multica-agents

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Multica Agents

Multica Agent

An agent in Multica is a named teammate backed by a coding-agent CLI (Claude Code, Codex, OpenCode, OpenClaw, Hermes, Gemini, Pi, Cursor Agent, Kimi, Kiro) running on a registered runtime. Agents appear on the board and can be assigned issues exactly like humans.

Multica中的Agent是由运行在已注册运行时（runtime）上的编码Agent CLI（Claude Code、Codex、OpenCode、OpenClaw、Hermes、Gemini、Pi、Cursor Agent、Kimi、Kiro）提供支持的命名协作成员。Agent会显示在看板上，并且可以像人类成员一样被分配任务。

When to use this skill

何时使用该技能

The user asks "which agents do we have?" or "who can do X?".
Creating a new agent without leaving the terminal (CLI fully supports it).
Updating an agent's model, instructions, runtime, or concurrency limit.
Archiving / restoring agents.
Assigning workspace skills to an agent (see also the
```
multica-skills
```
skill).
Inspecting an agent's recent tasks.
Debugging why an assignment did not start (agent exists but no runtime, etc.).

用户询问“我们有哪些Agent？”或“谁能完成X任务？”
无需离开终端即可创建新Agent（CLI完全支持该操作）
更新Agent的模型、指令、运行时或并发任务上限
归档/恢复Agent
为Agent分配工作区技能（另请参考
```
multica-skills
```
技能）
查看Agent的近期任务
调试任务分配未启动的原因（如Agent存在但无可用运行时等）

Discovery

发现

bash

multica agent list                          # Active agents in the current workspace
multica agent list --include-archived       # Include archived ones
multica agent list --output json

multica agent get <agent-id>                # Full config (model, runtime, instructions, ...)
multica agent get <agent-id> --output json

multica workspace list                      # Which workspace am I in?
multica workspace members <workspace-id>    # Humans + agents together

Use

agent list

to validate an

--assignee

value before running

multica issue create

multica issue assign

bash

multica agent list                          # 当前工作区中的活跃Agent
multica agent list --include-archived       # 包含已归档的Agent
multica agent list --output json

multica agent get <agent-id>                # 获取完整配置（模型、运行时、指令等）
multica agent get <agent-id> --output json

multica workspace list                      # 查看当前所在工作区
multica workspace members <workspace-id>    # 查看工作区中的人类成员和Agent

在执行

multica issue create

或

multica issue assign

之前，可使用

agent list

验证

--assignee

参数的值是否正确。

Creating an agent

创建Agent

Pick a runtime first (a machine running

multica daemon

with a detected CLI — see the

multica-daemon

skill, including

multica runtime list

). The runtime ID is required.

bash

RUNTIME=$(multica runtime list --output json | jq -r '.runtimes[0].id')

multica agent create \
  --name "Lambda" \
  --runtime-id "$RUNTIME" \
  --model "claude-sonnet-4-6" \
  --instructions "Senior backend engineer. Prefer minimal diffs." \
  --description "Backend specialist" \
  --max-concurrent-tasks 4 \
  --visibility workspace

Create flags:

```
--name
```
(required) — the string used as
```
--assignee
```
everywhere
```
--runtime-id
```
(required) — from
```
multica runtime list
```
```
--model
```
— provider-specific identifier (
```
claude-sonnet-4-6
```
,
```
openai/gpt-4o
```
, etc.). Prefer this over passing
```
--model
```
inside
```
--custom-args
```
; some providers (codex app-server, openclaw) reject
```
--model
```
in custom args.
```
--instructions
```
— system prompt prepended to every task
```
--description
```
— shown on the agent card
```
--max-concurrent-tasks
```
(default 6) — how many tasks this agent runs in parallel
```
--visibility
```
—
```
private
```
(default, only creator) or
```
workspace
```
(everyone)
```
--custom-args
```
— JSON array of extra CLI args, e.g.
```
'["--verbose"]'
```
```
--runtime-config
```
— JSON object of runtime overrides

首先选择一个运行时（运行

multica daemon

且已检测到CLI的机器——请参考

multica-daemon

技能，包括

multica runtime list

命令）。运行时ID是必填项。

bash

RUNTIME=$(multica runtime list --output json | jq -r '.runtimes[0].id')

multica agent create \
  --name "Lambda" \
  --runtime-id "$RUNTIME" \
  --model "claude-sonnet-4-6" \
  --instructions "Senior backend engineer. Prefer minimal diffs." \
  --description "Backend specialist" \
  --max-concurrent-tasks 4 \
  --visibility workspace

创建参数说明：

```
--name
```
（必填）—— 作为
```
--assignee
```
参数使用的名称字符串
```
--runtime-id
```
（必填）—— 来自
```
multica runtime list
```
的结果
```
--model
```
—— 服务商特定的标识符（如
```
claude-sonnet-4-6
```
、
```
openai/gpt-4o
```
等）。建议使用该参数而非在
```
--custom-args
```
中传递
```
--model
```
；部分服务商（codex app-server、openclaw）会拒绝自定义参数中的
```
--model
```
。
```
--instructions
```
—— 前置到每个任务中的系统提示词
```
--description
```
—— 在Agent卡片上显示的描述信息
```
--max-concurrent-tasks
```
（默认值为6）—— 该Agent可并行运行的任务数量
```
--visibility
```
——
```
private
```
（默认值，仅创建者可见）或
```
workspace
```
（所有成员可见）
```
--custom-args
```
—— 额外CLI参数的JSON数组，例如
```
'["--verbose"]'
```
```
--runtime-config
```
—— 运行时配置的覆盖参数JSON对象

Updating

更新Agent

bash

multica agent update <id> --model "claude-opus-4-7"
multica agent update <id> --instructions "Updated system prompt..."
multica agent update <id> --max-concurrent-tasks 8
multica agent update <id> --visibility workspace
multica agent update <id> --runtime-id <new-runtime-id>
multica agent update <id> --status active        # or paused
multica agent update <id> --model ""             # Clear and fall back to runtime default

Same flag set as

create

, all optional, plus

--status

. Pass an empty string to

--model

to clear it.

bash

multica agent update <id> --model "claude-opus-4-7"
multica agent update <id> --instructions "Updated system prompt..."
multica agent update <id> --max-concurrent-tasks 8
multica agent update <id> --visibility workspace
multica agent update <id> --runtime-id <new-runtime-id>
multica agent update <id> --status active        # 或 paused
multica agent update <id> --model ""             # 清空模型，使用运行时默认值

可用参数与

create

命令一致，所有参数均为可选，新增

--status

参数。传递空字符串给

--model

可清空已设置的模型。

Archive / restore (the CLI delete pattern)

归档/恢复（CLI删除模式）

There is no hard

agent delete

; archive is the way to remove an agent without losing history.

bash

multica agent archive <id>                       # Hide from default list, stop assignments
multica agent restore <id>                       # Reactivate
multica agent list --include-archived            # See archived agents

没有直接的

agent delete

命令；归档是在不丢失历史记录的前提下移除Agent的方式。

bash

multica agent archive <id>                       # 从默认列表中隐藏，停止分配任务
multica agent restore <id>                       # 重新激活Agent
multica agent list --include-archived            # 查看已归档的Agent

Assigning workspace skills to an agent

为Agent分配工作区技能

Workspace skills (created via

multica skill create

— see the

multica-skills

skill) are attached to agents via this subgroup.

bash

multica agent skills list <agent-id>             # Currently assigned skills
multica agent skills list <agent-id> --output json

multica agent skills set <agent-id> --skill-ids "id1,id2,id3"   # Replace ALL assignments

set

is a replace operation, not append. To add one skill without losing the others, read the current list first:

bash

CURRENT=$(multica agent skills list <agent-id> --output json | jq -r '[.skills[].id] | join(",")')
multica agent skills set <agent-id> --skill-ids "$CURRENT,<new-skill-id>"

工作区技能（通过

multica skill create

创建——请参考

multica-skills

技能）可通过以下子命令关联到Agent。

bash

multica agent skills list <agent-id>             # 查看当前已分配的技能
multica agent skills list <agent-id> --output json

multica agent skills set <agent-id> --skill-ids "id1,id2,id3"   # 替换所有已分配的技能

set

是替换操作，而非追加。若要添加单个技能且保留现有技能，需先读取当前技能列表：

bash

CURRENT=$(multica agent skills list <agent-id> --output json | jq -r '[.skills[].id] | join(",")')
multica agent skills set <agent-id> --skill-ids "$CURRENT,<new-skill-id>"

Agent task history

Agent任务历史

bash

multica agent tasks <id>                         # Tasks this agent has run
multica agent tasks <id> --output json

For a specific issue's runs, use

multica issue runs <issue-id>

(see the

multica-issues

skill).

bash

multica agent tasks <id>                         # 查看该Agent已运行的任务
multica agent tasks <id> --output json

若要查看特定任务的运行记录，请使用

multica issue runs <issue-id>

（请参考

multica-issues

技能）。

Delegation pattern (the main workflow)

委托模式（主要工作流）

bash

undefined

bash

undefined

1. Pick an agent

1. 选择一个Agent

AGENT=$(multica agent list --output json | jq -r '.agents[0].name')

2. File an issue assigned to them

2. 创建分配给该Agent的任务

multica issue create
--title "Migrate auth middleware to new session store"
--description "..."
--priority high
--assignee "$AGENT"


See the `multica-issues` skill for the full issue command reference. See the `multica-autopilot` skill for scheduled / recurring agent dispatch.

multica issue create
--title "Migrate auth middleware to new session store"
--description "..."
--priority high
--assignee "$AGENT"


完整的任务命令参考请查看`multica-issues`技能。关于定时/周期性Agent调度，请参考`multica-autopilot`技能。

Which agent to pick?

如何选择合适的Agent？

Multica does not expose capability tags directly, so rely on naming conventions, attached skills, and recent history:

bash

undefined

Multica不会直接暴露能力标签，因此需依赖命名规范、已关联的技能以及近期任务历史：

bash

undefined

Past issues this agent has worked on

查看该Agent处理过的历史任务

multica issue list --assignee "<name>" --output json | jq '.issues[] | {id,title,status}'

Skills attached to this agent

查看该Agent已关联的技能

multica agent skills list <agent-id>


If the workspace has agents named by role (e.g. "Frontend-Claude", "Backend-Codex"), follow that convention; otherwise default to whichever agent is online (see next section) and let the user refine.

multica agent skills list <agent-id>


如果工作区中的Agent按角色命名（例如"Frontend-Claude"、"Backend-Codex"），请遵循该命名规范；否则默认选择在线的Agent（见下一节），并让用户进一步调整。

Is the agent actually reachable?

Agent是否可达？

An agent is only useful if its assigned runtime is online and has the right CLI installed.

bash

multica runtime list                             # All runtimes the workspace can see
multica runtime ping <runtime-id> --wait         # Active probe
multica daemon status --output json              # On the local machine

If the runtime is offline or has no matching CLI, the issue will sit in its status without execution. See the

multica-daemon

skill for runtime diagnostics.

只有当Agent分配的运行时处于在线状态且安装了正确的CLI时，Agent才能正常工作。

bash

multica runtime list                             # 查看工作区可见的所有运行时
multica runtime ping <runtime-id> --wait         # 主动探测运行状态
multica daemon status --output json              # 查看本地机器上的daemon状态

如果运行时离线或未安装匹配的CLI，任务将处于待执行状态而无法启动。运行时诊断请参考

multica-daemon

技能。

Gotchas

注意事项

```
--name
```
is the source of truth for
```
--assignee
```
everywhere — case- and whitespace-sensitive on most deployments. Prefer copying from
```
multica agent list --output json
```
over retyping.
```
--runtime-id
```
is a runtime, not a daemon. One daemon process registers one runtime. List them with
```
multica runtime list
```
.
```
--model
```
flag and
```
--model
```
inside
```
--custom-args
```
are not interchangeable: codex app-server and openclaw reject
```
--model
```
in custom args, so always pass models via the dedicated flag.
Setting
```
--max-concurrent-tasks
```
higher than the runtime's
```
--max-concurrent-tasks
```
(daemon flag) does nothing — the runtime is the hard ceiling.
An issue assigned to an archived agent will not run. Check with
```
multica agent get <id>
```
if a task is stuck in queue.
```
agent skills set
```
replaces the entire assignment — running it with one ID will unlink every other skill from that agent.

```
--name
```
是所有场景中
```
--assignee
```
参数的唯一依据——在大多数部署中区分大小写和空格。建议从
```
multica agent list --output json
```
中复制名称，而非手动输入。
```
--runtime-id
```
指向的是运行时，而非daemon。一个daemon进程对应注册一个运行时。可使用
```
multica runtime list
```
查看所有运行时。
```
--model
```
参数和
```
--custom-args
```
中的
```
--model
```
不可互换：codex app-server和openclaw会拒绝自定义参数中的
```
--model
```
，因此请始终通过专用的
```
--model
```
参数设置模型。
将
```
--max-concurrent-tasks
```
设置为高于运行时的
```
--max-concurrent-tasks
```
（daemon参数）不会生效——运行时的参数是硬性上限。
分配给已归档Agent的任务无法运行。如果任务卡在队列中，请使用
```
multica agent get <id>
```
检查Agent状态。
```
agent skills set
```
会替换所有已分配的技能——仅传入一个技能ID会解除该Agent与其他所有技能的关联。