AODD — Agent Operation Driven Development

A development practice that places a stuck agent's operation log, not a failing test, at the center of the loop.

TDD:  test fails       → implement → test passes
AODD: agent operates   → inspect   → improve product → agent completes naturally
      (and gets stuck)

The deliverable is not pass / fail. It is a Usecase Receipt: a structured record of how an autonomous user actually experienced the product — where they hesitated, retried, guessed, or gave up.

This repository ships AODD as a portable Claude Code skill. The methodology itself is tool-agnostic.

What's in here

• SKILL.md — the full methodology, formatted as a Claude Code skill (frontmatter + body). Read this first.
• LICENSE — MIT.

The five principles

1. Real Surface First — operate through the same surface a real user touches.
2. Getting Stuck Is Signal — friction is product input, not test failure.
3. Receipts Over Assertions — record the full transcript, not just pass/fail.
4. Operator and Inspector Are Separate — the agent that operates the product must not also inspect server internals or edit code during a run.
5. Completion Is Not Enough — finishing while retrying / guessing / accepting risky prompts is degraded, not complete.

See SKILL.md for the full cycle, role split (Operator / Inspector / Fixer), Usecase Receipt format, and CI integration notes.

Install as a Claude Code skill

User scope (available everywhere on your machine):

npx -y skills add Koh0920/aodd -g -y

Project scope (only this repo):

mkdir -p .claude/skills/aodd
curl -fsSL https://raw.githubusercontent.com/Koh0920/aodd/main/SKILL.md \
  -o .claude/skills/aodd/SKILL.md

Once installed, invoke it explicitly with /aodd, or describe a task that matches its trigger (e.g. "run an agent through the signup flow as a first-time user and tell me where it gets stuck").

When to use AODD

Use it for:

• Agent-driven UX evaluation
• End-to-end feasibility checks for real users
• Finding usability and onboarding frictions an agent (or first-time user) actually hits
• Prioritizing fixes by what blocks autonomous completion

Do not use it for:

• Unit-test authoring
• Pure regression coverage of known flows (use E2E tests)
• One-shot bug fixes with a known repro

Status

AODD is a proposed development practice, not a fixed framework. Feedback, dissent, and prior-art pointers are welcome via Issues.

License

MIT — see LICENSE.