Skip to content

Latest commit

 

History

27 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Ultimate SDLC Framework for Claude Code

A structured, gate-driven, multi-council orchestration system for AI-assisted software development. Manages the full lifecycle from requirements through deployment: Planning → Development → Audit → Validation.

171 workflow commands, 247 knowledge skills, 19 specialist agents, 13 quality gates, 5 cycle types.

Install

git clone https://github.com/UnpaidAttention/claude-code-ultimate-sdlc-framework.git ~/ultimate-sdlc && cd ~/ultimate-sdlc && ./setup

The setup script copies the framework into the Claude Code plugin cache (~/.claude/plugins/cache/), registers it as a plugin, and installs agents and rules. The source repo at ~/ultimate-sdlc stays separate — safe to modify, update, and push without affecting the installed copy.

Restart Claude Code after setup. All /sdlc-* commands will be available.

Quick Start

In any Claude Code session:

/sdlc-init              # Initialize a new project
/sdlc-planning-start    # Begin the Planning Council
/sdlc-status            # Check current progress
/sdlc-continue          # Auto-advance to next step
/sdlc-help              # Full command reference

If you chose short naming mode during setup, drop the sdlc- prefix: /init, /planning-start, etc.


How It Works

The framework guides AI-assisted development through four sequential councils. Each council has phases or tracks, quality gates that block progression, and specialized workflow commands. You advance through each council in order — Planning produces specifications, Development implements them, Audit verifies quality, and Validation certifies release readiness.

The Four Councils

1. Planning Council — Requirements → Specifications

Transforms a product concept into implementation-ready specifications. Walks through discovery, architecture, feature specification, and AIOU (Atomic Implementation Output Unit) decomposition.

Phase Purpose Command
Phase 1 Discovery — understand the problem, users, constraints /sdlc-planning-start
Phase 1.5 Deliberation — multi-perspective analysis /sdlc-planning-phase-1-5
Phase 2 Architecture — system design, tech stack, data models /sdlc-planning-phase-2
Phase 2.5 Feature deep-dive — detailed analysis per feature /sdlc-planning-feature-deep-dive
Phase 3 Features — create FEAT-XXX specifications with acceptance criteria /sdlc-planning-phase-3
Phase 3.5 AIOU decomposition — break features into atomic implementation units /sdlc-planning-phase-3-5
Phases 4-7 Supporting specs — security, testing strategy, infrastructure, sprint planning /sdlc-planning-supporting-specs
Phase 8 Launch ready — final review and handoff preparation /sdlc-planning-phase-8

Gates: 1.5 (feature completeness), 3.5 (AIOU decomposition), 8 (launch ready)

2. Development Council — Specifications → Implementation

Implements the planned specifications wave by wave, from foundational types through UI to full integration.

Wave Purpose Command
Wave 1 Types, interfaces, utilities /sdlc-dev-wave-1
Wave 2 Data layer — models, schemas, migrations /sdlc-dev-wave-2
Wave 3 Services — business logic, domain operations /sdlc-dev-wave-3
Wave 4 API layer — endpoints, controllers, middleware /sdlc-dev-wave-4
Wave 5 UI components — pages, forms, interactions /sdlc-dev-wave5-start
Wave 6 Integration — end-to-end wiring, system tests /sdlc-dev-wave-6

Gates: I4 (services complete), I8 (pre-deployment ready)

Additional dev commands: /sdlc-dev-aiou (execute a single AIOU), /sdlc-dev-test (run tests), /sdlc-dev-checkpoint (save progress), /sdlc-dev-verify-feature (verify a feature)

3. Audit Council — Implementation → Quality Assessment

Independently tests and assesses the implementation for correctness, completeness, and quality.

Track Purpose Commands
Testing (T1-T5) Functional, GUI, integration, security, performance testing /sdlc-audit-t2 through /sdlc-audit-t5
Audit (A1-A3) Purpose alignment, completeness assessment, quality scorecard /sdlc-audit-a1 through /sdlc-audit-a3

Gates: T3 (GUI analysis), A2 (completeness), A3 (quality assessment)

Additional audit commands: /sdlc-audit-defect (log a defect), /sdlc-audit-report (generate report), /sdlc-audit-trace (verify spec-to-implementation traceability)

4. Validation Council — Assessment → Release Readiness

Validates corrections, hardens for production, polishes UX, and certifies release readiness.

Track Purpose Commands
Validation (V1-V5) Gap analysis, completeness, correction planning /sdlc-validate-v2 through /sdlc-validate-v5
Correction (C1-C4) Targeted fixes, edge cases, verification, regression /sdlc-validate-c1 through /sdlc-validate-c4
Production (P1-P4) Operational assessment, failure modes, performance, security hardening /sdlc-validate-p1 through /sdlc-validate-p4
Enhancement (E1-E4) Feature richness, innovation, UX polish /sdlc-validate-e1 through /sdlc-validate-e4
Synthesis (S1-S2) Documentation updates, final release certification /sdlc-validate-s1, /sdlc-validate-s2

Gates: V5, C4, P4, E4, S2 (release ready)


Quality Gates

13 binary pass/fail checkpoints that block progression until criteria are met. Each gate has a specific checklist — you can't skip or hand-wave past them.

Gate Council What It Checks
1.5 Planning Every requirement captured, no truncation, user confirmed scope
3.5 Planning All features decomposed into AIOUs with acceptance criteria
8 Planning Specs complete, handoff document ready for Development
I4 Development Backend services implemented, tests passing, security verified
I8 Development Full integration complete, no vulnerabilities, ready to hand off
T3 Audit GUI/UX analysis complete (skipped if no frontend)
A2 Audit All features tested, no untested code paths
A3 Audit Quality scorecard across 6 dimensions meets thresholds
V5 Validation All issues mapped, correction plan approved
C4 Validation All corrections applied and regression-tested
P4 Validation Production hardened — monitoring, rollback, security scans pass
E4 Validation UX polished — accessibility, visual consistency, flow verified
S2 Validation Final release gate — everything passes, documentation complete

Run any gate check with: /sdlc-planning-gate-1-5, /sdlc-dev-gate-i4, /sdlc-validate-gate-s2, etc.


Specialist Agents (19)

19 specialist agents are invoked automatically during workflow execution. Each carries deep domain expertise, embedded rules (anti-slop, integrity, security), and actionable patterns — not just evaluation questions. They are installed to ~/.claude/agents/ and wired into 138 of 167 workflow skills.

Analytical Lenses (8)

Applied during evaluation, review, and gate verification phases.

Agent Expertise When Invoked
Architecture System design patterns, ADRs, tech stack evaluation, scalability Planning phases 2-3, dev wave setup, Audit A3
Security OWASP Top 10, STRIDE/DREAD, secrets scanning, hardening All gates, every AIOU, Audit T5, Validation P4
Quality Code review checklists, quality scoring (6 dimensions), anti-patterns Dev code review, Audit A3, Validation C1-C4
Performance Web Vitals targets, load testing, profiling, query optimization Audit T5, Validation P1-P3
UX WCAG 2.2 AA, Nielsen's heuristics, anti-slop visual rules Audit T3, Validation E4, Wave 5 (frontend only)
Operations Deployment checklists, FMEA, monitoring, runbooks, SLOs Validation P1-P2, deployment workflows
Requirements Elicitation, INVEST criteria, scope-lock enforcement, traceability Planning phases 1-1.5, gate reviews, Audit A1-A2
Documentation API docs (OpenAPI), changelogs, handoff templates, ADRs Validation S1, all handoff generation

Action Specialists (11)

Invoked during implementation, testing, debugging, and correction phases.

Agent Expertise When Invoked
Planner Feature decomposition, AIOU creation, batch planning, sprint orchestration Planning phases 1-8, correction planning
Code Reviewer AIOU-aware review against acceptance criteria, anti-slop code rules Every AIOU completion, post-correction review
TDD Guide Test-first enforcement, Red-Green-Refactor, coverage analysis, test quality Every AIOU in waves 1-6, test failures, corrections
Database Specialist Schema design, migrations, indexing, N+1 prevention, query optimization Wave 2, Audit T4, Validation P3
API Designer REST conventions, all HTTP status codes, OpenAPI, middleware, rate limiting Wave 4, planning phase 3.5 API spec
Frontend Specialist Component architecture, state management, anti-slop visual rules, a11y Wave 5, UI audit/polish/retheme, Validation E4
Backend Specialist Service patterns, DI, clean architecture, error handling, domain logic Wave 3, Audit A3
Build Resolver Minimal-diff build fixes, no architectural drift, PRH-002/003 compliance Any build failure across all waves
Gate Keeper All 13 gate criteria, evidence-based verification, binary PASS/FAIL All gate verification skills
Debugger 4-phase root cause analysis, hypothesis testing, "3+ fixes" escalation /debug, Validation C1 corrections
Integration Tester E2E test design, 8-layer verification, API contracts, regression strategy Wave 6, Audit T4, Validation C3-C4

Agent Pipeline Per Wave

During development, each AIOU follows a specialist pipeline:

Wave Pipeline
Wave 1 (Types/Utils) TDD Guide → Code Reviewer
Wave 2 (Data Layer) Database Specialist → TDD Guide → Code Reviewer
Wave 3 (Services) Backend Specialist → TDD Guide → Code Reviewer
Wave 4 (API Layer) API Designer → TDD Guide → Code Reviewer + Security
Wave 5 (UI) Frontend Specialist → UX → Code Reviewer
Wave 6 (Integration) Integration Tester → Code Reviewer

Technical detail: Agent definitions live at ~/.claude/agents/sdlc-*.md and in the plugin's agents/ directory. Each agent carries 200-500 lines of domain expertise with embedded framework rules.


Governance Modes

Automatically selected based on feature count (or manually overridden):

Mode Feature Count What Changes
Lightweight <8 features Abbreviated pipeline — 7 gates active, combined planning phases, simplified audit
Standard 8-25 features Full pipeline — all 13 gates, multi-run support, full handoff documents
Enterprise 25+ features Full pipeline + formal ADRs, mandatory multi-run, external security/penetration testing required

Override during init: /sdlc-init --mode enterprise


Cycle Types

Not every change needs the full pipeline. Choose the right cycle type for your work:

Type Use Case What It Runs Command
Full Initial build, major versions All 4 councils, full planning /sdlc-init
Feature Adding 1-5 features All 4 councils, full planning /sdlc-new-cycle
Patch Bug fixes Planning (abbreviated) → Dev → Validation /sdlc-patch-planning
Maintenance Dependency updates, security patches, infra Planning (abbreviated) → Dev → Validation /sdlc-maintenance-planning
Improvement Refactoring, performance, tech debt All 4 councils, abbreviated planning /sdlc-improvement-planning

Cycle management: /sdlc-close-cycle archives the current cycle, /sdlc-new-cycle starts fresh.


Knowledge Skills (245)

The framework includes 245 reference knowledge skills covering patterns, best practices, and domain expertise. These aren't invoked as slash commands — they're automatically loaded by workflow skills when relevant context is needed.

Categories include: clean code, TDD, architecture patterns, security, accessibility, API design, database patterns, performance optimization, and many more.

Browse the full index in the knowledge/ directory.


Key Concepts

AIOUs (Atomic Implementation Output Units)

The smallest unit of work in the framework. Each feature is decomposed into AIOUs during Planning Phase 3.5. Each AIOU has:

  • A specific wave assignment (1-6)
  • Clear acceptance criteria
  • Estimated effort
  • Dependencies on other AIOUs

Handoffs

Formal documents that transfer work between councils. Each handoff is validated against a schema to ensure nothing is lost:

  • Planning → Development: Feature specs, AIOUs, architecture decisions
  • Development → Audit: Implementation summary, test results, known issues
  • Audit → Validation: Quality assessment, defect log, recommendations
  • Validation → Release: Final certification, deployment checklist

State Management

Per-project state persists in .ultimate-sdlc/ across sessions:

  • project-context.md — Current council, phase, and status (the "cursor")
  • project-manifest.md — Persistent identity that survives cycle archives
  • council-state/ — Per-council working memory and progress
  • specs/ — Feature specs (FEAT-XXX), AIOUs (AIOU-XXX), ADRs
  • progress.md — Append-only session history
  • feedback/ — User corrections with reasoning (see next section)
  • framework-revisions-proposed/ — Framework edit proposals awaiting user review

CLI utilities: sdlc-state current (show position), sdlc-state advance (move forward), sdlc-state gate-check (verify gate)


Feedback-Driven Learning

The framework captures user corrections with reasoning as persistent feedback entries, separate from specs. Requirements describe what the system does; feedback describes how the team wants work done. They are orthogonal — gates catch what you know can break, feedback catches what the project teaches you over time.

Storage

.ultimate-sdlc/feedback/
├── INDEX.md                          # Authoritative index grouped by status
├── FB-001-<slug>.md                  # Individual entries (frontmatter + body)
├── ...
└── REJECTED.md                       # Log of entries rejected at write time

.ultimate-sdlc/framework-revisions-proposed/
├── INDEX.md
├── FR-001-<slug>.md                  # Proposed edits to framework files
└── ...

Four Feedback Types

Type When to capture
user-correction User corrects agent output AND provides reasoning
user-preference User states a preference unprompted by an error
gate-learning A quality gate fails because of a rule/process gap
pattern Cycle-end synthesis of ≥2 recurring entries

Workflow Commands

Command Purpose When
/sdlc-feedback-log Capture a new feedback entry with required "why" reasoning Anytime the user corrects or states a preference
/sdlc-feedback-review Surface active entries matching current context Session start, pre-AIOU, pre-gate (often auto-invoked)
/sdlc-feedback-promote Cluster recurring entries into pattern entries Validation Council S1 (end of cycle)
/sdlc-framework-retro Draft proposed framework edits from patterns /sdlc-close-cycle Step 6b

Guardrails

  • Every entry requires the user's reasoning. No "why" → no entry. A correction without reasoning is patching, not learning (FBP-001).
  • Feedback cannot bypass INTEGRITY-RULES. Entries that would enable PRH-001..PRH-009 violations are rejected at write time and logged to REJECTED.md (FR-2, FBP-004).
  • Patterns require ≥2 corroborating entries — no single-entry promotions (FBP-005).
  • Framework self-modification is proposal-only. /sdlc-framework-retro writes proposals to the project under framework-revisions-proposed/. Agents never write to the framework repo directory. The user reviews and applies manually (FR-3).
  • Feedback is per-project. It does not propagate across projects (FR-4).

Integration Points

Feedback is read at:

  • Every council's Session Protocol (/sdlc-feedback-review auto-invoked)
  • Before any AIOU in Planning phases 2.5 / 3 / 3.5 and all Development waves
  • Before every gate verification
  • At /sdlc-new-cycle bootstrap (carries forward pattern entries)

Feedback is written at:

  • User correction or stated preference (any time)
  • Gate failure caused by rule/process gap
  • Validation S1 (pattern synthesis)

See rules/feedback-rules.md for full governance and contexts/feedback-schema.md for schema.


UI Development Workflows

Special workflows for frontend projects with comprehensive UI lifecycle support:

Workflow Purpose Command
UI Research Study design patterns, competitor analysis /sdlc-dev-ui-research
UI Design Plan Page layouts, component hierarchy, interaction maps /sdlc-dev-ui-design-plan
UI Audit Find gaps in existing UI (6 phases) /sdlc-dev-ui-audit
UI Polish Remove AI slop, improve visual quality (6 phases) /sdlc-dev-ui-polish
UI Retheme Complete visual makeover (5 phases) /sdlc-dev-ui-retheme
UI Redesign Full restart of UI development /sdlc-dev-ui-redesign

Each maintenance workflow has individually invocable phases (e.g., /sdlc-dev-ui-audit-scan, /sdlc-dev-ui-audit-gaps, /sdlc-dev-ui-audit-fix).


Project Structure

Per-project state lives in .ultimate-sdlc/ within your project directory:

.ultimate-sdlc/
├── project-manifest.md          # Persistent identity (survives cycles)
├── project-context.md           # Current council/phase/status
├── progress.md                  # Append-only session history
├── config.yaml                  # Project-specific config overrides
├── council-state/
│   ├── planning/current-state.md
│   ├── development/current-state.md
│   ├── audit/current-state.md
│   └── validation/current-state.md
├── specs/
│   ├── features/FEAT-001.md     # Feature specifications
│   ├── aious/AIOU-001.md        # Atomic implementation units
│   ├── adrs/ADR-001.md          # Architecture decision records
│   └── scope-lock.md            # Canonical in-scope feature list
├── feedback/                    # User corrections with reasoning (per-cycle)
│   ├── INDEX.md
│   ├── FB-001-<slug>.md
│   └── REJECTED.md
├── framework-revisions-proposed/ # Framework edit proposals (cross-cycle, user-reviewed)
│   ├── INDEX.md
│   └── FR-001-<slug>.md
├── handoffs/                    # Council transition documents
└── .cycles/                     # Archived cycle state

Plugin Structure

~/ultimate-sdlc/                   # Source repo (your working copy)
├── .claude-plugin/                # Plugin manifest (plugin.json, marketplace.json)
├── setup                          # Installation script
├── SKILL.md                       # Root skill (framework overview + routing)
├── commands/                      # 167 slash command stubs (/sdlc-init, etc.)
├── skills/                        # 167 workflow skills (full instructions)
├── knowledge/                     # 245 reference knowledge skills
├── agents/                        # 19 specialist agent definitions
├── contexts/                      # Gate criteria, governance modes, state management
├── rules/                         # Anti-slop, integrity, council rules (9 files)
├── bin/                           # CLI utilities (sdlc-config, sdlc-state, etc.)
├── templates/                     # Project initialization templates
└── scripts/                       # Generator scripts

Installed to: ~/.claude/plugins/cache/ultimate-sdlc/ultimate-sdlc/3.1.0/

Configuration

Global config: ~/.ultimate-sdlc/config.yaml

# View all settings
sdlc-config list

# Change settings
sdlc-config set governance_mode enterprise
sdlc-config set project_type api-service

Supported project types: web-app, cli-tool, library, api-service, ml-pipeline, mobile-app

Naming Modes

During setup you choose how commands appear:

  • Namespaced (default, recommended): /sdlc-init, /sdlc-planning-start, /sdlc-dev-wave-1
  • Short: /init, /planning-start, /dev-wave-1

Namespaced mode is recommended to avoid conflicts with other plugins.

Update

cd ~/ultimate-sdlc && git pull && ./setup

Uninstall

~/.claude/plugins/cache/ultimate-sdlc/ultimate-sdlc/3.1.0/bin/sdlc-uninstall

Requirements

License

MIT

About

Antigravity SDLC Framework for Claude Code — structured, gate-driven AI-assisted software development with 167 workflow commands

Resources

Stars

6 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages