How to ask another agent from a skill or a script

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81

# How to ask another agent from a skill or a script

This guide shows how to make a skill — or any shell script — put a question to another mm agent and use its answer. It assumes the agents are already declared under `mmConfigs` (see [how to ask another agent](ask-another-agent.md)). For your model to hand work to another agent on its own, the [`ask_agent` tool](let-the-agent-call-other-agents.md) is the simpler way: no command to copy, and the window shows which agent is working. A skill is for a fixed procedure, a script, or several agents in one command.

## Steps

1. In the skill, call the same mm back through the variables mm exports to the commands its `bash` tool runs. `MM_BIN` is the running `mm`, and `MM_CONFIG` its configuration, so `--ask` finds the agent names without a path:

   ```bash
   "$MM_BIN" --ask jan -p "What is the capital of Peru?"
   ```

   stdout carries the answer and nothing else; the label and the agent's tool calls go to stderr.

2. To send the conversation along, write it on stdin, then a `---` line, then the question:

   ```bash
   "$MM_BIN" --ask-with-context reviewer <<'CONTEXT'
   User: I think Go's goroutines are OS threads.
   Assistant: They are not: the Go runtime schedules them onto a few OS threads.
   ---
   Is the assistant right?
   CONTEXT
   ```

   The model writes this summary itself: a command cannot read the conversation.

3. Tell the model, in the skill, to run the command once and to report the answer as the other agent's. A complete skill (`skills/second-opinion/SKILL.md`):

   ```markdown
   ---
   name: second-opinion
   description: Get a second opinion from the reviewer agent on a claim or an answer from this conversation.
   keywords: second opinion, review, check, deuxième avis, vérifie, relis
   ---

   # Second opinion

   Call your `bash` tool ONCE with this command, the conversation summarised in English above the `---` line, the question below it:

       "$MM_BIN" --ask-with-context reviewer <<'CONTEXT'
       User: <the claim>
       Assistant: <your answer, if any>
       ---
       Is this right? Correct it if not.
       CONTEXT

   What it prints is the reviewer's answer. Report it, starting with "The reviewer says:"; never answer in its place. If it prints `[ask: …]`, tell the user that line.
   ```

   Keep such a command short: a small model copies a long one inexactly (quotes, prefixes).

4. Raise the time limits in the asking agent's configuration. A local model can take a minute to answer, and both limits default to well under that:

   ```yaml
   bashTimeout: 3m       # one bash command (default 30s)
   watchdogTimeout: 3m   # keeps counting while a command runs (ticket 57)
   ```

5. Run the asking agent from the directory the skills expect and ask in plain words:

   ```text
   > I think Go's goroutines are OS threads. Get a second opinion from Lucy on that claim.
   📖 read_skill: second-opinion
   🛠️  bash: "$MM_BIN" --ask-with-context reviewer <<'CONTEXT' …
      │ 🗣 asking reviewer (llmman/hf.co/Menlo/Lucy-gguf:Q4_K_M), with the conversation…
      │ The assistant's answer is correct. Go goroutines are not OS threads; …
   ```

## Variants

- **From a terminal or a script, outside mm.** Name the configuration: `mm -config demos/llmman/agent.llmman.yaml --ask lucy -p "…"`, or put the question on stdin. Check the exit code: 0 answered, 1 failed (the reason is on stderr), 130 interrupted.
- **An agent that should only answer.** Give it a configuration with `bashTool: false`, `editTools: false`, `planTool: false` and no skills — `demos/llmman/agent.lucy-reviewer.yaml` is one. A small model with every tool tends to wander (read files, load skills) when it is only asked for an opinion.
- **Capture the answer in a variable.** `answer=$("$MM_BIN" --ask jan -p "…")` — stdout holds the answer alone.

## See also

- Every flag, stream and exit code: [command-line reference](../reference/cli.md#ask-mode-specifics)
- `bashTimeout` and `mmConfigs`: [configuration reference](../reference/configuration.md)
- Why the agent is a child process, and what `--ask` adds for scripts: [asking another agent — explanation](../explanation/asking-another-agent.md)
- `preAsk` and `postAsk` hooks run for `mm --ask` too; with a `postAsk` hook, stdout carries the answer once the hooks have run: [control the hand-offs to other agents with hooks](control-agent-calls-with-hooks.md)