# Tanjiren MCP Prompts

Prompts are reusable workflow templates exposed by the MCP server.

They are not tools and they do not mutate state by themselves. They teach an AI client how to use the Tanjiren surface safely and effectively.

## Available prompts

### `start_here`

Use when:

- a session has just started
- the task is still vague
- you want the agent to orient itself before touching anything

Outcome:

- identity and scope summary
- plan and policy constraints
- best next read actions

### `diagnose_workstation`

Args: `instanceId` (required), `symptom` (optional)

Use when:

- a workstation is failing
- a server appears offline
- a host is behaving strangely and you need a read-first diagnosis

Outcome:

- workstation state summary
- likely explanations
- whether a bounded diagnostic task is justified

### `investigate_task_execution`

Args: `taskId` (required), `goal` (optional)

Use when:

- a task is failing
- a task appears stuck or partially complete
- you want a step-by-step review before deciding whether to wait, cancel, or retry

Outcome:

- overall task status
- step-level outcomes
- lease and execution summary
- recommended next step (wait, cancel, retry, or follow up)

### `prepare_safe_task`

Args: `objective` (required), `mode` (required: prompt|command), `targetHint` (optional)

Use when:

- you think you need to create a task
- you want to verify targets, plan, scopes, and policy first

Outcome:

- whether task execution is allowed
- which workers or workstations are appropriate
- a proposed bounded `create_task` payload or an explanation of blockers

### `summarize_org_activity`

Args: `focus` (optional)

Use when:

- you need an operational summary
- you want to brief a human operator
- you want a concise narrative of current org activity

Outcome:

- concise state summary
- main risks or blockers
- best next action

## Recommended workflow

Prompts are best used together with resources:

1. Read stable resources for context
2. Use the matching prompt to get a structured workflow
3. Execute only the tools justified by the prompt-guided workflow