills are attached.
Base model for runs executed by this agent. The precedence order for model resolution is:
- The model specified on the run itself
- The agent's base model
- The team's default model
MCP server configurations attached to this agent by default. Run-level MCP config takes precedence over this agent-level default.
Default harness for runs executed by this agent. The precedence order for harness resolution is:
- The harness specified on the run itself
- The agent's base harness
- Oz Deprecated - use harness instead, which carries the full {type, model_id, reasoning_level} default.
Default credential strategy for runs executed by a named agent.
- EXECUTOR: runs authenticate with the named agent's own credentials (e.g. a GitHub App installation token for the agent's team).
- CREATOR: runs authenticate with the credentials of the principal that created the run. Unlike the factory default, an agent may leave this unset. The strategy applied to a run is resolved in this order: the run's config.credential_strategy, then the agent's default, then the factory's default for factory-seeded agents, and finally EXECUTOR. The inherited strategy is validated at run creation time (the required credential must be mintable), like an explicit run-level value.
Whether runs created with this agent's API key may use the on_behalf_of field to attribute runs to another team member.
Default worker host for runs executed by this agent, or empty when unset. The precedence order for worker host resolution is:
- The host specified on the run itself
- The agent's default host
- The workspace default host