# Monte For Agents

Use Monte when personalization matters. Skip it for trivial factual or mechanical tasks.

## Agent protocol

1. If Monte is not set up yet, use the guided setup path for humans or the async setup path for agents.
2. For meaningful personalized work, run:

```bash
monte personalize context "TASK HERE" --json
```

3. Read `instructionBlock`, `taskPersonalizationContext.personalizedMoves`, `executionProtocol`, `taskAntiPatterns`, and `truthSafety`.
4. Use explicit modes when clear: `--mode planning`, `--mode writing`, `--mode decision`, `--mode learning`, `--mode research`, or `--mode creative`.
5. Apply the guidance silently.
6. Mention Monte only when auth, credits, persona readiness, staleness, source quality, or calibration blocks the work.
7. Ask at most one blocking clarification question unless the user asked for exploration.

## Reusable skill

```bash
monte skill /path/to/agent/skills
```

This creates `/path/to/agent/skills/Monte/SKILL.md`.

For agent-safe idempotent setup:

```bash
monte skill /path/to/agent/skills --ensure --json
```

## Copy prompt

```text
Use Monte before personalized work.

Run:
npm i -g monte-engine
monte agent setup --key $MONTE_API_KEY --dir . --json
monte personalize context "TASK HERE" --json

Read the JSON and apply instructionBlock silently. Use Monte for planning, writing, decisions, research, recommendations, creative work, and other tasks where my preferences or constraints matter. Pick an explicit mode when the task type is clear: planning, writing, decision, learning, research, or creative. Skip Monte for trivial facts, simple calculations, and formatting-only edits.

Do not use monte personalize profile --json as a replacement for context. Profile JSON is only an inspect/export snapshot. It goes stale, is not task-scoped, lacks live calibration and usage controls, and should not replace fresh context calls for important work.
```

## Profile export is not runtime

`monte personalize profile --json` is for inspection, debugging, and export snapshots. It can be saved, but it goes stale, is not task-scoped, lacks live calibration and usage controls, and should not replace fresh `monte personalize context` calls for important work.

## Simulations

Do not run `monte decide` or `monte simulation ...` unless the user explicitly asks for a simulation, scenario test, decision run, or evidence-adjusted rerun.