Agile Iteration Method (AIM) 2.2

AIM is a delivery method for AI-assisted software work. You describe the outcome. AIM plans one useful increment, builds it, reviews it, validates it, and asks for the decisions that still belong to you.

It works with Codex, Claude Code, and GitHub Copilot.

Install

Install the complete, self-contained AIM Agent Skill through the open skills ecosystem:

npx skills add joneri/agile-iteration-method \
  --skill agile-iteration-method

This is full AIM, not AIM Lite. The package is generated from the same canonical workflow, adapter, installer, and schema sources as the adaptive AIM distribution. It does not require a separate AIM npm publication.

Target one supported agent explicitly when needed:

# Codex
npx skills add joneri/agile-iteration-method \
  --skill agile-iteration-method --agent codex --yes

# GitHub Copilot
npx skills add joneri/agile-iteration-method \
  --skill agile-iteration-method --agent github-copilot --yes

# Claude Code
npx skills add joneri/agile-iteration-method \
  --skill agile-iteration-method --agent claude-code --yes

Update an installed public skill with:

npx skills update agile-iteration-method --yes

The adaptive AIM installer remains available when you want reviewed repository calibration, native project specialists, or a broader supplier-specific footprint. Clone the public source so you can inspect exactly what will run:

git clone --depth 1 https://github.com/joneri/agile-iteration-method.git aim-source
cd aim-source
python3 scripts/aim_install.py --dry-run

Review the source and preview, then rerun with --apply only when the plan is correct. The installer asks for a repository and the adapters you use. Native project specialists can also be refreshed later through /aim configure-agents from aim.roles.yaml.

Already using AIM? Run:

/aim upgrade

Start

First verify the repository knowledge AIM will reuse:

/aim calibrate-repo

Record a durable project rule when AIM should remember it on later runs:

/aim remember-repo habits "Keep user-facing language direct and calm."

Then start with an outcome, not a task list:

/aim start "EPIC: Make checkout recovery clear and reliable when payment confirmation is delayed"

Use /aim help when you want the next useful action. The same command family also covers continue, status, validation, configuration, upgrade, memory, execution mode, cost depth, and replanning.

How AIM works

PO -> TDO -> Dev -> Reviewer -> TDO -> PO

• PO owns the outcome and acceptance.
• TDO chooses the next end-to-end Done Increment and validates delivery.
• Dev implements the approved increment.
• Reviewer looks for correctness problems, regressions, and risk.

Gate A approves the Epic. Gate B approves the next increment. Gate E accepts the result. Review and technical validation happen before acceptance.

Strict pauses at every hard gate. Auto continues while the approved direction remains clear, but still stops for risk, scope changes, and final Epic acceptance.

Repository-aware, not repository-heavy

AIM keeps four things separate:

Surface	Purpose
docs/workflow/agile-iteration-method.md	canonical AIM method
aim.profile.yaml	reusable repository knowledge
aim.roles.yaml	project-specific PO, TDO, Dev, and Reviewer expertise
.aim/	active local runtime state and review evidence

The standard installation adds the selected supplier skills and native project specialists. It never needs to create AGENTS.md or CLAUDE.md.

Native adapters

Platform	AIM skill	Project specialists
Codex	~/.agents/skills/agile-iteration-method/	.codex/agents/aim-*.toml
Claude Code	.claude/skills/aim/	.claude/agents/aim-*.md
GitHub Copilot	.github/skills/aim/	.github/agents/aim-*.agent.md

All adapters use the same AIM roles, gates, state ownership, and /aim command semantics. Supplier-specific files define how each project specialist works.

What is new in v2.2.1

• one adaptive installation instead of Personal, Team, and Enterprise editions
• project-specific native role specialists generated from aim.roles.yaml
• supplier-native skills for the full /aim command family
• adapter readiness receipts after install and upgrade
• current Codex skill discovery through $HOME/.agents/skills
• shorter product documentation and stronger release/documentation checks

The AIM runtime contract remains 2.0. Product release, runtime contract, installer manifest, and profile schema versions are tracked separately.

Safety

• one main AIM thread owns .aim/state.json and gate transitions
• native specialists never accept work or create parallel AIM runtimes
• existing files are collision-protected
• apply is rollback-protected and idempotent
• unavailable native delegation falls back to the same sequential role loop
• tags, releases, deploys, and other external changes still need explicit scope

Documentation

• Feature guide
• First-time journey
• Platforms and project specialists
• Install and upgrade
• Canonical AIM method
• Troubleshooting
• Release and publication
• Public Agent Skill distribution

Maintainers regenerate and verify the public package with:

python3 scripts/build_public_skill.py
python3 scripts/build_public_skill.py --check

Current product release: v2.2.1. See CHANGELOG.md.

Documentation is licensed under CC BY 4.0.