uFawkesAI

An implementation of Anthropic's AI-Native SDLC Playbook (Claxton, 2026): six stages, each ending in a committed artifact, built on the Superpowers skill suite. It ships as a template and as a signed devcontainer image (fawkes-space) with Claude Code, OpenCode, Codex and Gemini CLI installed.

What you get

• AGENTS.md — one page of instructions. CLAUDE.md, GEMINI.md, .cursorrules and .github/copilot-instructions.md are symlinks to it, so Claude Code, Gemini CLI, Cursor, Copilot and Codex all load the same file.
• .agents/agents/ — four agents, each a boundary that can block, write or change repo state: planner, builder, verifier, operator.
• .agents/commands/ and .agents/skills/ — slash commands and on-demand skills, shared by Claude Code and OpenCode through symlinks.
• REVIEW.md — the review rubric. /review-local applies it before you commit; CI applies it again on the PR.
• .devcontainer/ — uses the fawkes-space image (below): Claude Code, OpenCode, Codex, Gemini CLI, language servers, pre-commit and scanners.
• CI guardrails: PR size limit, artifact-chain check, agent evals, secret scan, documentation freshness.

The six stages

Each stage commits an artifact the next stage reads, so the commit history is the audit trail. Agents never approve their own work or merge their own PRs; a human code owner does.

Stage	Artifact	Superpowers skill	Where it lives here
1. Plan	docs/ai-sdlc/<feature>/intent.md	brainstorming	discovery skill, @planner, /plan
2. Design	spec.md	brainstorming	@planner; CI requires an intent.md beside every spec.md
3. Build	plan.md, then the diff	writing-plans, test-driven-development, subagent-driven-development	@builder, /implement, /tdd
4. Test	tests, builds, eval results	verification-before-completion, systematic-debugging	@verifier, /verify; agent-config-ci.yml replays evals when agent configuration changes
5. Deploy	PR with review findings	requesting-code-review, finishing-a-development-branch	REVIEW.md, the review-local command before commit, @operator, /ship; a human merges
6. Maintain	an incident or metric that becomes the next intent.md	systematic-debugging	/release files the follow-up issue; the metrics live in uFawkesObs

Stage 6 is partial: the loop back into intent.md is manual today.

Quick start

1. Click Use this template on GitHub, or clone the repo.
2. Run ./scripts/setup.sh (creates symlinks, installs git hooks).
3. Replace the [PLACEHOLDER] sections in AGENTS.md with your project's facts.
4. Open the folder in VS Code and choose Reopen in Container.

DORA AI Capabilities implemented

DORA AI capability	Where
Clear and communicated AI stance	AI_STANCE.md, AGENTS.md
Healthy data ecosystems	docs/CHANGE_IMPACT_MAP.md, docs/API_SURFACE.md
AI-accessible internal data	AGENTS.md context index, .vscode/settings.json
Strong version control practices	ci-quality.yml PR size check, docs/COMMIT_CONVENTIONS.md
Working in small batches	400-line PR limit, docs/GOLDEN_PATH.md
User-centric focus	.agents/skills/discovery/SKILL.md
Quality internal platforms	.devcontainer/devcontainer.json, .agents/agents/

Playbook components

Component (from the playbook)	Where it lives
CLAUDE.md project knowledge	AGENTS.md, with CLAUDE.md and other harness files as symlinks
Skills	.agents/skills/ and the Superpowers suite baked into the image
Hooks	.pre-commit-config.yaml, .claude/settings.json, scripts/hooks/protected-paths.json
Subagents	.agents/agents/ (four agents: planner, builder, verifier, operator)
Evals	.agents/evals/, run by agent-config-ci.yml
REVIEW.md	REVIEW.md; applied locally by the review-local command
Sandbox	.devcontainer/devcontainer.json using ghcr.io/paruff/fawkes-space (non-root)

Container images

Built, signed (cosign, keyless) and published by build-cde-space-image.yml on every v<semver> tag.

Image	Use
ghcr.io/paruff/fawkes-space	The devcontainer: core + Claude Code, OpenCode, Codex, Gemini CLI, language servers, qmd
ghcr.io/paruff/fawkes-core	Linters, secret/vuln/SAST scanners and pre-commit only

The devcontainer's base layer is the same package under the tag base-X.Y.Z. Tags: X.Y.Z, X.Y, X, latest. Pin a version or digest from the Releases page and verify with cosign verify (cosign v3 or later; the command is in each release's notes).

Host requirements: 2 CPUs and 4 GB of memory for the container. A smaller VM kills agent sessions with no error. Colima: colima start --cpu 4 --memory 8.

Works with

Claude Code, OpenCode, Codex, Gemini CLI, GitHub Copilot, Cursor, and anything else that reads AGENTS.md.

AWS's AI-DLC is installed in the image as an opt-in alternative lifecycle; nothing is configured until you run aidlc config.

Related repositories

uFawkesPipe (CI/CD pipeline) and uFawkesObs (observability and DORA dashboards). See docs/UFAWKES_INTEGRATION.md.

Contributing

See CONTRIBUTING.md. Security issues: SECURITY.md.

Research basis

• DORA, AI Capabilities Model (2025): the seven capabilities above.
• DORA, 2025 State of AI-assisted Software Development: AI amplifies what is already there.
• Anthropic, The AI-Native SDLC Playbook (2026): the artifact chain.

License

MIT