freshcrate
Skin:/
Home > AI Agents > grace-marketplace

grace-marketplace

GRACE (Graph-RAG Anchored Code Engineering): open Agent Skills for contract-driven AI code generation with semantic markup, knowledge graphs, and support for Claude Code, Codex CLI, and Kilo Code.

Why this rank:Strong adoptionRelease freshnessHealthy release cadence

Description

GRACE (Graph-RAG Anchored Code Engineering): open Agent Skills for contract-driven AI code generation with semantic markup, knowledge graphs, and support for Claude Code, Codex CLI, and Kilo Code.

README

GRACE Marketplace and CLI

GRACE means Graph-RAG Anchored Code Engineering: a contract-first AI engineering methodology built around semantic markup, shared XML artifacts, verification planning, and knowledge-graph navigation.

This repository ships the GRACE skills plus the optional grace CLI. It is a packaging and distribution repository, not an end-user application.

Current packaged version: 3.10.0

What This Repository Ships

  • Canonical GRACE skills in skills/grace/*
  • Packaged Claude marketplace mirror in plugins/grace/skills/grace/*
  • Marketplace metadata in .claude-plugin/marketplace.json
  • Packaged plugin manifest in plugins/grace/.claude-plugin/plugin.json
  • OpenPackage metadata in openpackage.yml
  • Optional Bun-powered CLI package @osovv/grace-cli

The published CLI currently gives you:

  • grace lint for integrity checks
  • grace lint --profile autonomous for autonomy-readiness checks
  • grace status for project health, autonomy gate, and next-action guidance
  • grace module find for module resolution across shared docs and file-local markup
  • grace module show for shared/public module context
  • grace file show for file-local/private implementation context

Why GRACE

GRACE is designed for AI-assisted engineering where agents need stable navigation, explicit contracts, and reusable verification evidence.

Core ideas:

  • shared artifacts define the public module boundary
  • file-local markup defines private implementation detail
  • contracts describe expected behavior before code changes spread
  • verification is planned, named, and reused instead of improvised per task
  • semantic blocks give agents precise read and patch targets
  • execution packets, checkpoints, and failure handoffs make long autonomous runs recoverable
  • preferred stacks and named surfaces keep agents inside high-reliability project conventions

GRACE is process-first, not prompt-first:

  • do more work before launch so the agent has less ambiguity during execution
  • give the agent named contracts, flows, markers, and checkpoints instead of abstract exhortations
  • treat autonomy as a governed execution mode that must pass an explicit readiness gate

This makes it easier to:

  • plan modules and execution order
  • hand work across agents without losing context
  • review for drift between code, graph, and verification
  • debug failures from named blocks and planned evidence

GRACE was designed by Vladimir Ivanov (@turboplanner).

Install

Install skills first.

  • Skills are the core GRACE product surface.
  • The CLI is optional, but highly recommended once the skills are installed.
  • Installing only skills is a valid setup.
  • Installing only the CLI is usually not useful without the GRACE skills and workflow.

Install Skills

Skills and CLI are complementary, but they are distributed differently.

OpenPackage

opkg install gh@osovv/grace-marketplace
opkg install gh@osovv/grace-marketplace -g
opkg install gh@osovv/grace-marketplace --platforms claude-code

Claude Code Marketplace

/plugin marketplace add osovv/grace-marketplace
/plugin install grace@grace-marketplace

Agent Skills-Compatible Install

git clone https://github.com/osovv/grace-marketplace
cp -r grace-marketplace/skills/grace/grace-* /path/to/your/agent/skills/

Install CLI

The CLI is a companion to the GRACE skills, not a replacement for them.

Requires bun on PATH.

bun add -g @osovv/grace-cli
grace lint --path /path/to/grace-project

Quick Start

For a new GRACE project:

  1. Run $grace-init
  2. Design docs/requirements.xml and docs/technology.xml together with your agent
  3. Run $grace-plan
  4. Run $grace-verification
  5. Run grace lint --profile autonomous --path /path/to/project
  6. Run grace status --path /path/to/project
  7. Run $grace-execute or $grace-multiagent-execute

For an existing GRACE project, the CLI is often the fastest way to orient yourself:

# Integrity gate
grace lint --path /path/to/project
grace lint --profile autonomous --path /path/to/project

# Health + next action
grace status --path /path/to/project

# Resolve the relevant module
grace module find auth --path /path/to/project
grace module find src/provider/config-repo.ts --path /path/to/project --json

# Read shared/public context
grace module show M-AUTH --path /path/to/project
grace module show M-AUTH --path /path/to/project --with verification

# Read file-local/private context
grace file show src/auth/index.ts --path /path/to/project
grace file show src/auth/index.ts --path /path/to/project --contracts --blocks

Skills Overview

Skill Purpose
grace-init Bootstrap the GRACE docs, templates, and agent guidance
grace-plan Design modules, phases, flows, dependencies, and contracts
grace-verification Build and maintain verification-plan.xml, tests, traces, and log evidence
grace-execute Execute the plan sequentially with scoped review and commits
grace-multiagent-execute Execute parallel-safe waves with controller-managed synchronization
grace-refactor Rename, move, split, merge, and extract modules without shared-artifact drift
grace-fix Debug issues from graph, contracts, tests, traces, and semantic blocks
grace-refresh Refresh graph and verification artifacts against the real codebase
grace-reviewer Review semantic integrity, graph consistency, and verification quality
grace-status Report overall project health and suggest the next safe action
grace-ask Answer architecture and implementation questions from project artifacts
grace-cli Use the optional grace binary as a fast lint and artifact-query layer for GRACE projects
grace-explainer Explain the GRACE methodology itself
grace-setup-subagents Scaffold shell-specific GRACE worker and reviewer presets

CLI Overview

Command What It Does
grace lint --path <root> Validate current GRACE artifacts, semantic markup, unique XML tags, and export/map drift
grace lint --profile autonomous --path <root> Enforce autonomy readiness for execution packets, verification coverage, observable evidence, and operational-packet presence
grace status --path <root> Report artifact health, codebase metrics, integrity snapshot, autonomy gate, recent changes, and the next safe action
grace module find <query> --path <root> Search by module ID, name, path, purpose, annotations, dependency IDs, verification IDs, and LINKS
grace module show <id-or-path> --path <root> Show the shared/public module record from plan, graph, steps, and linked files
grace module show <id> --with verification --path <root> Include verification excerpt when a V-M-* entry exists
grace file show <path> --path <root> Show file-local MODULE_CONTRACT, MODULE_MAP, and CHANGE_SUMMARY
grace file show <path> --contracts --blocks --path <root> Include scoped contracts and semantic block navigation

Current output modes:

  • grace lint: text, json
  • grace status: text, json
  • grace module find: table, json
  • grace module show: text, json
  • grace file show: text, json

Agentic Reliability

GRACE 3.8 pushes more of the autonomous-execution discipline into the product surface:

  • grace lint --profile autonomous acts as a cheap readiness gate before long runs
  • grace status surfaces whether the project is healthy enough for execution or needs planning, verification, or refresh work first
  • technology.xml should name preferred stacks, test tools, and observability surfaces so workers stay on approved, high-reliability paths
  • operational-packets.xml should define assumptions, stop conditions, retry budgets, and checkpoint fields so workers can stop or replan without hidden reasoning
  • semantic anchoring matters: meaningful module names, block names, contracts, and examples are better agent guidance than abstract IDs or vague prompts

Public Shared Docs vs File-Local Markup

GRACE works best when shared docs stay public and stable, while private detail stays close to code.

Use shared XML artifacts for:

  • module IDs and module boundaries
  • public module contracts and public interfaces
  • dependencies and execution order
  • verification entries, commands, scenarios, and required markers
  • project-level flows and phases

Use file-local markup for:

  • MODULE_CONTRACT
  • MODULE_MAP
  • CHANGE_SUMMARY
  • function and type contracts
  • semantic block boundaries
  • implementation-only helpers and orchestration details

Rule of thumb:

  • grace module show is the shared/public truth
  • grace file show is the file-local/private truth

Core GRACE Artifacts

Artifact Role
docs/requirements.xml Product intent, scope, use cases, and requirements
docs/technology.xml Stack, tooling, constraints, runtime, and testing choices
docs/development-plan.xml Modules, contracts, implementation order, phases, and flows
docs/verification-plan.xml Verification entries, test commands, scenarios, and required markers
docs/knowledge-graph.xml Module map, dependencies, public annotations, and navigation graph
docs/operational-packets.xml Canonical execution packet, delta, and failure handoff templates
src/**/* and tests/**/* with GRACE markup File-local module context, contracts, and semantic block boundaries

Typical Workflows

Bootstrap a New Project

$grace-init
design requirements.xml and technology.xml together with your agent
$grace-plan
$grace-verification
$grace-execute or $grace-multiagent-execute

Inspect One Module Quickly

grace module find <name-or-path>
grace module show M-XXX --with verification
grace file show <governed-file> --contracts --blocks

Review or Refresh After Code Drift

grace lint --path <project-root>
grace status --path <project-root>
$grace-reviewer
$grace-refresh

Debug a Failing Flow

grace module find <error-or-path>
grace module show M-XXX --with verification
grace file show <governed-file> --contracts --blocks
$grace-fix

Repository Layout

Path Purpose
skills/grace/* Canonical skill sources
plugins/grace/skills/grace/* Packaged mirror used for marketplace distribution
.claude-plugin/marketplace.json Marketplace entry and published skill set
plugins/grace/.claude-plugin/plugin.json Packaged plugin manifest
src/grace.ts CLI entrypoint
src/grace-lint.ts grace lint command
src/grace-module.ts grace module find/show commands
src/grace-verification.ts grace verification find/show commands
src/grace-file.ts grace file show command
src/query/* Artifact loader, index, and render layer for CLI queries
scripts/validate-marketplace.ts Packaging and release validation
scripts/release-checklist.ts Release hygiene checklist for the current version and workflow coverage
.github/workflows/validate.yml CI workflow for tests, CLI validation, and marketplace checks
examples/cli/* Sample CLI flows and packet examples
RELEASING.md Manual release checklist and validation commands

For Maintainers

  • Treat skills/grace/* as the source of truth unless the task is explicitly about packaged output.
  • Keep plugins/grace/skills/grace/* synchronized with the canonical skill files.
  • Keep versions synchronized across README.md, package.json, openpackage.yml, .claude-plugin/marketplace.json, and plugins/grace/.claude-plugin/plugin.json.
  • Validate packaging changes with bun run ./scripts/validate-marketplace.ts.
  • Validate CLI changes with bun run validate:ci.
  • Do not assume every directory under skills/grace/ is published; the actual shipped set is declared in .claude-plugin/marketplace.json.

Development and Validation

Install dependencies:

bun install

Run the test suite:

bun test

Run the CLI against the repository itself:

bun run validate:cli

Run marketplace and packaging validation:

bun run validate:marketplace

Run the full CI-compatible validation stack:

bun run validate:ci
bun run release:checklist

Smoke test the query layer against a real GRACE project:

bun run ./src/grace.ts module show M-AUTH --path /path/to/grace-project --with verification
bun run ./src/grace.ts file show src/auth/index.ts --path /path/to/grace-project --contracts --blocks

License

MIT

Release History

VersionChangesUrgencyDate
v4.2.1## <small>4.2.1 (2026-09-24)</small> ### Summary Version 4.2.1 fixes a consistency bug where `grace lint` and navigation commands disagreed about how assertion diagnostics in archived plans are evaluated: lint accepted projects whose archived plans carried superseded `--assertions current` target commands, while module and verification navigation rejected the same projects as `invalid-project`. A single shared assertion lifecycle policy is now applied by both paths, treating archived plans as High9/24/2026
v4.2.0## <small>4.2.0 (2026-09-22)</small> ### Summary Version 4.2.0 improves command-gate reliability by adding per-command time budgets, warnings for redundant commands, commit and process metadata for run evidence, and retention controls that preserve the latest passing evidence for each change. Lint marker detection now recognizes Go standard logging and avoids false credit for error construction, while documentation tests verify that published CLI flags remain valid and .gitignore safely excludHigh9/22/2026
v4.1.0## <small>4.1.0 (2026-09-03)</small> ### Summary Users working with .NET, C/C++, or PowerShell codebases gain meaningful lint coverage in 4.1.0: C, C++, C#, and PowerShell source files can now opt into structural governance—markup validity, contract completeness, and MODULE_MAP shape are enforced the same way they already are for Go, Rust, Java, and shell—while the default ignored-directory set now prunes .NET's `obj` intermediate output, where generated sources no author edits would otherwiseHigh9/3/2026
v4.0.5## <small>4.0.5 (2026-08-23)</small> ### Summary Grace CLI 4.0.5 makes linting faster, more robust, and less noisy across ecosystems. The default ignored-directory list grows to 64 well-known tooling, build-output, and test-report names (covering Playwright, Allure, Newman, Cucumber, Stryker, Python, and JavaScript/TypeScript tooling), so grace lint automatically skips generated artifacts without extra configuration. Repeated lint runs and CI re-runs are now faster thanks to a content-addresseHigh8/23/2026
v4.0.4## <small>4.0.4 (2026-07-27)</small> ### Summary This release fixes a lint gap that could silently accept phase-incompatible command evidence, preventing `--assertions current` from being used inside target assertions or `MustPassCommand` entries, which are meant only for leaf project checks like tests and builds—this enforces the intended execution lifecycle where current mode is strictly a pre-write active-baseline check and keeps the plan phase rules unambiguous for all users. CI workflows Medium7/27/2026
main@2026-07-22Latest activity on main branchHigh7/22/2026
0.0.0No release found — using repo HEADHigh4/5/2026

Dependencies & License Audit

Loading dependencies...

Similar Packages

agent-orchestrator Agentic orchestrator for parallel coding agents — plans tasks, spawns agents, and autonomously handles CI fixes, merge conflicts, and code reviews.v0.14.6
skiller-desktop-skills-managerAI agent skills manager for Claude Code, Cursor, Codex and more — install, sync, and manage skills from one desktop app.v0.3.8
cc-sddTurn approved specs into long-running autonomous implementation. A minimal, adaptable SDD harness with Agent Skills for Claude Code, Codex, Cursor, Copilot, Windsurf, OpenCode, Gemini CLI, and Antigrav3.1.0
skillctrlCLI to manage and deploy AI agent skillsv1.6.1
ralphy-openspec🤖 Enhance AI coding with ralphy-openspec, a spec-driven tool that enables iterative execution for predictable, efficient development.main@2026-06-15

More in AI Agents

claude-memA Claude Code plugin that automatically captures everything Claude does during your coding sessions, compresses it with AI (using Claude's agent-sdk), and injects relevant context back into future ses
AIGC-Interview-Book【三年面试五年模拟】AIGC算法工程师面试秘籍。涵盖AIGC、LLM大模型、AI Agent、传统深度学习、自动驾驶、机器学习、计算机视觉、自然语言处理、强化学习、大数据挖掘、具身智能、元宇宙、AGI等AI行业面试笔试干货经验与核心知识。
letta-codeThe memory-first coding agent
nano-banana-pro-prompts-recommend-skillAI skill for OpenClaw & Claude Code — recommend from 10000+ Nano Banana Pro (Gemini) image prompts. Smart search by use case, content remix, sample images.