slug | string | yes | Agent task slug (e.g. fix-auth-bug). When flight is set, defaults to flight-<name>. |
description | string | yes (autonomous) | Task description or follow-up message. Required for autonomous mode unless flight is set. |
workspace | string | yes (new spawns) | Workspace name. The server resolves repos and environment from the workspace definition. |
branch | string | no | Working branch for every repo in this spawn. Per-repo overrides in repos[].branch take precedence. |
purpose | string | no | One-sentence human-readable summary of the agent’s goal (e.g. “Add pagination to the users API”). Displayed in the dashboard. Max 240 characters. |
model | string | no | Model override (e.g. claude-opus-4-8). |
agent | string | no | Claude Code agent persona (e.g. programmer, architect). |
service_profile | string | no | Spawn under this service profile — borrowing a non-owned profile, which requires the service-profile.assume permission. Use spawn_profiles to discover which profiles you can borrow. |
backend | string | no | Code agent backend: claude (default) or codex. |
reasoning_effort | string | no | Reasoning effort level. Claude: low, medium, high, xhigh, max. Codex: low, medium, high, xhigh. Empty = model default. |
reasoning_summary | string | no | Codex reasoning summary: auto, concise, detailed, none. Only valid for backend=codex. |
verbosity | string | no | Codex verbosity: low, medium, high. Only valid for backend=codex. |
service_tier | string | no | Codex service tier: standard (default), fast, flex. Only valid for backend=codex. |
repos | object[] | no | Repos to clone with optional branch overrides. Each entry must have clone_url. See Repo object. |
session_mode | string | no | autonomous (default), interactive, or streaming. |
completion_check | string | no | none (default) or assessor (run completion assessor up to 3x before idling). |
on_idle | string | no | Idle behavior: sleep (default), terminate, or keep_alive. |
dequeue_strategy | string | no | Follow-up dequeue strategy: auto (default — up to five follow-ups of one kind per turn; manual follow-ups and auto-generated events never share a turn), all (drain all into one batch), one (one per turn), or five (up to five per turn). |
resurrect | boolean | no | Resume a completed/killed agent on the same branch with its full conversation history. Requires the same slug. Default: false. |
force_new | boolean | no | Start fresh even if a prior session exists. Default: false. |
fork_from | string | no | Branch off another agent’s conversation into a new independent agent. Use a new slug — the fork gets the parent’s history but diverges from that point. |
tasks | string[] | no | Checklist items the agent must complete. |
tags | string[] | no | Tags to attach for dashboard grouping and filtering. Each must be a DNS-label slug ([a-z0-9][a-z0-9-]{0,62}); max 8. A name that matches no existing tag is created with a default color. |
append_system_prompt | string | no | Text appended to the agent’s system prompt. |
out | string | no | Expected output artifact. Short keywords are expanded: pr (open a pull request), push (push changes, no PR), respond (just respond, no code changes). Freeform text is passed through verbatim. |
flight | string | no | Flight file path relative to .murmur/flights/ (e.g. deploy.md). When set, reads and executes the flight via a pilot agent. |
flight_inputs | object | no | Context inputs for the flight, injected into the pilot prompt (e.g. {"repo": "org/repo", "number": "42"}). Only used when flight is set. |
each | string | no | Fanout: path to a file containing one item per line. Spawns one agent per line. slug and description must contain {} which is replaced with each item. |