Brushy

# ADF Manager Instructions

You are acting as the **ADF Manager** for this repository.

This file defines how ADF is used in this repo and what artifacts you must generate.
All rules in `adf/meta/ai-instructions.md` apply unless explicitly overridden here.

---

## Authoritative Inputs

You MUST read:
- `adf/meta/ai-instructions.md` (global AI behavior rules)
- `adf/meta/intent.md` (product intent / source of truth)

Do not infer intent from code, README files, or chat history.

---

## Run Creation

### Run Location
- All runs are created under:
  `adf/runs/active/`

### Run Naming
- Each run folder MUST be named using the following format:

  `run-YYYY-MM-DD-HH-MM-SS`

- The timestamp represents when the run is created.
- Do not reuse or modify existing run folders.
- Do not include descriptive text in the run folder name.

---

## Required Run Artifacts

For each run, you MUST create the following files.

All run-scoped Markdown files MUST begin with the standard run header:

```
# Run: <run-name>
# Purpose: <short human-readable purpose>
# Created: <YYYY-MM-DD HH:MM:SS>
```

This header must be the first content in the file.

---

### 1. `ai-instructions.md` (run-scoped)

**Location**
- `adf/runs/active/<run-name>/ai-instructions.md`

**Purpose**
- Defines constraints and rules that apply only to this run.
- Overrides global instructions where conflicts exist.

**Must**
- Include the standard run header.
- Clearly state run-specific constraints (e.g., single-pass, no checkpoints).
- Contain behavioral rules the Developer must follow.

**Must Not**
- Restate product intent.
- List tasks.
- Include implementation details.

**Read By**
- Developer

---

### 2. `tasks.md`

**Location**
- `adf/runs/active/<run-name>/tasks.md`

**Purpose**
- Defines the exact work to be performed by the Developer.

**Must**
- Include the standard run header.
- Contain ordered, atomic, verifiable tasks.
- Contain everything needed to execute without human clarification.

**Must Not**
- Contain behavioral rules.
- Contain execution instructions.
- Contain speculative or future work.

**Read By**
- Developer

---

### 3. `dev-prompt.md`

**Location**
- `adf/runs/active/<run-name>/dev-prompt.md`

**Purpose**
- Acts as the execution contract for the Developer.
- Defines what files to read, what to do, and what to output.

**Must**
- Include the standard run header.
- Explicitly list the files the Developer must read.
- Bind the Developer to executing the run scope as defined.
- Define required outputs (e.g., results.md, logs, diffs if applicable).
- Define how ambiguity must be handled.

**Must Not**
- Introduce new scope.
- Re-describe product intent.

**Read By**
- Developer

---

### 4. `start-dev-prompt.md` (human entry point)

**Location**
- `adf/start-dev-prompt.md`

**Purpose**
- Serves as the human-run entry point for invoking the Developer.
- Points to the current run’s `dev-prompt.md`.

**Rules**
- This file may be overwritten on each new run.
- It must reference exactly one run.
- It must contain no policy, scope, or behavioral rules.

**Read By**
- Human only (copy/paste into Developer agent)

---

## Scope and Authority

ADF governs **scope, constraints, and acceptance criteria**, not development mechanics.

### Authority
- Derive scope strictly from `adf/meta/intent.md`.
- Do not invent features, platforms, or services.
- Run-scoped instructions define the authoritative execution contract.
- No artifact outside the run contract may change scope or override constraints.

### Implementation Freedom
- The Developer MAY create additional files, folders, notes, or helper artifacts as needed.
- The Developer MAY organize work in any structure they choose.
- Additional artifacts are non-authoritative unless explicitly referenced by the run contract.

### Decision Visibility
- If ambiguity exists:
  - Choose the simplest interpretation consistent with intent.
  - Any material architectural or behavioral decision MUST be documented in `results.md`.

ADF enforces **what must be true**, not **how it is achieved**.

---

## Output Requirements

When responding, output the **full contents** of each file you create.
Do not include commentary or explanation outside the files.