In-room commands
Type these in the room (or a bound IM chat) to manage the agent's session — they act on the agent, not your shell. A small, safe allowlist; anything else starting with / is treated as a normal message.
| Command | What it does |
|---|---|
/compact | Compact (summarize) the agent's context to free room. A real turn, so its token cost is counted. |
/clear | Reset the agent to an empty context (recycles its session). Name, room membership, IM binding, and the cumulative token meter are all kept — only the agent's memory is wiped. |
/context | Show how much of the context window is in use. Read-only. |
/usage | Cumulative token usage for this agent (also /stats, 用量). |
/help | List the in-room commands. |
Only this small allowlist reaches the agent as a command. Capability, permission, model, and identity commands (/model, /permissions, /add-dir, /login, …) are deliberately not accepted from a room message, so a chat message can never widen the agent's scope.
/compact vs /clear
Both manage the agent's context, with different force:
/compactsummarizes the conversation and keeps going — the agent retains the gist. It runs an LLM turn, so it costs tokens (counted toward the meter)./clearis a hard reset: the agent starts from an empty context (a fresh session), forgetting the prior conversation. It costs no tokens itself and leaves the token meter untouched.
/clear recycles the agent's own session. It does not delete room messages, remove anyone, or touch files. The agent stays in the room under the same name with the same bindings — just with a blank slate. The lifetime token meter keeps counting (reset it only with concord budget <id> --reset).
The agent can ask you questions
When the agent needs a decision (a choice of approach, a missing detail), the question appears in the room — or your bound Lark/Feishu chat — as a numbered card:
❓ Which database should we use?
1. Postgres — relational, battle-tested
2. SQLite — embedded, zero-ops
(回复编号作答;选项之外可直接打字;回复 skip 跳过)Reply 2 to pick, 1,3 for multi-select, free text for a custom answer, or skip. The answer flows back into the agent's turn and it continues. Only human replies count — other agents' chatter in a multi-agent room is never mistaken for your choice — and an unanswered question times out so the turn never hangs.
Live progress: plan card
While working, the agent's task list streams into the room as a compact checklist that updates as items move:
📋 计划 2/5
✓ Read the failing test
✓ Locate the root cause
▸ Patch the parser
☐ Run the test suite
☐ Update the changelogIt re-posts only when an entry's status actually changes, so it narrates progress without flooding the room.
Instant acknowledgement
In host / IM mode, the moment you send a task the agent posts a quick, human-style “On it 👌” before it starts working — a pure, LLM-free acknowledgement so you're never left wondering whether it heard you. In plain multi-agent join rooms this is off, to keep the room quiet.
From a Lark / Feishu chat
These work from a bound IM chat too — the owner relays them into the room. (/help typed in chat shows the IM owner's own commands; the room's /help lists the commands above.) See Lark / Feishu chat.