Agents
What an agent is in Habakkuk, and where to read about identity, teams, memory, skills, schedules, the board, approvals, sessions and retros.
On this page
An agent is a configured assistant: its models, permission mode, workspace, toolsets, and the files that shape how it talks. One Habakkuk can run several agents, each with its own memory and skills, and they can hand each other work.
yaml
agents:
habakkuk:
model: anthropic/claude-opus-5-5
permission: workspace
workspace: ~/habakkuk-workspaceEvery message an agent handles is a turn: a durable job that runs the model, calls tools, and ends with a reply. Turns belong to a session, the conversation's history.
In this section#
Agents and identity define an agent, give it a personality and instructions, and how the first chat lets you name it.Multi-agent teams route conversations to agents, delegate to sub-agents, and give each agent a home chat.Memory Markdown notes, the user profile, semantic search, background review, and undo for every change.Skills
SKILL.md folders an agent loads before a task, and writes for itself.Schedules prompts and scripts on a cron, at most once per slot.Task board cards one agent hands another, that outlive the conversation.Permissions and approvals what an agent may do on its own, the reviewer model, hooks and safety guards.Sessions and chat commands titles, compaction, fork and rewind, standing goals, and the / commands.Retros an agent looks back over its own work, finds causes in the evidence, and proposes changes you approve.Defaults worth knowing#
- An agent without
toolsetsgetscore,memory,skills,schedule,delegateandboard. A config that liststoolsetsgets exactly that list. - The default permission mode is
workspace: reads and writes inside the workspace run on their own, commands ask first. - Messages from chat apps need approval for commands even in
fullmode, unless you opt in withtrust: full. See Security.
In this section
- Agents and identityDefine agents, give them a personality and standing instructions, use personas, and how the first-chat welcome lets the user name the agent.
- Multi-agent teamsRun several agents in one Habakkuk: route conversations to them, switch agents in a chat, delegate to sub-agents, give each a home chat, and talk to agents elsewhere over A2A.
- MemoryLong-term memory as Markdown notes, the user profile, untrusted notes, semantic search, background review, and a history of every change you can undo.
- SkillsSkills are SKILL.md folders an agent loads before a task it covers: where they live, how to run and write them, installing outside skills, and the curator.
- SchedulesScheduled prompts and scripts: cron expressions, intervals and one-off times, at most one run per slot, script schedules that skip the model, and where results go.
- Task boardThe task board agents use to hand each other work that outlives a conversation: cards, dependencies, dispatch, results back to the creator, and the rules.
- Permissions and approvalsPermission modes, how approvals work in chats and the terminal, the approval reviewer model, hooks that can block a tool call, and the guards that always apply.
- Sessions and chat commandsHow conversations map to sessions: idle timeouts, titles, compaction, fork and rewind, standing goals, export and search, and the chat commands that steer a turn.
- RetrosRetrospectives: an agent looks back over its own work, finds what went wrong and why from Habakkuk's records, proposes changes, applies the safe ones, and checks next time whether they helped.