By sc30gsw
Verified Coherence Spec-Driven Development — adversarial quality gates for AI-assisted development
Run Phase 3 (adversarial review) for the active VCSDD feature. Spawns a fresh vcsdd-adversary agent with zero Builder context to review implementation against spec. Produces binary PASS/FAIL verdict per dimension.
Run BFS forward-impact analysis from changed nodes through the CEG.
Rebuild the Conditioned Evidence Graph (CEG) from `coherence:` or `codd:`
Validate the CEG for reference integrity and circular dependencies.
Create a git commit for the current VCSDD phase. Auto-generates a conventional commit message from state.json including phase, sprint, gate verdict, and affected bead IDs. Tags with vcsdd/<feature>/phase-<id>.
VCSDD adversarial reviewer with fresh context. Use this agent ONLY for Phase 1c (spec review), strict-mode contract review, and Phase 3 (implementation review). This agent must be spawned as a NEW instance with zero Builder context. It reviews artifacts from disk and writes verdict/findings only under the feature review output directory. It produces binary PASS/FAIL verdicts per dimension with concrete findings.
VCSDD spec author and TDD implementer. Use this agent to write behavioral specifications, generate failing tests (Red phase), implement code to pass tests (Green phase), refactor code, write sprint contracts, and create verification architectures. Invoked for all code-writing and spec-writing tasks in the VCSDD pipeline.
VCSDD pipeline state manager and gate enforcer. Use this agent to coordinate the overall VCSDD workflow, track phase transitions, manage sprint contracts, and enforce quality gates. Invoke when you need to advance the pipeline, check gate prerequisites, route adversary feedback, or manage the .vcsdd/ state directory.
VCSDD formal verification coordinator. Use this agent for Phase 5 (Formal Hardening). It reads the language profile, detects installed verification tools, writes proof harnesses, runs proof/security/purity checks, and produces verification-report.md, security-report.md, and purity-audit.md. Invoked by /vcsdd-harden command.
Use this skill when running or interpreting VCSDD adversarial reviews. Provides calibration guidance for the Adversary agent, finding severity classification, and anti-leniency enforcement patterns.
Run Phase 3 (adversarial review) for the active VCSDD feature. Spawns a fresh vcsdd-adversary agent with zero Builder context to review implementation against spec. Produces binary PASS/FAIL verdict per dimension.
Given a set of changed or affected spec nodes, perform BFS forward-impact
Rebuild the Conditioned Evidence Graph (CEG) from `coherence:` or `codd:`
Validate the CEG for structural integrity:
Executes bash commands
Hook triggers when Bash tool is used
Modifies files
Hook triggers on file write and edit operations
Uses power tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Uses Bash, Write, or Edit tools
Uses Bash, Write, or Edit tools
Languages: 日本語
A Claude Code plugin that brings Verified Coherence Spec-Driven Development (VCSDD) methodology to any project. It enforces spec-first, test-first, adversarial review, and formal verification as sequential quality gates.
AI-assisted development has a structural problem: there are no quality gates. Language models produce code that passes surface-level review but routinely harbors spec mismatches, untested edge cases, and structural debt. This is "AI slop" -- code that looks correct but conceals hidden deficiencies.
VCSDD is a methodology that fuses four disciplines into a single workflow:
These are joined by an adversarial review gate: a fresh-context agent running on a more capable model that reviews all artifacts with zero tolerance and produces binary verdicts. The adversary is structurally isolated from the builder -- it reads only from disk and cannot be influenced by the builder's conversational context.
The result is a systematic process for eliminating the gap between "looks correct" and "is correct."
6-phase pipeline Spec Crystallization -> Test-First Implementation -> Adversarial Review -> Feedback Integration -> Formal Hardening -> Convergence. Each phase has explicit prerequisites and produces file artifacts that serve as the handoff to the next phase.
Two operating modes
strict -- full VCSDD ceremony for high-assurance work: sprint contracts, multiple adversary passes, proof obligations, all 6 phases enforcedlean -- all 6 phases remain in place, but approvals, sprint contracts, and proof obligations are lighter for product work and faster iterationThese modes are plugin-specific extensions, not a claim that canonical VCSDD defines two ceremonies. Canonical VCSDD assumes the human Architect signs off on the Phase 1c spec gate; this plugin keeps that as a hard requirement in strict mode and relaxes it in lean mode for faster product iteration.
Fresh-context adversary agent
The adversary (vcsdd-adversary) runs on the Opus model and is always spawned as a new agent instance with zero conversational history from the builder. It reads review artifacts from disk, produces findings, and terminates. It cannot say "overall looks good" -- it must cite concrete evidence for every verdict.
Binary PASS/FAIL verdicts across 5 operational dimensions
These are stable machine-readable buckets for the plugin, not a claim that the original article names these exact five dimensions. In methodology terms they compress the broader Phase 3 concerns: spec fidelity, test quality including edge cases, code quality, security surface, and spec gaps / verification readiness.
Chainlink bead traceability system
Every requirement, test, implementation block, adversary finding, and formal proof is assigned a bead identifier and linked in a directed graph. Any line of code can be traced back to its originating requirement. The full chain is preserved in an append-only history.jsonl audit log.
Completion is blocked if any persisted adversary finding lacks a matching adversary-finding bead.
Gate enforcement via Claude Code hooks
The vcsdd-gate-check.js hook runs on PreToolUse for Write/Edit/MultiEdit and for Bash when the command targets phase-restricted paths. It blocks direct writes, shell redirects, in-place edits, and common path-based mutation commands such as cp into restricted areas. Gate strictness is controlled by the VCSDD_HOOK_PROFILE environment variable.
npx claudepluginhub sc30gsw/vcsdd-claude-code --plugin vcsddIntelligently compose and deploy Claude Code Agent Teams. Auto-selects optimal team composition from 24+ agents and 83+ skills across 5 scopes, generates task dependency graphs, and orchestrates multi-agent workflows with a single command.
Verifiable AI-Augmented Engineering framework with traceable requirements, independent verification, and compliance-ready artifacts
Acceptance Test Driven Development for Claude Code. Enforces the ATDD methodology: write Given/When/Then specs first, generate a project-specific test pipeline, maintain two test streams (acceptance + unit). Inspired by Robert C. Martin's acceptance test approach from empire-2025.
Spec-Driven Development Workflow for Claude Code. A 6-step pipeline: Requirements → Code Analysis → Design → Implement → Verify → Self-Improve. Specifications are the source of truth, code is a verified artifact.
Specification-Driven Development with Process Discipline for Claude Code
Persona-driven AI development team: orchestrator, team agents, review agents, skills, slash commands, and advisory hooks for Claude Code
Code transformation: Dev SDLC orchestrator (code-shipping pipeline), plan, assert, audit, review, test, refactor, debug, for-sure. Hosts engineering agents.