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
- Click Use this template on GitHub, or clone the repo.
- Run
./scripts/setup.sh (creates symlinks, installs git hooks).
- Replace the
[PLACEHOLDER] sections in AGENTS.md with your project's facts.
- 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
License
MIT
