AI-DLC Methodology Skill

English · 简体中文

A PM-friendly Agent Skill that brings the AWS AI-DLC (AI-Driven Development Life Cycle) workflow into Claude Code and Codex. Use it when you want an AI assistant to clarify scope, write or review a PRD, generate a question file, and prepare an engineering handoff — without rushing into code.

This is an independent, AI-DLC-inspired adapter, not an official AWS artifact and not endorsed by AWS. Upstream reference: awslabs/aidlc-workflows (MIT-0).

What it does

When invoked, the skill:

1. Detects whether the request is a new requirement, a PRD review, a rewrite, or implementation.
2. Picks the lightest workflow that can handle the request safely.
3. Generates a question file when key decisions are missing — instead of scattering them through chat.
4. Produces a PRD, review report, or engineering handoff using the bundled templates.
5. Stops at each major stage boundary and asks for Request Changes / Approve and Continue.
6. Only enters Construction (code planning + implementation) when the user explicitly asks.

Artifacts are written to an aidlc-docs/ directory in the user's workspace, so handoffs are durable and resumable.

Install

Claude Code

Copy the folder into your skills directory:

# global
cp -R aidlc-methodology ~/.claude/skills/

# or per-project
mkdir -p .claude/skills && cp -R aidlc-methodology .claude/skills/

Claude Code discovers it automatically on the next launch. Trigger it by describing the kind of work (e.g. "review this PRD", "help me shape this requirement using AI-DLC").

Codex

The bundled agents/openai.yaml exposes a display name and a default prompt for Codex environments that support skill manifests. Place the folder where your Codex client looks for skills.

Layout

aidlc-methodology/
├── SKILL.md                          # entrypoint loaded by the agent (source of truth)
├── SKILL.zh-CN.md                    # human-readable Chinese mirror
├── agents/
│   └── openai.yaml                   # optional Codex manifest
├── references/
│   ├── phase-stage-matrix.md         # which stages to run and when
│   ├── phase-stage-matrix.zh-CN.md
│   ├── pm-workflow.md                # PM-facing patterns and defaults
│   ├── pm-workflow.zh-CN.md
│   ├── artifact-contracts.md         # aidlc-docs/ directory contract
│   ├── interaction-patterns.md       # question files, approval gates, resume
│   ├── source-map.md                 # how to fall back to official rules
│   ├── example-third-party-sso.md    # worked example
│   └── example-third-party-sso.zh-CN.md
└── templates/
    ├── prd.md / prd.zh-CN.md
    ├── question-file.md / question-file.zh-CN.md
    ├── review-report.md / review-report.zh-CN.md
    └── engineering-handoff.md / engineering-handoff.zh-CN.md

English files are the source of truth for skill behavior. Chinese files are localized output shapes and human-readable mirrors. Three reference files (artifact-contracts, interaction-patterns, source-map) are English-only — translations welcome via PR.

How to invoke

Triggers the skill is tuned for (English and Chinese):

• "Apply AI-DLC to ..."
• "Review this PRD" / "PRD 评审" / "审一下这个 PRD"
• "Shape this requirement" / "需求梳理" / "需求拆解"
• "Prepare an engineering handoff" / "工程交接"
• "Scope this feature" / "划一下范围"
• "Generate a question file" / "出一个问题清单"

The skill assumes a PM posture by default. If you want it to plan or write code, say so explicitly — there are stop conditions designed to prevent it from drifting into implementation on its own.

Customizing

The skill is intentionally generic. To adapt it to your team:

• Templates (templates/*.md) are starting shapes — fork them to match your team's PRD conventions, JIRA/Linear fields, or rollout checklist.
• Worked example (references/example-third-party-sso.md) is a third-party SSO walk-through. Add or replace examples in your own fork for the kinds of features you ship most often.
• PM defaults (references/pm-workflow.md) include guidance on language and questioning style — adjust to your house style.

For higher-fidelity behavior, keep a local checkout of awslabs/aidlc-workflows and set AIDLC_WORKFLOWS_DIR to its path. The skill will load the official rules from there when available and fall back to the bundled summaries when not.

Compatibility

• Claude Code (skills)
• Codex (via agents/openai.yaml)
• Any AI client that respects the Agent Skill convention of SKILL.md + supporting files

Only name and description are used as frontmatter, so the skill stays portable.

License

MIT — see LICENSE.

This skill is an independent adaptation of the AI-DLC methodology. The official upstream workflow files are © Amazon Web Services and distributed under the MIT-0 license; see awslabs/aidlc-workflows. MIT-0 does not require attribution, but this project credits the upstream work voluntarily. This skill is not an official AWS artifact and is not endorsed by AWS.

Contributing

Issues and PRs welcome. Useful contributions:

• Additional worked examples in references/ for non-login features (payments, exports, notifications, search filters, billing changes).
• Localized templates (e.g. Chinese / Japanese / Spanish PRD templates in templates/).
• Tightening trigger phrases in SKILL.md based on real usage.