> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mobile-starter.amisi.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Agents

> Use AMiSi role contracts for planning, implementation, review, and documentation

Agents define who performs a role, what context that role needs, which actions it may take, and what it must return. AMiSi stores portable agent contracts under `.ai/agents` without selecting a model or provider.

## Available agents

| Agent                        | Responsibility                                                                    | Worktree access    |
| ---------------------------- | --------------------------------------------------------------------------------- | ------------------ |
| **Architect**                | Investigate ownership, dependencies, risks, tests, and implementation scope.      | Read-only          |
| **Implementer**              | Apply an approved task or plan as the only worktree writer.                       | Read and write     |
| **Reviewer**                 | Independently assess correctness, security, architecture, tests, and regressions. | Read-only          |
| **Documentation specialist** | Create and validate Mintlify content against shipped source.                      | Documentation only |

Each contract defines its inputs, responsibilities, restrictions, and output format. Read the relevant contract before performing that role.

## Invoke an agent

Give your AI tool the complete task and the contract path:

```text theme={null}
Read AGENTS.md, .ai/README.md, and .ai/agents/architect.md. Investigate this
request without modifying files. Return the current behavior, proposed approach,
exact files, verification strategy, risks, and unresolved questions.
```

For implementation, provide the approved plan and require `.ai/agents/implementer.md`. Do not run multiple implementers against one worktree.

## Use native delegation

A tool with native sub-agents can delegate independent architecture, platform, testing, or review work concurrently. Every worker must receive:

* the complete task and acceptance criteria
* the relevant agent contract
* the selected skill or workflow
* the repository files needed to support its conclusions

A tool without native delegation follows the same roles sequentially in one conversation.

<Warning>
  Keep one worktree writer. Parallel agents should remain read-only until the
  parent consolidates their findings into one approved implementation plan.
</Warning>

## Next steps

* [Choose a skill](/ai/skills) for the repeatable procedure an agent should follow.
* [Structure a task](/ai/task-specs) with stable goals, constraints, and acceptance criteria.