Custom Agent Support

Hive can run any command in a terminal. To have it identify a custom coding agent, show its state, and restore it after a reboot, add an agent-detection manifest.

Add a manifest

Create ~/.config/hive/agent-detection/my-agent.toml:

toml

id = "my-agent"
name = "My Agent"

[detect]
process = ["my-agent"]

[start]
command = ["my-agent"]

[resume]
last = ["--continue"]

[[rules]]
state = "working"
region = "bottom:3"
[rules.match]
contains = "thinking"

The id becomes the --kind value. [detect] describes the running process, [start] is the command Hive launches, and [resume] describes how to continue the agent’s latest conversation. Add screen rules for the agent’s visible working, blocked, done, or idle indicators.

Load and inspect it

Ask Hive to reload agent manifests, or restart the daemon:

sh

hive server reload-agent-manifests
hive agent manifests

Start the custom agent in an environment and inspect what Hive detects:

sh

hive agent start --kind my-agent --env app
hive agent explain <agent>

explain includes the matched rule and terminal screen so you can tune detection against what the agent actually displays. A custom manifest can replace or extend built-in detection definitions.

Report state directly

If the agent has a reliable hook or wrapper, it can report state directly instead of relying on screen rules. From inside an agent pane:

sh

hive pane report-agent --state working

Direct reports are useful for agents whose screen output changes often or is difficult to classify. Keep the report aligned with the same states Hive uses: working, blocked, done, and idle.

Give the agent Hive tools

If the agent should also operate other agents, connect it to Hive’s MCP server or provide the Hive Agent Skill. For example, Claude Code can register the server with:

sh

claude mcp add hive -- hive mcp

See Integrations for the supported hooks and Orchestration for agent-to-agent workflows.