Compound Engineering

Compounding Engineering Plugin

AI-powered development tools that get smarter with every use. Make each unit of engineering work easier than the last. Includes 36 specialized agents, 28 commands, and 27 skills.

This Claude plugin install surface contains only Claude-relevant files. Codex plugin metadata and Copilot assets are generated only on explicit target builds/installs from the canonical portable source in portable/compound-engineering/.

Support policy

• OpenCode first-class, GitHub Copilot and Codex second, Claude Code third.
• Keep: OpenCode-first portable source, this generated Claude Code plugin, and explicit Copilot/Codex exports.
• Codex: explicit full local export plus repo marketplace packaging through .codex-plugin/plugin.json, codex-skills/, and .agents/plugins/marketplace.json.
• De-emphasize: compatibility exporters for Droid, Pi, Gemini, and Kiro, plus legacy Claude-home sync mirrors.
• Removed legacy surfaces: .github_gpt/ and dormant Cursor-specific export/sync code that no longer fit the supported target matrix.

Workflow contract highlights

• /workflows:architecture is the architecture-improvement handoff between planning and /deepen-plan.
• /workflows:to-issues turns plans into local ticket artifacts with compact execution context, a dependency graph, and conservative execution batches before implementation, using focused-ticket-priming and ticket-flow-auditor.
• Full workflow track: constitution -> brainstorm -> grill-with-docs -> plan -> architecture -> deepen-plan -> to-issues -> work -> review -> triage -> compound.
• Lite workflow track: brainstorm/plan --lite -> work -> review -> triage if review creates todos -> compound if reusable knowledge exists.
• The lite mode is for small, low-risk changes and preserves TDD/evidence and scope contracts while reducing intake, research, and ticketization ceremony.
• Every core workflow ends with workflow-next-step, which binds the active artifact chain, audits completion evidence, summarizes completed stages, and prints the exact next-session command/input handoff.
• Finalized brainstorms, plans, architecture handoffs, and reviews can offer optional local-only visual artifacts as MDX sidecars under docs/visual-artifacts/ without hosted Plan MCP infrastructure. The renderer loads the BuilderIO Agent-Native plan style guidance, generates the pinned block catalog with @agent-native/core@0.67.0, and writes structured Plan primitives such as diagrams, file trees, tabs, checklists, annotated code, diffs, schema/API blocks, and wireframes when the source supports them. It writes preview.html by default, and /visual-artifact <artifact-path> wraps check and static preview from only the artifact path; --serve requires a reachable local Plan UI on 127.0.0.1:3001 by default.
• /lrj is a Ralph-style coordinator for existing plans: ticketize, audit/repair the ticket set, then work/review/triage/validate and commit two ticket batches at a time until the ticket index is complete.
• /workflows:plan, /deepen-plan, and /workflows:work now default to issue-shaped execution slices, with the first slice acting as the tracer bullet, while still allowing explicit infra-track and fix-batch modes when slices would be fake.
• /workflows:work can execute the next safe batch directly from a ticket index while preserving parent plan and architecture refs.
• Vertical slices now carry a feature-home module contract: feature business logic stays co-located, while truly shared utilities and adapters stay global.
• /brownfield-maintenance is the on-demand repair path for inherited repos whose AI-layer docs, prompts, and reviewer coverage have drifted.
• /workflows:work is the Ralph-first execution path; /ralph-loop and /cancel-ralph are helpers, not a detached workflow.
• Plans default to unit + e2e evidence unless an explicit exception documents replacement evidence.

Full workflow guide

Use this sequence when you want the full compound workflow instead of an ad hoc prompt chain:

constitution -> brainstorm -> grill-with-docs -> plan -> architecture -> deepen-plan -> to-issues -> work -> review -> triage -> compound

1. /workflows:brainstorm -- clarify the problem, user story, and constraints.
2. grill-with-docs -- stress-test domain language, update CONTEXT.md, and enrich the brainstorm inline.
3. /workflows:plan -- choose the execution shape and define packets.
4. /workflows:architecture -- lock feature homes, shared/global boundaries, and architecture handoff details.
5. /deepen-plan -- harden the plan with research and review.
6. /workflows:to-issues -- generate docs/tickets/... with focused-ticket-priming, then write the dependency graph and batch cursor into index.md before ticket-flow-auditor gates the set.
7. /workflows:work <ticket-index> -- execute the next safe batch from the ticket index while preserving parent plan and architecture refs.
8. /workflows:review -- review code, ticket drift, architecture fit, and TDD evidence together.
9. /workflows:triage -- research and resolve review-created todos before follow-up work or compounding.
10. /workflows:compound -- turn the result into reusable team knowledge.

Lite workflow path

Use the compact track when the request is already clear and low risk:

brainstorm/plan --lite -> work -> review -> triage if review creates todos -> compound if reusable knowledge exists

Lite mode keeps the problem narrative, user story, success criteria, TDD/evidence contract, execution shape, and scope fences, but usually emits one or a few execution packets instead of a full ticketized backlog.

Ticketized execution guidance

• Prefer /workflows:to-issues after /deepen-plan for the sharpest execution packets.
• Use the ticket index as the default /workflows:work input once it exists.
• Treat feature-home ownership and scope fences as hard boundaries, not suggestions.
• Use /workflows:triage after /workflows:review when review produces todos or follow-up decisions.
• Use /brownfield-maintenance separately when an inherited repo needs workflow repair before normal feature delivery.

Migration notes

• /technical_review is no longer part of the supported workflow. Use /workflows:architecture between planning and /deepen-plan, then continue through /workflows:work and /workflows:review.
• OpenCode remains the canonical first-class surface for the source repo. This generated Claude output is the third-class compatibility surface, while Copilot and Codex are supported second-class generated outputs.
• .github_gpt/ and dormant Cursor-specific export/sync code have been removed from the supported workflow. Droid, Pi, Gemini, and Kiro remain de-emphasized compatibility exporters.
• Ralph evidence is now part of the normal work/review contract: red, green, and post-refactor green proof are expected unless a plan records an explicit exception.

Verification guidance

bun run build:platforms
bun run verify:generated
bun test

Components

Component	Count
Agents	36
Commands	28
Skills	27
Hooks	0
MCP Servers	1

Agents

Agents are organized into categories for easier discovery.

Review (21)

Agent	Description
agent-native-reviewer	Verify features are agent-native (action + context parity)
architecture-strategist	Analyze architectural decisions and compliance
code-simplicity-reviewer	Eliminate complexity, wrong abstractions, and readability regressions
constitution-guardian	Enforce repo rules derived from constitutions, architecture docs, and governing markdown
data-integrity-guardian	Database migrations and data integrity
data-migration-expert	Validate data migrations and ID mappings
deployment-verification-agent	Create Go/No-Go deployment checklists for risky data changes
rabak-frontend-races-reviewer	Review React, Vue, and async UI code for race conditions and stale state bugs
rabak-java-reviewer	Java/JVM code review with a high bar for deep modules, contracts, and maintainable performance
rabak-laravel-reviewer	Modern Laravel 11+ code review (architecture, Eloquent, testing)
rabak-nest-reviewer	NestJS code review enforcing simplicity, performance, and security
rabak-python-reviewer	Python code review with strict conventions
rabak-rust-reviewer	Rust code review for ownership, safety, idiomatic patterns, and zero-cost abstractions
rabak-typescript-reviewer	TypeScript code review with strict conventions
rabak-vue-reviewer	Vue 3 / Nuxt 3 code review (Composition API, TypeScript, Pinia)
pattern-recognition-specialist	Analyze code for patterns and anti-patterns
performance-oracle	Performance analysis and optimization
schema-drift-detector	Detect unrelated schema dump and schema artifact drift in database PRs
security-sentinel	Security audits and vulnerability assessments
ticket-flow-auditor	Review plan-to-ticket and ticket-to-implementation alignment, dependency order, scope fences, and execution drift
uncle-bob	Clean-code reviewer focused on naming, cohesion, side effects, boundaries, and tests that keep code changeable

Research (6)

Agent	Description
best-practices-researcher	Gather external best practices and examples
framework-docs-researcher	Research framework documentation and best practices
git-history-analyzer	Analyze git history and code evolution
issue-intelligence-analyst	Analyze GitHub issues to surface recurring themes and pain patterns
learnings-researcher	Search institutional learnings for relevant past solutions
repo-research-analyst	Research repository structure and conventions

Design (3)

Agent	Description
design-implementation-reviewer	Verify UI implementations match Figma designs
design-iterator	Iteratively refine UI through systematic design iterations
figma-design-sync	Synchronize web implementations with Figma designs

Workflow (6)

Agent	Description
bug-reproduction-validator	Systematically reproduce and validate bug reports
e2e-test-strategist	Design, harden, advise on, and brutally audit real end-to-end test suites across plan, deepen, work, and review — real app, real infra, no fakes, no hardcoded passes
execution-agent	Execute scoped /workflows:work tickets and units with strict clean-code, DRY, SOLID, and Ralph-aware delivery discipline
local-visual-artifact-renderer	Convert finalized workflow Markdown artifacts into source-loyal local visual MDX sidecars
pr-comment-resolver	Address PR comments and implement fixes
spec-flow-analyzer	Analyze user flows and identify gaps in specifications

Commands

Workflow Commands

Core workflow commands use workflows: prefix to avoid collisions with built-in commands:

Command	Description
/workflows:constitution	Create or update the repo-level constitution that governs downstream workflows
/workflows:brainstorm	Explore requirements and approaches before planning
/workflows:plan	Create implementation plans with issue-shaped execution slices and structured project inputs
/workflows:architecture	Produce a dedicated architecture improvement artifact before deepening and execution
/workflows:to-issues	Convert plans into local vertical-slice ticket artifacts with scoped execution context
/workflows:work	Execute execution slices systematically
/workflows:review	Run comprehensive code reviews
/workflows:triage	Research todos, record chosen actions, then execute safe batches in swarm mode
/workflows:debug	Orchestrate reproduction, diagnosis, fix decisions, and design escalation for bugs and failures
/workflows:compound	Document solved problems to compound team knowledge
/workflows:compound-refresh	Refresh stale learnings and pattern docs in docs/solutions/

Utility Commands

Command	Description
/lrj	Ralph-style coordinator from an existing plan through ticket audit, two-batch work/review/triage/validation loops, and per-window commits
/deepen-plan	Enhance plans and harden execution slices with parallel research
/changelog	Create engaging changelogs for recent merges
/create-agent-skill	Create or edit Claude Code skills
/generate_command	Generate new slash commands
/heal-skill	Fix skill documentation issues
/brownfield-maintenance	Audit and fill brownfield AI-layer gaps outside the main feature workflow
/report-bug	Report a bug in the plugin
/reproduce-bug	Reproduce bugs and collect structured evidence for diagnosis or escalation
/resolve_parallel	Resolve TODO comments in parallel
/resolve_todo_parallel	Resolve todos in parallel
/test-browser	Run browser tests on PR-affected pages
/feature-video	Record video walkthroughs and add to PR description
/visual-artifact	Render or serve a local visual artifact from only its artifact path
/ralph-loop	Start a self-referential loop until completion promise is met
/cancel-ralph	Cancel an active ralph loop
/deploy-docs	Deploy documentation site

Skills

Architecture & Design

Skill	Description
agent-native-audit	Run a scored audit of agent-native architecture principles
agent-native-architecture	Build AI agents using prompt-native architecture
adversarial-plan-audit	Run a deep adversarial review of plans and architecture before implementation

Development Tools

Skill	Description
compound-refresh	Refresh stale learnings and pattern docs in docs/solutions/ against the current codebase
compound-docs	Capture solved problems as categorized documentation
create-agent-skills	Expert guidance for creating Claude Code skills
frontend-design	Create production-grade frontend interfaces
focused-ticket-priming	Turn one plan packet into one compact ticket-local execution packet
laravel-conventions	Modern Laravel coding standards reference
skill-creator	Guide for creating effective Claude Code skills

Content & Workflow

Skill	Description
brainstorming	Explore requirements and approaches through collaborative dialogue
caveman	Ultra-compressed communication mode for terse, technically exact responses
document-review	Review workflow artifacts with artifact-aware lenses and headless refinement
file-todos	File-based todo tracking system
finishing-branch	Structured branch completion with verification, merge/PR options, and cleanup
git-worktree	Manage Git worktrees for parallel development
grill-me	Stress-test plans and designs by interviewing one decision at a time
resolve-pr-parallel	Resolve PR review comments in parallel
session-history	Recover prior attempts from repo artifacts first, then bounded recent session history
setup	Configure which review agents run for your project
systematic-debugging	Structured debugging with explicit causal chains, prediction tests, and design escalation gates
ubiquitous-language	Build a DDD-style glossary and canonical domain vocabulary from the conversation
workflow-next-step	Inspect workflow artifacts, summarize completed stages, apply routing gates, and hand off the exact next-session command

Multi-Agent Orchestration

Skill	Description
orchestrating-swarms	Comprehensive guide to multi-agent swarm orchestration

File Transfer

Skill	Description
rclone	Upload files to S3, Cloudflare R2, Backblaze B2, and cloud storage

Browser Automation

Skill	Description
agent-browser	CLI-based browser automation using Vercel's agent-browser

Image Generation

Skill	Description
gemini-imagegen	Generate and edit images using Google's Gemini API

MCP Servers

Server	Description
context7	Framework documentation lookup via Context7

Context7

Tools provided:

• resolve-library-id - Find library ID for a framework/package
• get-library-docs - Get documentation for a specific library

Supports 100+ frameworks including Laravel, Vue, Nuxt, Django, React, and more.

MCP servers start automatically when the plugin is enabled.

Browser Automation

This plugin uses agent-browser CLI for browser automation tasks. Install it globally:

npm install -g agent-browser
agent-browser install  # Downloads Chromium

The agent-browser skill provides comprehensive documentation on usage.

Installation

Claude Code:

claude /plugin install compound-engineering

Codex full local export:

bun run cli:install ./portable/compound-engineering --to codex

Codex repo marketplace output is generated explicitly with bun run build:codex at .agents/plugins/marketplace.json and points at this plugin folder's .codex-plugin/plugin.json.

Codex plugin package caveat: native Codex plugins package installable skills. The complete compound-engineering Codex environment also needs custom agents in .codex/agents, MCP config in .codex/config.toml, hooks in .codex/hooks.json, and local marketplace metadata in .agents/plugins/marketplace.json. Use the direct Codex installer command above for the complete setup; bun run build:codex generates the ignored repo-local marketplace and skill package explicitly.

Known Issues

MCP Servers Not Auto-Loading

Issue: The bundled Context7 MCP server may not load automatically when the plugin is installed.

Workaround: Manually add it to your project's .claude/settings.json:

{
  "mcpServers": {
    "context7": {
      "type": "http",
      "url": "https://mcp.context7.com/mcp"
    }
  }
}

Or add it globally in ~/.claude/settings.json for all projects.

Version History

See CHANGELOG.md for detailed version history.

License

MIT