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).

Steps

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:
```
"$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.

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

"$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.

Tell the model, in the skill, to run the command once and to report the answer as the other agent's. demo-llmman/skills/ has three complete examples:
- delegate-to-jan — a question delegated to Jan;
- second-opinion — the conversation sent to a reviewer;
- compare-agents — the same question to Jan and Lucy, in one command.
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:
```
bashTimeout: 3m       # one bash command (default 30s)
watchdogTimeout: 3m   # keeps counting while a command runs (ticket 57)
```

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

> 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 demo-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 — demo-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.

 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


# 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)).

## 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. `demo-llmman/skills/` has three complete examples:

   - `delegate-to-jan` — a question delegated to Jan;
   - `second-opinion` — the conversation sent to a reviewer;
   - `compare-agents` — the same question to Jan and Lucy, in one command.

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 demo-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 — `demo-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)