---

An extension for GitHub Spec Kit that enforces the V-Model methodology: every development specification has a simultaneously generated, paired testing specification with full traceability.

The AI drafts. The human decides. The scripts verify. Git remembers.

Why

AI-native teams ship fast but produce no traceability. Regulated teams have full traceability but move too slowly. The V-Model Extension Pack closes this gap — from a single specification, it generates traceable requirements, paired test plans, hazard analysis, and a deterministic traceability matrix in minutes, not months.

Features

Category	Commands
Specification	requirements · system-design · architecture-design · module-design
Test Planning	acceptance · system-test · integration-test · unit-test
Cross-Cutting	hazard-analysis · impact-analysis · peer-review
Verification	trace · test-results · audit-report
Bridge to Implementation	plan · tasks · implement

17 commands across 4 V-Model levels plus the bridge to implementation, with deterministic coverage validation, 5 traceability matrices (A–D + H), an 8-stage validation gate (run-v-model-gate.sh), 4 lifecycle hooks, and compliance gating for regulated industries.

Quick Start

# Install the extension
specify extension add v-model \
  --from https://github.com/leocamello/spec-kit-v-model/archive/refs/tags/v0.7.0.zip

# Generate requirements from your spec
/speckit.v-model.requirements

# Generate paired acceptance tests (100% coverage validated by script)
/speckit.v-model.acceptance

# Build the traceability matrix
/speckit.v-model.trace

That's Level 1. Go deeper with system-design → system-test → architecture-design → integration-test → module-design → unit-test, running trace after each pair.

Built for Regulated Industries

Standard	Domain	Use Case
IEC 62304	Medical Devices	Software safety classes A/B/C
ISO 26262	Automotive	ASIL A–D functional safety
DO-178C	Aerospace	DAL A–E design assurance

Configure your domain in v-model-config.yml:

domain: iec_62304  # or iso_26262, do_178c

The Core Principle

Scripts handle all deterministic logic — coverage calculations, matrix generation, gap detection. AI handles creative translation — turning specifications into structured requirements and test scenarios. Humans review every artifact. Git remembers everything.

What	Who
Generate requirements & test plans	AI + Human review
Coverage calculation & matrix generation	Deterministic scripts
Quality evaluation	LLM-as-judge (advisory)
Audit trail	Git (cryptographic hashes)

Compliance & Hybrid Modes

The extension supports two modes of operation. Pick one and be explicit about it in your repository's README and CI configuration — this is the difference between a compliant audit trail and a prototyping shortcut.

Compliant mode (recommended for regulated work)

Use the V-Model bridge end-to-end:

/speckit.v-model.plan       →  plan.md (V-Model-enriched, schema-validated)
/speckit.v-model.tasks      →  tasks.md (TDD-ordered, with `Implements` headers)
/speckit.v-model.implement  →  source + tests + hallucination guard + trace post-hook

Every step runs the V-Model gates (run-v-model-gate.sh: status → domain → matrix → 5 coverage validators), the deterministic Implements-directive hallucination guard, and the trace post-hook. Artifacts under specs/<feature>/v-model/ must carry **Status**: Approved (enforced by validate-artifact-status.sh since v0.7.0). This is the path designed for regulated work targeting IEC 62304, ISO 26262, or DO‑178C.

Hybrid mode (prototyping only)

It is technically possible to feed a tasks.md produced by /speckit.v-model.tasks to core /speckit.implement — the schema round-trips. This is a hybrid mode that bypasses the V-Model gates, the Implements-directive hallucination guard, and the trace post-hook. Use it only for prototyping or for non-safety-critical features inside an otherwise-V-Model repository.

Audit guidance: a build that ran /speckit.implement against a V-Model tasks.md is not evidence of V-Model compliance, regardless of how the artifacts look. The deterministic gates left no record because they were not invoked.

v0.8.0 will introduce a compliance_mode: strict profile (see EPIC-1 in the v0.8.0 backlog) that makes this distinction mechanically detectable in CI.

Testing

Layer	Tests	What it validates
BATS	455	Bash script logic across all script files
Pester	431	PowerShell script parity
Structural	89	ID format, template conformance, section completeness
LLM evals	53	Requirements quality, BDD quality, traceability
End-to-end (E2E)	32	Bash + golden-output fixtures (advisory in v0.7.0; gated in v0.8.0 via EPIC-7)

📖 Documentation

Full documentation, tutorials, and guides are available at the project website.

• Getting Started — Install and create your first V-Model project in 15 minutes
• Guides — V-Model concepts, level-by-level guides, CI integration
• Compliance — IEC 62304, ISO 26262, DO-178C artifact mapping
• Tutorials — End-to-end walkthroughs for medical, automotive, and aerospace
• Reference — All 17 commands, ID schema, scripts, templates, configuration
• Contributing — Development setup, testing, PR process

License

MIT