skraft-plugin

agent
Guvenlik Denetimi
Uyari
Health Uyari
  • No license — Repository has no license file
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Pipeline SDLC agentique déterministe (DISCOVER→DELIVER) : agents spécialisés, reviewers adversariaux, skills Outside-In TDD/Clean Architecture et garde-fous hooks portés sur Claude Code, Copilot & Cursor.

README.md

skraft

Deterministic agentic SDLC pipeline — DISCOVER → DISCUSS → DESIGN → DISTILL → DELIVER

Specialized agents, adversarial reviewers, discipline skills (Outside-In TDD, Clean Architecture)
and mechanical guardrails (hooks) ported to Claude Code, GitHub Copilot and Cursor.

skraft-framework CI
Release
Latest release
Conventional Commits
semantic-release


Description

skraft turns a coding assistant into a disciplined software production line.
Each lifecycle phase is driven by a specialized agent, reviewed by a dedicated adversarial
reviewer, and locked down by deterministic guardrails (hooks) that block — before paying
for the sub-agent — any out-of-sequence dispatch, any mandatory skill left unloaded, or any
progression without a real artifact, an APPROVED verdict and a verified git commit.

The runtime core follows a hexagonal architecture (Clean Architecture) with zero external
dependency, tested boundary-to-boundary and hardened with mutation testing.

Key features

  • 🔁 5-phase SDLC pipeline orchestrated by skraft-orchestrator: DISCOVER → DISCUSS → DESIGN → DISTILL → DELIVER.
  • 🤖 Specialized phase agents: backlog-discoverer, backlog-planner, solution-architect, acceptance-designer, software-engineer — each with its dedicated adversarial reviewer.
  • 🔬 Independent reviewer lenses (quality-gates, architecture-boundaries, test-integrity, cold-reader) synthesized into a weighted verdict.
  • 📚 Discipline skills: Outside-In TDD, Clean Architecture testing, BDD/Gherkin, mutation testing, contract testing, ADR, issue refinement…
  • 🛡️ Mechanical guardrails G1–G8 (fail-closed hooks): dispatch ordering, forced skill loading + JSONL audit, artifact/verdict/commit verification, state protection.
  • 🎯 Multi-harness portability: the same guardrails on Claude Code, Copilot CLI and Cursor.
  • 💸 Token economy: state write-through model (rehydration once per session), model routing by cost class, repo-wide depthTier configurator.

Installation

skraft ships as a marketplace plugin. The plugin source lives in plugins/.

Claude Code

# Add the marketplace, then install the "skraft" plugin
/plugin marketplace add SebastienDegodez/skraft-plugin
/plugin install skraft

GitHub Copilot / Cursor

Equivalent manifests are provided (.cursor-plugin/marketplace.json, .github/hooks/).
See docs/architecture.md for the per-harness porting details.

Quick start

Once the plugin is installed, run the full pipeline through the orchestrator:

/skraft

The orchestrator automatically resumes from the last persisted state, manages phase transitions,
reviewer verdicts (with retry), and the engineer ↔ reviewer loop.

Documentation

All documentation lives in docs/.

Topic Link
📑 Documentation index docs/README.md
🏗️ Plugin architecture docs/architecture.md
🛠️ Guardrail framework (hexagonal, G1–G8, genesis anchoring) plugins/README.md
🛣️ Roadmap (13 US + status) docs/roadmap.md
🤝 Engineer/Reviewer cross-cutting view docs/agents/software-engineer-and-reviewer.md
🎨 Documentation conventions docs/conventions.md

Current status — summary

Component Status
SDLC pipeline orchestrated by skraft-orchestrator ✅ Implemented
Specialized phase agents (backlog-*, solution-architect*, acceptance-designer*, software-engineer*) ✅ Implemented
Reviewer lenses (quality-gates, architecture-boundaries, test-integrity, cold-reader) ✅ Implemented
Operational skills (plugins/skills/*) ✅ Implemented
Hook guardrails G1–G5 + G4/G5 (artifact/verdict/commit) ✅ Implemented
Guardrails G6–G8, observability, recovery 🚧 Roadmap

Development

# Tests (boundary-to-boundary, 0 runtime dependency)
node --test tests/skraft-framework/*.test.mjs

# Mutation testing (Stryker)
npm --prefix plugins/src ci && node plugins/src/node_modules/.bin/stryker run plugins/src/stryker.config.mjs

# Policy checks (data-driven config, models by cost class)
node plugins/src/cli/build-config-bin.mjs --check
node plugins/src/cli/resolve-model-bin.mjs --check

Test placement and Stryker configuration rules are described in AGENTS.md.

Versioning & releases

This project follows SemVer and publishes releases automatically via
semantic-release.

  • Commit messages must follow Conventional Commits:
    • feat:minor bump; fix: / perf: / refactor: / docs:patch bump;
    • feat!: or a BREAKING CHANGE: footer → major bump.
  • Automatic online-documentation commits (docs(sync): / docs(gaps):) never trigger a release.
  • On every push to main, the release.yml workflow:
    1. computes the next version from the commit history,
    2. updates CHANGELOG.md and the version in plugins/.claude-plugin/plugin.json + plugins/src/package.json,
    3. creates the vX.Y.Z tag and the GitHub Release with the release notes,
    4. commits everything with chore(release): X.Y.Z [skip ci].

See the Releases for the per-version
change history.

Contributing

  1. Branch off main.
  2. Use Conventional Commits (required for automatic versioning).
  3. node --test tests/skraft-framework/*.test.mjs must pass.
  4. Open a Pull Request — CI checks tests, config policy and models.

License

See the repository for licensing information.

Yorumlar (0)

Sonuc bulunamadi