Skip to content

MCKRUZ/claude-code-sdlc

Repository files navigation

claude-code-sdlc

SDLC orchestration plugin for Claude Code. Drives projects through a structured software development lifecycle — gated opening and closing phases around a continuous Build loop — with company-configurable profiles, compliance gates, and quality enforcement.

Why This Exists

No existing tool combines specification-driven development + quality enforcement + compliance gates + company configurability into one executable workflow. This plugin makes the AI-SDLC methodology executable inside Claude Code.

Features

  • 9 SDLC phases — Discovery, Requirements, Design, Foundation, the continuous Build loop, Documentation, Deployment, Monitoring, Close & Transfer. The gap where batch Implementation/Quality/Testing used to be is intentional — checking happens per change inside the Build loop, never as a later batch phase.
  • Project type routing — testing, deployment, and monitoring adapt automatically based on project_type (service, app, library, skill, cli)
  • Company profiles — YAML configs for stack, quality thresholds, compliance, conventions
  • 7-gate validation — Integrity, completeness, metrics, compliance, consistency, quality checks at every transition, plus the phase's own exit criteria rendered for the human who signs
  • Frozen layers — Token-efficient phase summaries enabling cross-phase context continuity
  • 3-tier context architecture — Foundation (always), frozen layers (per-phase), references (on-demand)
  • Narrative enhancement — Optional stakeholder-friendly .narrative.md companions for technical artifacts
  • Conversational coaching — Adaptive dialogue mode for guided phase completion
  • NFR measurement basis enforcement — Every numeric threshold must declare its measurement source; aspirational thresholds require an agreed validation plan in the Build loop
  • Test traceability — the Build loop requires scenario-to-requirement mapping and redundancy audit before test execution
  • Compliance enforcement — SOC 2, HIPAA, GDPR, PCI-DSS gate definitions
  • Skill orchestration — Maps phases to existing skills (/deep-plan, /deep-implement, /tdd, /code-review, /security-review, /e2e)
  • State machine — Tracks progress in .sdlc/state.yaml with full audit trail
  • Artifact checksums + dirty tracking — SHA-256 hashes detect changed artifacts; gate checks skip unchanged files for faster validation
  • Smart repair — Gate failures on structural issues (missing templates, placeholders) are auto-fixed before escalating to human
  • Dependency graph enforcement — Section dependency DAG validated in the Build loop; circular dependencies detected; implementation order enforced
  • Multi-perspective review/sdlc-review with adversarial, edge-case, and council modes for thorough artifact review
  • Brainstorming techniques — SCAMPER, reverse brainstorming, constraint removal, and 3 more structured ideation methods with anti-bias protocol
  • Brownfield detection — Auto-detects existing codebases and generates workspace analysis before Discovery begins
  • Structured error specs — P0/P1 requirements must define Accepts/Returns/Errors explicitly
  • Cross-artifact validation — File-level reference checking between artifacts (SHOULD severity)
  • Question-to-file audit trail — Open questions written to persistent markdown files for team collaboration
  • Test expansion protocol — Healer (diagnose failures), Expander (add edge cases), bug-to-test pipeline (RED first)
  • Parallel review resolution — Batch review findings by file, resolve in parallel, re-verify after fix
  • Advanced elicitation — Constraint analysis, inverse thinking, pre-mortem, 5 whys, and more for coaching mode
  • Phase-aware context — Hooks inject current phase reminders and conventions
  • Session health check — Opt-in pre-flight build verification at session start (Build loop) catches broken builds before the agent compounds them with new work
  • Visual regression testing — Optional screenshot capture during Build-loop E2E testing with baseline comparison (manual or pixel-diff)
  • Single-change guardrail — Explicit constraint preventing simultaneous work on multiple specs in the Build loop unless explicitly parallelizable
  • Phase-scoped evaluation criteria — Quality rubrics that apply to non-code artifacts (requirements, design, foundation) in addition to code
  • Empirical metrics logging — JSONL instrumentation in gates, frozen layer validation, and section evaluation for evidence-based harness optimization
  • Document intake — Opt-in Phase 0 corpus analysis for external reference materials (RFPs, API specs, vendor docs, compliance handbooks) with per-document summaries, DOC-NNN traceability IDs, token-budgeted session-start index (Tier 1.5), and Phase 1 requirement-to-source linking

Installation

Via the plugin marketplace (recommended)

One marketplace add, one install — brings the orchestration commands and the delivery harness:

/plugin marketplace add MCKRUZ/claude-code-sdlc
/plugin install claude-code-sdlc@mckruz

Then, per project: /sdlc-setup initializes .sdlc/ and installs the full delivery harness (governance CLAUDE.md, .claude/{settings,hooks,agents,skills}, 7 CI workflows in .github/workflows/ — 5 rail gates plus 2 eval workflows — of which five checks (build-and-test, spec-gate, grader, correctness-review, security-review) block at merge via the branch-protection ruleset, and infra/ starters). To add or refresh just the harness in an existing repo, use /sdlc-harness. Requires Claude Code v2.1.196+ (source: ".").

As a Claude Code Skill (local dev)

# Clone the repo
git clone https://github.com/MCKRUZ/claude-code-sdlc.git

# Symlink to your Claude Code skills directory
# Windows
mklink /D "%USERPROFILE%\.claude\skills\claude-code-sdlc" "path\to\claude-code-sdlc"

# macOS/Linux
ln -s /path/to/claude-code-sdlc ~/.claude/skills/claude-code-sdlc

Dependencies

The plugin uses Python scripts via uv:

# Install uv (if not already installed)
pip install uv
# or
brew install uv

# Install script dependencies
cd claude-code-sdlc/scripts
uv sync

Quick Start

1. /sdlc-setup          → Select profile, initialize .sdlc/ AND install the delivery harness
2. /sdlc                → See Phase 0: Discovery guidance
3. Create artifacts     → Write problem-statement.md in .sdlc/artifacts/00-discovery/
4. /sdlc-gate           → Check if exit criteria are met
5. /sdlc-next           → Advance to Phase 1: Requirements
6. /sdlc-status         → View progress dashboard

Documentation

For in-depth technical documentation, see the guides in docs/:

Guide Description
Architecture Plugin anatomy, component relationships, data flow diagrams, two-directory model, progressive disclosure strategy
Phase Lifecycle All 9 phases in depth — workflows, artifacts, HITL gates, skills, agents, handoff protocol, project type adaptations
Gate System 7-gate validation — integrity, completeness, metrics, compliance, consistency, quality, exit criteria — severity levels, override protocol
Profiles Schema reference (every field), built-in profiles, custom profile creation, compliance framework integration, evaluation criteria
Commands All 21 slash commands — internal flow, state changes, Python scripts called, error scenarios, examples
Agents 13 custom agents + built-in subagent orchestration, phase-to-agent mapping, parallel execution rules, mandatory spawns
State Machine state.yaml format, transition rules, history tracking, session-handoff.json, the spec backlog
Templates & Artifacts Template directory structure, per-phase artifact details, handoff document protocol, artifact lifecycle
Scripts All Python scripts (incl. phase_model.py, the phase-identity source of truth) — CLI args, inputs/outputs, exit codes, gate implementation details, uv runtime
Integrations How /deep-plan, /deep-implement, /tdd, /code-review map into SDLC phases, artifact transformation pipeline
Hooks Session-start and phase-inject hooks — what they read, what they inject, session continuity, convention reminders

Commands

Command Purpose
/sdlc-setup Interactive setup wizard — select profile, initialize project
/sdlc Show current phase guidance, next action, required artifacts
/sdlc-status Progress dashboard with phase table and completion %
/sdlc-gate Run 7-gate exit criteria check (does not advance)
/sdlc-next Advance to next phase if all MUST gates pass
/sdlc-enhance Generate narrative companions for stakeholder review (optional)
/sdlc-coach Interactive coaching mode — adaptive dialogue for current phase
/sdlc-intake Catalog and summarize an external document corpus (Phase 0, opt-in)
/sdlc-brief Prep a stakeholder workshop brief from the document corpus
/sdlc-spec Author a ready Build-loop spec (specs/NNNN-name.md) and enforce the Definition of Ready
/sdlc-phase-report Generate phase HTML report with artifact inventory
/sdlc-review Multi-perspective artifact review (council, adversarial, or edge-case modes)
/sdlc-audit Analyze gate effectiveness across completed phases
/sdlc-feature Decompose an epic into channel-aware features and specs (feature-brief.md)
/sdlc-experience Author per-channel journeys, surface layout, and the channel interaction spec
/sdlc-data Author the data contract, data-readiness, and lineage/audit (Data)
/sdlc-rules Capture business rules (BR-NN) and golden scenarios (SCEN-NN) (Bizreq)
/sdlc-channel Bind a spec's channel: and inject that channel's acceptance dimensions
/sdlc-evals Author a versioned golden set next to a spec (wraps the harness eval-builder)

Profiles

microsoft-enterprise

Full enterprise stack with compliance:

  • Stack: C#/.NET 8, Angular 17, SQL Server, Azure
  • Quality: 80% coverage minimum, 100% critical paths, TDD required
  • Compliance: SOC 2 gates at every phase transition
  • Conventions: Conventional commits, immutable patterns, no console.log
  • Health check: dotnet build --no-restore --verbosity quiet at session start (Build loop)
  • Visual verification: Screenshot capture with manual baseline comparison
  • Evaluation criteria: Phase-scoped rubrics for requirements (testability, traceability), design (ADR completeness, interface specificity), foundation (section plan verifiability), and Build-loop code (Result pattern, immutable state, FluentValidation, API docs)

starter

Minimal profile for quick start:

  • Stack: Configurable (defaults to TypeScript/Node)
  • Quality: 60% coverage minimum, no TDD requirement
  • Compliance: None
  • Conventions: Conventional commits

creative-tooling

Python profile for creative-pipeline tooling (ComfyUI technique registry / inventory projects):

  • Stack: Python (uv-run scripts), pytest
  • Quality: 80% coverage minimum, TDD + code review + security review required, 400-line file cap
  • Compliance: None (audit trail kept)
  • Evaluation criteria: Registry/profile schema compliance, skill prompt clarity, inventory resolution via inventory.resolve(), cross-reference integrity, data freshness

Creating Custom Profiles

Copy profiles/starter/profile.yaml and modify for your stack. Validate with:

uv run scripts/validate_profile.py profiles/my-profile/profile.yaml

See profiles/_schema.yaml for the full schema.

Project Types

Phase 0 captures project_type, which is stored in state.yaml and controls how testing (in the Build loop), Deployment, and Monitoring execute:

Type Description Testing (Build loop) Deployment (8) Monitoring (9)
service Backend API / server process Unit + integration + E2E + API contract tests Staging server deploy + DB migrations Full infrastructure monitoring + alerting
app User-facing application with UI Same as service Same as service Same as service
library Shared code package / SDK Unit tests for public API surface Package registry publish Download counts + issue triage
skill Claude Code skill / AI plugin Scenario-based testing against requirement list File distribution + install verification Qualitative feedback channels + issue triage
cli Command-line tool Unit + integration + invocation tests Binary distribution Same as library

SDLC Phases

# Phase Primary Skills Key Artifacts
0 Discovery /plan problem-statement.md, constitution.md (incl. project_type)
1 Requirements /deep-project requirements.md, non-functional-requirements.md (with Measurement Basis)
2 Design /deep-plan design-doc.md, ADRs
3 Foundation /deep-plan foundation-report.md, risk-tier-map.md, cadence-plan.md, build-handoff.md
build Build Loop /tdd, /code-review, /security-review, /e2e specs/NNNN-*.md, phase7-handoff.md (no batch gate — human declares feature-complete)
7 Documentation /update-docs README, RUNBOOK
8 Deployment CI/CD release notes (approach varies by project_type)
9 Monitoring Manual monitoring config (approach varies by project_type)
close Close & Transfer final-handoff-report.md, harness-audit.md, close-gate-evidence.md, access-revocation-checklist.md

7-Gate Validation System

Every phase transition runs through seven gates:

  1. Integrity — Required artifacts exist and are well-formed
  2. Completeness — Artifacts contain all required sections, no placeholders
  3. Metrics — Quantitative thresholds met (coverage, file size)
  4. Compliance — Correct labeling (priorities, ADR status, compliance mapping)
  5. Consistency — Locked metrics checked against frozen layers from prior phases
  6. Quality — Phase-scoped evaluation criteria from profile (e.g., requirement testability in Phase 1, ADR completeness in Phase 2, code patterns in the Build loop)
  7. Exit criteria — The phase's own prose exit conditions from the registry, rendered as review items for the human who signs the advance

Gates 1–6 pass or fail. Gate 7 always reports REVIEW: its conditions are judgments ("The rails are proven, not just present") that no script can settle, so it shows the approver the checklist rather than deciding for them. Phases 0–2 declare no prose conditions, so Gate 7 is silent there.

All gate results are logged to .sdlc/metrics/gate-log.jsonl for empirical tracking of gate effectiveness.

Gates have severity levels:

  • MUST — Blocks transition if failed
  • SHOULD — Generates warning
  • MAY — Informational

Project Structure

claude-code-sdlc/
├── plugin.json              # Plugin manifest
├── SKILL.md                 # Main skill entry point
├── commands/                # 21 slash commands (/sdlc, /sdlc-setup, /sdlc-status, /sdlc-next, /sdlc-gate, /sdlc-enhance, /sdlc-coach, /sdlc-review, /sdlc-intake, /sdlc-brief, /sdlc-spec, /sdlc-phase-report, /sdlc-audit, /sdlc-feature, /sdlc-experience, /sdlc-data, /sdlc-rules, /sdlc-channel, /sdlc-evals, /sdlc-harness, /sdlc-upgrade)
├── agents/                  # 13 agents (orchestrator, requirements-analyst, compliance-checker, section-evaluator, narrative-enhancer, gate-repair, multi-reviewer, discovery-analyst, feature-architect, visual-designer, conversation-designer, data-analyst, bizreq-analyst)
├── profiles/                # Company/stack YAML profiles
├── channels/                # Channel descriptor library (ag-ui, voice, chat) + schema
├── phases/                  # Phase definitions (0,1,2,3,build,7,8,9,close)
├── references/              # Progressive disclosure docs
├── templates/               # Artifact templates
├── scripts/                 # Python automation (uv)
└── hooks/                   # PowerShell hooks for context injection

Architecture

This plugin follows ADR-001: claude-code-sdlc is a Claude Code plugin that executes AI-SDLC methodology.

  • AI-SDLC = the methodology (phases, validation, templates)
  • claude-code-sdlc = the execution engine (commands, hooks, profiles, scripts)
  • The plugin references AI-SDLC patterns but adapts them for Claude Code's command/hook/agent system

Contributing

  1. Fork the repo
  2. Create a feature branch: feat/my-feature
  3. Follow the conventional commit format: type: description
  4. Submit a PR with a clear description

License

MIT

About

Claude Code plugin that orchestrates the full SDLC lifecycle using company-configurable profiles

Resources

License

Stars

2 stars

Watchers

0 watching

Forks

Packages

 
 
 

Contributors