A skills-based Software Development Lifecycle framework for autonomous coding agents. Drop it into any project and your agent follows a structured, auditable process — from requirements through retrospective — instead of ad-hoc coding.
Systematic over ad-hoc. Every project goes through the same stages: inception, design, architecture, planning, implementation, review, testing, and retrospective. The framework enforces this by making skills mandatory, not optional.
Evidence over claims. Every stage writes artifacts to disk. No stage relies on conversation memory alone. All decisions are documented — requirements in brd.md, architecture decisions in ADRs, test results in test-plan.md.
Gates over momentum. Each stage has an explicit exit checklist. If it fails, the agent loops within the stage. Proceeding with a failed gate is a process violation.
Human in the loop at irreversible decisions. Architecture approval, destructive operations, and ambiguous decisions require explicit human sign-off before the agent proceeds.
Check for a relevant skill BEFORE taking any action — including asking clarifying questions.
If there is even a 1% chance a skill applies, invoke it. Skills override default behavior. User instructions override skills.
Instruction priority:
- User's explicit instructions — highest
- Agentic SDLC skills — override default agent behavior
- Default agent behavior — lowest priority
Each stage is a skill — a SKILL.md file the agent reads and follows. Skills contain:
- A
description(YAML frontmatter) that tells the agent when to invoke it - Behavioral instructions, checklists, and output formats
- Anti-rationalization "Red Flags" tables that help agents resist skipping discipline
The agent loads AGENTS.md at session start, which instructs it to check for relevant skills before any action.
In this source repository, the repo-root AGENTS.md is for contributors working on agentic-sdlc itself. The packaged project-root AGENTS.md that framework users receive via asdlc init lives at src/agentic_sdlc/core/AGENTS.md.
The framework includes a built-in dashboard to monitor project progress.
# Start the local dashboard
asdlc serve .See docs/installation.md for the supported install paths, verification steps, and troubleshooting notes.
After installing, bootstrap your project with:
asdlc initFor contributors and CI, a separate asdlc-dev CLI exposes eval and validation tools. It is installed with the package entrypoints, including standard installs and editable dev installs:
# Validate all SKILL.md files for structural correctness
asdlc-dev validate-skills .
# Run deterministic scenario fixtures against skills
asdlc-dev eval-skills .
# Eval a single skill
asdlc-dev eval-skills . --skill asdlc-implementationasdlc-dev also includes all public commands (init, serve).
| Skill | Stage | Trigger |
|---|---|---|
asdlc-using-agentic-sdlc |
Meta | Starting any session |
asdlc-inception |
1 | New project with unclear requirements |
asdlc-design-system |
2 | Establishing visual/interaction language |
asdlc-ui-mockups |
2a | User-facing redesign, new landing page, or major visual contract change |
asdlc-tech-architecture |
3 | Making technology or architecture decisions |
asdlc-coding-constitution |
3a | Establishing coding standards |
asdlc-implementation-planning |
4 | Creating execution plan from approved architecture |
asdlc-story-breakdown |
5 | Decomposing plan into executable tasks |
asdlc-implementation |
6 | Writing production code |
asdlc-critical-review |
7 | Adversarial quality review of completed implementation |
asdlc-testing |
8 | Executing test plan and verifying end-to-end behavior |
asdlc-code-review |
9 | Final structured review before merge |
asdlc-retrospective |
10 | Closing a project or story cycle |
| Skill | Stage | Trigger |
|---|---|---|
asdlc-context-harvest |
0 | Starting on an unfamiliar existing codebase |
asdlc-brownfield-brainstorm |
1 | Understanding business impact of a story |
asdlc-brownfield-design |
2 | Story introduces new UI |
asdlc-ui-mockups |
2a | UI redesign or major visual contract change needs approval |
asdlc-brownfield-tech-plan |
3 | Planning technical approach for a story |
| (Stages 4–10 same as greenfield) |
| Skill | Trigger |
|---|---|
asdlc-stage-gates |
Evaluating whether a stage's exit criteria are met |
asdlc-hitl-protocol |
Irreversible action or ambiguous decision needing human input |
asdlc-git-discipline |
Any git operation — branch creation, commits, merges |
asdlc-writing-skills |
Creating or editing skills |
Source repo (packaged skills live under src/agentic_sdlc/skills/):
agentic-sdlc/
├── src/agentic_sdlc/
│ ├── skills/ ← Packaged skills (asdlc- prefixed, 21 total)
│ ├── core/ ← Packaged entry-point docs copied into target projects
│ └── fixtures/ ← Packaged eval fixtures for developer workflows
└── docs/
├── agentic-sdlc-framework.md ← Full framework reference
├── getting-started.md
├── installation.md
├── workflow-greenfield.md
├── workflow-brownfield.md
├── skill-reference.md
├── platforms/ ← Per-platform setup guides (7 platforms)
└── future/ ← Deferred roadmap and unimplemented plans
├── README.md
├── future-platforms.md
├── multi-agent-orchestration-draft.md
├── release-plan.md
└── skill-evals-implementation-plan.md
Deferred roadmap items and non-canonical plans live under docs/future/. Canonical workflow artifacts such as docs/sdlc/epics/implementation-plan.md remain in their standard paths.
Installed project (after asdlc init, standard Agent Skills layout):
your-project/
├── .agents/
│ └── skills/ ← Natively discoverable by Codex, Gemini, Amp, Cursor
│ ├── asdlc-using-agentic-sdlc/SKILL.md
│ ├── asdlc-inception/SKILL.md
│ ├── asdlc-implementation/SKILL.md
│ └── ... (21 skills total)
├── AGENTS.md ← Copied from src/agentic_sdlc/core/AGENTS.md
└── docs/
docs/
architecture/domain-model.md, product/features/brd.md ← asdlc-inception
product/design-system.md, product/accessibility.md ← asdlc-design-system
product/mockups.md ← asdlc-ui-mockups
architecture/tech-architecture.md, architecture/adrs/ ← asdlc-tech-architecture
architecture/coding-standards.md ← asdlc-coding-constitution
sdlc/epics/implementation-plan.md, architecture/data-domain.md ← asdlc-implementation-planning
sdlc/epics/task-graph.md ← asdlc-story-breakdown
sdlc/stories/STORY-*.md ← asdlc-story-breakdown
sdlc/retrospectives/critical-review.md ← asdlc-critical-review
sdlc/test-plans/test-plan.md ← asdlc-testing
sdlc/retrospectives/retrospective.md ← asdlc-retrospective
architecture/existing-system.md ← asdlc-context-harvest (brownfield)
sdlc/workspaces/tech-plan-*.md ← asdlc-brownfield-tech-plan (per story)
| Platform | Status | Instructions |
|---|---|---|
| Codex | ✅ Available | docs/platforms/codex.md |
| Claude Code | ✅ Available | docs/platforms/claude.md |
| OpenCode | ✅ Available | docs/platforms/opencode.md |
| Cursor | ✅ Available | docs/platforms/cursor.md |
| Gemini CLI | ✅ Available | docs/platforms/gemini.md |
| Antigravity | ✅ Available | docs/platforms/antigravity.md |
| Amp | ✅ Available | docs/platforms/amp.md |
MIT