# Agent Tasks

Bounded, approval-aware prompts for agents working in `tegwick-control`.

## When to use

Create an agent task when:

- the work stays inside this repository (drafting, structuring, summarizing);
- boundaries and forbidden changes are clear;
- human approval gates are explicit;
- the task is ready to hand to an agent without further triage.

Do **not** use agent tasks for:

- implementation work in other repos (prepare the prompt here, execute elsewhere);
- legal, tax, or financial conclusions;
- converting options into commitments without approval.

## Layout

```text
agent-tasks/
├── README.md                          this file
├── <area>-<action>-<subject>.md       ready agent prompts
└── archive/                           completed or superseded tasks
```

Flat layout by default. Move finished tasks to `archive/` during review.

## Naming

```text
<area>-<action>-<subject>.md
```

| Segment | Examples |
|---|---|
| area | `tegwick`, `binky`, `helix`, `coulomb` |
| action | `review`, `surface`, `draft`, `classify` |
| subject | `critical-path`, `area-cards`, `migration-slice` |

Examples:

- `binky-surface-critical-path.md`
- `tegwick-review-area-cards.md`

## File format

Copy `templates/agent-task.md` and fill every section. Required sections:

- Goal
- Context
- Allowed Changes
- Forbidden Changes
- Expected Output
- Acceptance Criteria
- Human Approval Required For

## Lifecycle

| State | Meaning | Location |
|---|---|---|
| Seed | Bootstrap or historical reference | root or `archive/` |
| Ready | Approved for agent execution | root |
| Done | Completed; move to `archive/` | `archive/` |

Track state in the task file header:

```markdown
Status: ready | done | superseded
Workplan: TEGWICK-WP-NNNN (if linked)
Area: binky | tegwick | helix | ...
```

## Intake path

```text
inbox/ → triage (INTAKE.md) → agent-tasks/ → agent session → archive/
```

See `INTAKE.md` for the full pipeline.