Exploration Protocol

Exploration Protocol

A structured approach to explore features before production development. The goal is to learn, build acceptance criteria, experiment with verifiable prototypes, and codify learnings.

Task Folder Structure

tasks/NNN-taskname/
├── 01-objectives.md
├── 02-acceptance-criteria.md
├── 03-architecture.md
├── 04-verifiable-prototype.md
├── 05-verification.md
├── 06-outcome.md
├── 99-research/
└── 99-findings/

Document Templates

You MUST read the template files in ./references/001-example-task/ before scaffolding a task folder. These templates show the exact structure and format to use:

• [01-objectives.md](./references/001-example-task/01-objectives.md)
• [02-acceptance-criteria.md](./references/001-example-task/02-acceptance-criteria.md)
• [03-architecture.md](./references/001-example-task/03-architecture.md)
• [04-verifiable-prototype.md](./references/001-example-task/04-verifiable-prototype.md)
• [05-verification.md](./references/001-example-task/05-verification.md)
• [06-outcome.md](./references/001-example-task/06-outcome.md)

Key Patterns

Stub Markers

For sections to be filled in later:

> Stub - to be filled in after objectives and acceptance criteria are approved.

Criteria Tables

Always pair criteria with verification method:

| Criterion | Verification Method |
|-----------|---------------------|
| User can log in | Manual test: complete flow, see logged-in state |
| API returns 401 on bad auth | `curl -H "Auth: bad"` shows 401 |

Checkpoint Lists

For prototypes, use checkboxes:

**Checkpoint 1:**
- [ ] Auth0 application configured
- [ ] PKCE enabled

ASCII Diagrams

Keep simple:

┌─────────────┐     ┌─────────────┐
│   Popup     │────▶│  Background │────▶ API
└─────────────┘     └─────────────┘

Phase Flow

1. Objectives → user reviews
2. Acceptance Criteria → user approves before work begins
3. Architecture → stub until criteria approved
4. Prototype → incremental with checkpoints
5. Verification → evidence for each criterion
6. Outcome → learnings and follow-on work

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: dwmkerr
• Source: dwmkerr/claude-toolkit
• License: MIT
• Homepage: https://www.skills.sh/dwmkerr/claude-toolkit

Install and usage instructions live in the source repository linked above.