1
GitHub's toolkit for spec-driven development with AI coding agents
★ 140k · +898 this week · MIT · updated 2026-10-02
8.5/10
What it does Spec Kit is GitHub's open-source toolkit for giving AI coding agents a structured process instead of ad-hoc prompts. Its core process is spec-driven development: you describe what you want and why, the agent turns that into a specification, a technical plan and a task list, and only then implements against those documents. Two optional processes sit alongside… Read more →
Pros
- Official GitHub project
- Works with most coding agents
- Keeps big features on track
Cons
- Extra ceremony for small changes
Pricing: Free · spec-driven
2
AI task management that turns a PRD into tasks your coding agent works through
★ 28k · +39 this week · MIT + Commons Clause · updated 2026-04-28
8.0/10
What it does Task Master is a task-management layer for AI-driven development. You give it a product requirements document, it breaks the document into a list of tasks with dependencies, priorities and subtasks, and your coding agent then works through them one at a time, asking what comes next rather than trying to build everything in one pass. It runs… Read more →
Pros
- Keeps agents focused on one task
- Works as MCP server or CLI
Cons
- Needs a good PRD to shine
Pricing: Free · tasks planning mcp
3
Python framework for building, orchestrating and evaluating MCP-native AI agents
★ 3.9k · +9 this week · Apache-2.0 · updated 2026-10-01
8.0/10
What it does fast-agent is a Python framework — not a packaged end-user app — for building, running and evaluating AI agents. Its own README describes it as "a flexible way to interact with LLMs, excellent for use as a Coding Agent, Development Toolkit, Evaluation or Workflow platform." Agents are declared with simple Python decorators and can be chained, parallelized,… Read more →
Pros
- Deep, first-class MCP support (OAuth 2.1, sampling, elicitations), maintained by a well-known MCP tooling author (evalstate)
- Built-in workflow patterns — chaining, parallelization, routing, orchestration, and a k-voting 'MAKER' pattern — beyond a single-agent loop
- Actively maintained: Apache-2.0, ~3.9k stars, 446 forks, 2,000+ commits
Cons
- It's a code-first framework, not a turnkey app — requires Python and some setup to get real value
- As with any agent framework, safety depends entirely on how the developer scopes credentials and vets connected MCP servers
Pricing: Open source · mcp agent-framework python orchestration workflows
4
Lightweight spec-driven development: agree on changes before the agent codes
★ 71k · +564 this week · MIT · updated 2026-10-02
7.9/10
What it does OpenSpec adds a lightweight specification layer between you and your coding agent. Before any code is written, each change is written up as a small set of documents that you and the agent review and agree on; once the work is done, the change is archived and its requirements are merged into the project's living specs. It… Read more →
Pros
- Lightweight and brownfield-friendly
- Works across many agents
Pricing: Free · spec-driven
5
Pre-execution guard that blocks destructive commands and secret access for coding agents
★ 1.6k · +11 this week · MIT · updated 2026-10-02
7.9/10
What it does CC Safety Net is a pre-execution guard for AI coding agents. Before a tool call runs, it inspects the command and blocks destructive operations and access to secrets. It parses what a command actually does, so wrapping it in bash -c or python -c, reordering flags, or otherwise disguising it does not slip past the check. It… Read more →
Pros
- Parses command intent so wrapping or flag reordering does not evade it
- Works across many CLIs with presets, a GUI, and a local audit log
Cons
- Denies calls only; does not sandbox, set permissions, or watch network egress
- Path matching is mostly POSIX, with limited PowerShell coverage
Pricing: Open source · ai-safety guardrails hooks secrets security
6
Agile AI-driven development with analyst, PM, architect, developer and UX agents
★ 54k · +269 this week · MIT · updated 2026-10-02
7.8/10
What it does BMad Method (Breakthrough Method for Agile AI-Driven Development) is an open-source workflow for turning an idea or change request into working software with an AI coding agent while keeping the important decisions explicit. Its central idea is right-sized planning: a clear, small change goes straight to a build step, while a large or fuzzy initiative gets research,… Read more →
Pros
- Complete agile workflow with documents
- Works across many apps
Cons
- Heavyweight for small projects
Pricing: Free · methodology agile
7
Manage multiple terminal agents in parallel in separate workspaces
★ 8.6k · +28 this week · AGPL-3.0 · updated 2026-08-20
7.7/10
What it does Claude Squad is a terminal app for running several AI coding agents at the same time without them tripping over each other. Each agent gets its own session and its own copy of your repository, so you can hand one a bug fix, another a refactor and a third a new feature, then review and push each… Read more →
Pros
- Parallel agents without conflicts
- Works with several agents
Pricing: Free · parallel worktrees
8
Spec-driven development harness with Agent Skills and long-running autonomous implementation
★ 3.7k · +14 this week · MIT · updated 2026-09-23
7.7/10
What it does cc-sdd installs a spec-driven development workflow as Agent Skills, turning approved specifications into long-running autonomous implementation. One command sets up an agentic SDLC: discovery, requirements, design, tasks, and then autonomous implementation with per-task independent review. It treats the spec as a contract that makes boundaries between parts of the system explicit, so humans approve at phase gates… Read more →
Pros
- Full SDLC from discovery to autonomous per-task TDD with independent review
- Same 17-skill set installs across eight coding agents
Cons
- Only Claude Code and Codex are stable; other platforms are beta
- Autonomous implementation edits code and needs gate review
Pricing: Open source · spec-driven agent-skills tdd workflow kiro
9
Agent orchestration platform for running swarms of Claude Code agents
★ 74k · +449 this week · MIT · updated 2026-10-02
7.6/10
What it does Ruflo, the project formerly called Claude Flow, describes itself as an agent meta-harness for Claude Code and OpenAI Codex. It wraps the coding agent in an orchestration layer that spawns teams of specialist agents, coordinates them as swarms, stores what they learn in a vector memory and keeps that memory across sessions. The idea is that after… Read more →
Pros
- Advanced multi-agent orchestration
- Very active development
Cons
- Complex; steep learning curve
- Can consume a lot of tokens
Pricing: Free · orchestration swarm
10
Persistent cross-platform memory for coding agents, backed by Markdown and Milvus
★ 2.7k · +42 this week · MIT · updated 2026-09-24
7.6/10
What it does memsearch gives AI coding agents a persistent, cross-platform memory layer. Each supported agent installs a plugin that automatically captures conversation turns, stores them as Markdown files, and lets the agent recall relevant history later, either through a /memory-recall command or by asking naturally when a question needs past context. Because memories are plain .md files, they are… Read more →
Pros
- Automatic capture and hybrid semantic recall with local, free embeddings
- Markdown source of truth shared across agents; optional PROJECT.md/USER.md upkeep
Cons
- Downloads a ~558 MB model on first launch
- Cloud or self-hosted Milvus backends send memory data off-machine
Pricing: Open source · agent-memory semantic-search milvus markdown recall
11
Development recipes that keep Claude Code's exploration focused on the approved outcome
★ 687 · +3 this week · MIT · updated 2026-10-01
7.6/10
What it does Claude Code Workflows is a set of development recipes that keep Claude Code's deep exploration pointed at an agreed outcome. On non-trivial work Claude can wander into a real side finding and leave the requested change vague; these workflows fix the scope and exclusions before design, check designs against the repository, verify each task before commit, and… Read more →
Pros
- Fixes scope before design and verifies each task before commit
- Independent review and fresh-context handoffs on larger changes
Cons
- Claude Code only (Codex has a separate repo)
- Adds agent calls and artifacts, so overkill for small or throwaway work
Pricing: Open source · workflow spec code-review planning verification
12
Multi-model agent orchestration for OpenCode and Codex: type ultrawork, agents run to done
★ 70k · +344 this week · updated 2026-10-02
7.5/10
What it does Oh My OpenAgent (OmO), previously called oh-my-opencode, is a multi-agent harness plugin. It turns a single coding-agent session into an orchestrated team that works across several models. You type ultrawork (or ulw) with your prompt. A main orchestrator then plans the work, hands pieces to specialist subagents chosen by category, such as visual work, deep work, quick… Read more →
Pros
- Parallel specialist agents with category-based model routing
- LSP, AST-Grep, tmux and built-in search MCPs integrated
- Loads existing Claude Code hooks, skills and MCPs in OpenCode
Cons
- Source-available Sustainable Use License, not OSI open source; telemetry on by default
- Complex, fast-moving setup that can enable autonomous full-permission mode
Pricing: Free · orchestration multi-agent opencode codex multi-model
13
Google's official CLI and skills for building AI agents on Google Cloud with any assistant
★ 6.0k · +40 this week · Apache-2.0 · updated 2026-09-30
7.5/10
What it does Agents CLI is Google's official command-line tool and skill suite that turns a general-purpose coding assistant (such as Claude Code or Google's own Antigravity CLI) into a specialist for creating, evaluating and deploying AI agents on Google Cloud. Rather than being a coding agent itself, it's an add-on that teaches existing coding assistants a specific, well-scoped workflow:… Read more →
Pros
- Official Google project with Apache-2.0 license and a credible backer
- Adds a well-scoped, documented skillset (ADK scaffolding, eval, deploy) to any coding assistant
- Covers the full lifecycle: scaffold, evaluate, deploy, and observe
Cons
- Value is narrowly tied to the Google Cloud / ADK ecosystem
- Deploys to billable cloud resources, so misconfiguration has real cost implications
Pricing: Free · google-cloud adk agent-deployment cli official
14
Visual inspector for Claude Code sessions: tool calls, diffs, tokens, subagents and context
★ 4.0k · +11 this week · MIT · updated 2026-09-26
7.5/10
What it does claude-devtools is a desktop and web viewer for Claude Code sessions. It reads the transcripts and logs Claude Code already writes under ~/.claude/ and rebuilds them into a navigable interface, so you can see which files were read, what patterns were searched, the exact diffs applied, subagent activity and how the context window filled up. It does… Read more →
Pros
- Reads existing ~/.claude logs with no wrapper or configuration
- Per-turn token attribution and compaction view explain context loss
- Docker mode with no outbound network calls
Cons
- Claude Code only
- Unsigned desktop builds need a manual trust step on first launch
Pricing: Open source · observability debugging session-logs token-usage claude-code
15
Configuration framework that adds commands, personas and modes to Claude Code
★ 24k · +3 this week · MIT · updated 2026-09-27
7.4/10
What it does SuperClaude is a configuration framework that adds a structured development process to Claude Code. It installs a family of /sc: slash commands, specialist agents and behavioural modes that steer how Claude plans, researches, implements and tests, so a session follows a more repeatable workflow than free-form prompting. The project states that it is not affiliated with Anthropic.… Read more →
Pros
- Many ready-made commands and modes
- Popular and documented
Cons
- Adds a lot of instructions to context
- Overlaps with native features
Pricing: Free · framework commands
16
Run multiple AI models against one task and surface disagreements before you ship
★ 4.1k · +40 this week · MIT · updated 2026-10-02
7.3/10
What it does Claude Octopus runs a task through multiple AI models and surfaces where they disagree before you ship. Claude Code handles the ordinary path; Octopus stays dormant until you explicitly invoke an /octo: command, then escalates to multi-model workflows: adversarial review, structured provider debates, consensus gating, and a "Dark Factory" mode that takes a spec and runs research,… Read more →
Pros
- Adversarial multi-model review and consensus gating on demand
- Dormant by default; ordinary prompts are untouched
Cons
- Invoke-mode routing can start paid external-provider workflows and share prompt context
- Large surface area with many providers to configure
Pricing: Open source · multi-model orchestration code-review consensus personas
17
Multi-agent orchestration with 39 specialists, phased workflows and persistent sessions
★ 463 · +1 this week · Apache-2.0 · updated 2026-08-07
7.3/10
What it does Maestro is a multi-agent development orchestration platform. Given a task, it classifies the work, chooses between a lightweight Express path and a four-phase Standard workflow (design, plan, execute, complete) with explicit approval gates, asks the required design questions, produces an implementation plan when needed, delegates execution to specialists, runs a quality gate, and archives the session state.… Read more →
Pros
- Express and Standard workflows with approval gates and blocking reviews
- Same command surface across four runtimes with persistent session state
Cons
- Setup spans multiple runtimes and needs Node.js 20+ for the MCP server
- Large specialist roster can be heavy for small tasks
Pricing: Open source · orchestration multi-agent workflow quality-gate specialists
18
A meta-skill that designs domain-specific agent teams and generates the skills they use
★ 9.1k · +46 this week · Apache-2.0 · updated 2026-09-28
7.2/10
What it does Harness is a meta-skill for Claude Code that designs domain-specific agent teams. You describe a domain in plain language ("build a harness for this project") and it turns that description into a coordinated team of specialized agents plus the skills those agents use. Rather than being a fixed set of subagents, it is a factory that generates… Read more →
Pros
- Generates a tailored agent team plus skills from a plain-language domain
- Six architectural patterns with dry-run and comparison validation
Cons
- Requires Claude Code's experimental Agent Teams feature
- Author's effectiveness numbers are a small self-measured study
Pricing: Open source · meta-skill agent-teams orchestration skill-generation claude-code
19
ByteDance's self-hosted agent harness with subagents, sandbox, memory and skills
★ 83k · +360 this week · MIT · updated 2026-10-02
7.0/10
What it does DeerFlow is ByteDance's open-source super agent harness. Version 2 is a full rewrite built on LangGraph and LangChain: a lead agent plans work, spawns subagents, runs code in a sandbox, keeps long-term memory and loads skills on demand. It started as a deep-research tool and now also writes code, builds web pages and produces reports and slides.… Read more →
Pros
- Backed by a large team with very active development
- Standard SKILL.md skills, MCP servers and sandboxed execution
- Clear security guidance for deployment
Cons
- Heavy multi-service stack aimed at general research tasks, not an editor plugin
- Gateway admin equals code execution, so it must stay locked down
Pricing: Open source · agent-harness langgraph subagents sandbox self-hosted
20
Multi-AI orchestration plugin coordinating Claude, Gemini and Codex workers inside Claude Code
★ 40k · +177 this week · MIT · updated 2026-10-02
7.0/10
What it does Oh My Claude Code is a Claude Code plugin that turns Claude into a conductor for a small team of AI workers. It launches Claude, Gemini and Codex CLI processes in parallel tmux panes and coordinates them through 19 specialized agents, dozens of skills and MCP-backed tools, aiming for "no manual intervention required" runs from a stated… Read more →
Pros
- Coordinates multiple CLI agents (Claude, Gemini, Codex) in parallel rather than just one
- Ships a large set of ready-made specialist agents and MCP tools out of the box
- Adds checkpoints/rollback for staged changes needing approval
Cons
- Tightly coupled to Claude Code plus tmux, so it's not usable from other editors or agents
- Running several CLI agents concurrently increases the blast radius if a step goes wrong; worth testing on a disposable branch first
Pricing: Open source · claude-code-plugin multi-agent orchestration tmux gemini
21
Multi-agent terminal UI that coordinates Claude, Codex, Gemini and other CLI agents in one workspace
★ 3.5k · +18 this week · updated 2026-09-30
7.0/10
What it does Claude Codex Bridge (CCB) is a multi-agent terminal UI that runs several coding-agent CLIs — Claude, Codex, Gemini, Cursor, GitHub Copilot, Kimi, Qwen and others — side by side in one workspace, and lets them hand work to each other in defined graphs through an in-terminal /ask command. Each provider keeps its own real, native CLI pane… Read more →
Pros
- Unusually transparent safety model for a cross-CLI bridge: loopback-only network defaults, explicit private-interface opt-in for LAN, restrictive token-file per
- Very actively developed with detailed, versioned release notes and commits within days of review, orchestrating a genuinely large set of real CLI agents (Claude
Cons
- AGPL-3.0: free to use, but any modified/hosted derivative must also be open-sourced, and closed-source commercial use needs a separate paid license from the mai
- Large, fast-moving surface area (desktop app + Android mobile gateway + many provider integrations + Windows beta) that makes it harder to fully audit any singl
Pricing: Freemium · multi-agent cli-orchestration mobile-remote-control agpl cross-agent-handoff
22
Code-defined agent workflows with quality gates, approval breakpoints and resumable journals
★ 1.8k · +14 this week · MIT · updated 2026-09-16
7.0/10
What it does Babysitter is an orchestration framework that makes coding agents follow a workflow defined in code rather than improvising. You describe a process in JavaScript, or let the agent plan one, and Babysitter runs it step by step: it enforces quality gates before moving on, pauses at breakpoints for human approval, and records every decision in an event-sourced… Read more →
Pros
- Deterministic process runs with breakpoints and an event-sourced journal
- Plugins for many coding harnesses
- Internal harness for CI without an external agent
Cons
- Complex multi-package setup with a learning curve
- 'Hallucination-free' claim overstates what orchestration can guarantee
Pricing: Open source · orchestration workflows quality-gates human-in-the-loop ci
23
evo Open source
Autoresearch loop that has coding agents run gated optimisation experiments on your codebase
★ 1.5k · Apache-2.0 · updated 2026-07-17
7.0/10
What it does evo turns a codebase into an automated optimisation loop. You point it at a repository, it works out what can be measured, sets up a benchmark, and then runs rounds of experiments in which coding agents try changes, keep the ones that improve the score and discard the rest. It builds on the autoresearch idea of an… Read more →
Pros
- Gates stop the search from gaming the metric
- Tree search with several frontier strategies and parallel subagents
- Local, SSH and cloud sandbox backends
Cons
- Anonymous telemetry on by default
- Only useful when there is a clear measurable target; unattended runs can burn compute
Pricing: Open source · autoresearch optimization benchmarks parallel-agents worktrees
24
Model-agnostic terminal coding agent and the Python framework behind it
★ 1.1k · +8 this week · MIT · updated 2026-09-15
7.0/10
What it does Pydantic Deep Agents is two things in one repository: a terminal coding assistant in the style of Claude Code, and the Python framework that powers it. Both are built on Pydantic AI and can use Claude, GPT, Gemini or local models. You can use the CLI directly to plan, edit files, run commands and connect MCP servers,… Read more →
Pros
- Same harness usable as a CLI or from one Python call
- Live run forking with test-based or judge-based merging
- Any model provider, Docker sandbox and MCP client support
Cons
- Comparison claims against other tools are the maintainers' own
- Forking multiplies model spend
Pricing: Open source · agent-framework python pydantic-ai coding-agent cli
25
Self-hosted real-time dashboard for Claude Code and Codex sessions, agents, tools and cost
★ 1.0k · +23 this week · MIT · updated 2026-10-02
7.0/10
What it does Claude Code Agent Monitor is a self-hosted dashboard for watching Claude Code and Codex sessions as they happen. It installs hook handlers that forward session, tool-use, stop and subagent events to a local Express server, which stores them in SQLite and pushes updates to a React interface over WebSockets. It also reads existing transcripts from ~/.claude and… Read more →
Pros
- Live view of sessions, subagent trees, tool use and estimated cost
- Imports existing Claude Code and Codex transcripts
- Local-first with SQLite; MCP server read-only by default
Cons
- Large, fast-moving codebase that needs Node.js 22+
- Hook installer edits your global Claude and Codex settings
Pricing: Open source · monitoring dashboard hooks observability cost-tracking
26
Routing, repo-local memory, guarded workflows and handoffs around Claude Code and Codex
★ 922 · +-1 this week · MIT · updated 2026-10-01
7.0/10
What it does Citadel is an operating layer that sits around Claude Code or OpenAI Codex inside a single git repository. Instead of choosing between dozens of commands, you describe the outcome through one /do entry point, and Citadel routes the request to a matching workflow, keeps repository-local state between sessions, and records what happened so a fresh session knows… Read more →
Pros
- Single /do entry point with durable repo-local state and resume actions
- Pinned, checksum-verified releases and a reviewable removal path
- Unusually candid about what its benchmarks do and do not prove
Cons
- Heavy process for short one-off edits
- Not a sandbox; runs with the agent's full permissions
Pricing: Open source · orchestration hooks project-memory worktrees claude-code-plugin
27
Self-hosted operating system for delegating work across a lead agent and Docker worker agents
★ 850 · +18 this week · MIT · updated 2026-10-02
7.0/10
What it does Agent Swarm is an open-source "operating system" for running AI coding agents as a persistent, multi-agent workforce rather than one-off terminal sessions. A lead agent receives work from Slack, a repository, an issue tracker, email or an API, then delegates it to worker agents that each run in an isolated Docker container with its own development environment.… Read more →
Pros
- Persistent multi-agent orchestration with memory and identity that survives across sessions, not just one-shot runs
- Broad agent support: Claude Code, Codex, Cursor, Gemini CLI, Devin, OpenCode and more can act as workers
- Very active development (2,300+ commits) and MIT licensed
Cons
- Self-hosted Docker stack with its own encryption key and env management, more operational overhead than a plugin or skill
- No hosted/managed option yet, so you own the infrastructure and its security
Pricing: Open source · multi-agent orchestration docker self-hosted open-source
28
nWave Open source
Seven-wave, human-gated workflow from idea to TDD-delivered code inside Claude Code
★ 617 · +1 this week · MIT · updated 2026-09-16
7.0/10
What it does nWave is a structured delivery workflow that runs inside Claude Code. It splits feature work into seven waves: discover, diverge, discuss, design, devops, distill and deliver. Each wave has its own agents and produces documents you review and approve before the next wave begins, so the agent never runs unsupervised from idea to merged code. Delivery is… Read more →
Pros
- Human approval gate between every wave
- Hooks enforce TDD phases during delivery
- Adjustable rigor profiles to control model cost
Cons
- Heavy process for small changes
- Plugin-marketplace install lacks the enforcement hooks
Pricing: Open source · tdd workflow subagents atdd requirements
29
Shared, searchable session memory that turns agent traces into team skills
★ 1.6k · +-1 this week · Apache-2.0 · updated 2026-09-28
6.8/10
What it does Hivemind is a shared memory layer for coding agents, built by Activeloop on its Deeplake storage. It captures each session's prompts, tool calls and responses as structured traces, lets agents search that history later, and runs background workers that summarise sessions and turn repeated patterns into reusable SKILL.md files. The aim is that a solution one teammate's… Read more →
Pros
- Automatic capture and recall across several agents and machines
- Mines traces into reusable SKILL.md files
- Per-session and per-directory capture switches
Cons
- Full session traces go to a hosted service unless you set up your own cloud storage
- Benchmark claims come from the vendor's own evaluation
Pricing: Freemium · memory traces skills team hooks
30
Claude Code-style background subagents and scripted multi-agent workflows for pi
★ 1.2k · +27 this week · MIT · updated 2026-09-03
6.8/10
What it does pi-subagents is an extension for the pi coding agent that adds Claude Code-style subagents and scripted orchestration. The main agent gets an Agent tool that spawns specialised agents in isolated sessions, each with its own system prompt, model, thinking level and tool set. Agents run in the background by default and report back when they finish, and… Read more →
Pros
- Parallel background agents with steering, resume and FleetView
- Scripted workflows via SubagentWorkflow
- Worktree isolation and per-agent tool scoping
Cons
- Only works with pi, which is not an app in this directory
Pricing: Open source · pi-extension subagents orchestration worktrees parallel-agents
31
All-in-one Claude Code kit of agents, commands, skills, safety hooks and adversarial review
★ 845 · +2 this week · MIT · updated 2026-09-03
6.8/10
What it does Claude Forge is a large, opinionated setup pack for Claude Code that the author compares to oh-my-zsh. One install adds specialist subagents, slash commands, skills, hooks, rule files and a few MCP connections, all wired into a plan, test, review, verify and ship pipeline. Its main idea since version 4 is an adversarial verification loop: a second… Read more →
Pros
- Maker-checker loop with an independent adversarial reviewer
- Safety hooks block leaked secrets and destructive commands
- Frequent releases with changelogs
Cons
- Full install overwrites parts of ~/.claude and imposes many conventions
- Plugin install omits agents, hooks and rules
Pricing: Open source · claude-code subagents hooks code-review tdd
32
Scripted, resumable multi-agent workflows with model routing for the Pi agent
★ 551 · +11 this week · MIT · updated 2026-09-29
6.8/10
What it does pi-dynamic-workflows, by Quintin Shaw, brings script-driven multi-agent workflows to the Pi coding agent. When you ask for a workflow, Pi writes a short JavaScript orchestration script that fans work out to fresh subagent sessions, routes each one to a suitable model, cross-checks the results and returns one answer. Intermediate results live in script variables rather than your… Read more →
Pros
- Journaled resume avoids paying again for finished agent calls
- Per-agent model tiers and git worktree isolation
- Built-in verify and judge-panel helpers
Cons
- Only works with the Pi agent
- Large fan-outs can burn tokens quickly
Pricing: Open source · pi-extension workflows subagents git-worktrees orchestration
33
Local orchestration for Claude and Codex subagent teams with durable, recoverable state
★ 179 · +36 this week · MIT · updated 2026-09-14
6.8/10
What it does Oh My Subagents (oms) is a local orchestration layer for running Claude Code and Codex as coordinated teams of subagents. Instead of ad-hoc delegation where a lost connection or closed browser tab means starting over, it commits assignments and "waves" of work to durable state so a run can be resumed exactly where it left off, without… Read more →
Pros
- Durable, resumable state survives interruptions instead of losing progress
- Reusable, publishable workflow definitions plus a visual designer
- Ships with 8 ready-made starter workflows
Cons
- Console UI's Sustainable Use License is more restrictive than the MIT core
- Smaller community (140 stars) than established orchestration frameworks
Pricing: Free · subagents orchestration claude-code codex multi-agent
34
Open-source AlphaEvolve-style framework that evolves code using an LLM ensemble
★ 7.5k · +41 this week · Apache-2.0 · updated 2026-09-29
6.5/10
What it does OpenEvolve is an open-source, evolutionary coding framework inspired by DeepMind's AlphaEvolve. You give it starting code and an evaluation function, and it uses an ensemble of LLMs to iteratively mutate and improve that code across generations, aiming to autonomously discover better algorithms rather than just answer one-off coding prompts. What is inside - An island-based architecture running… Read more →
Pros
- Concrete, documented evolutionary mechanism (MAP-Elites, Pareto optimization, seeded reproducibility) rather than vague self-improvement claims
- Works with Claude Code CLI and several other LLM providers as pluggable backends
- Real per-iteration cost estimates given upfront rather than hidden
Cons
- Requires writing a custom evaluation function, so it's less plug-and-play than a normal coding assistant
- Ongoing LLM API costs scale with iterations, which needs active budget monitoring
Pricing: Open source · evolutionary-algorithms alphaevolve llm-ensemble python research
35
Spec-first workflow engine that runs an interview-evaluate-evolve loop over coding agents
★ 6.2k · +69 this week · MIT · updated 2026-10-02
6.5/10
What it does Ouroboros is a specification-first workflow engine that sits on top of your existing coding agent (Claude Code, Codex CLI, OpenCode, Gemini, Kiro, Copilot and others) rather than replacing it. Instead of letting an agent start coding from a vague prompt, it runs a structured loop: Interview (Socratic questioning to expose hidden assumptions), Seed (turning answers into an… Read more →
Pros
- Concrete, documented mechanism (ambiguity scoring, ontology convergence, drift measurement) behind its evolve loop rather than vague hype
- Works across 8+ existing coding-agent CLIs instead of locking you into one
- Active development with full architecture docs and event-sourced replay
Cons
- "Self-improving" framing overstates what actually happens (spec refinement, not model learning)
- Upfront interview/spec process adds overhead that's unnecessary for small, simple tasks
Pricing: Open source · workflow-engine spec-driven multi-agent orchestration python
36
Installs a plan, build and independent-review loop into your coding agent
★ 1.3k · +6 this week · MIT · updated 2026-09-28
6.5/10
What it does Autoprompt, by Spielewoy, installs a structured multi-agent workflow into an existing coding agent. You give it a goal with constraints and a way to check success, and it runs the loop for you: choose a route, plan, build, check with independent reviewers, repair, and finish. The idea is that the agent doing the work should not also… Read more →
Pros
- Separates planning, execution and verification across agents
- One installer covers many harnesses with doctor and uninstall commands
- Concurrency and model controls per run
Cons
- Headline failure-reduction figures are the author's own v1 benchmarks
- Needs Node, Python and Bash 4.3+, and multi-agent runs cost more tokens
Pricing: Open source · agent-orchestration subagents verification multi-agent coding-workflow
37
Self-hosted mission control to run and monitor terminal AI coding agents from any browser
★ 782 · +14 this week · MIT · updated 2026-10-01
6.5/10
What it does Codeman is a self-hosted dashboard for running terminal AI coding agents from any browser, including a phone. It spawns Claude Code, OpenCode, Codex, Antigravity, Gemini, Pi, Grok, DeepSeek Harness or OMP inside persistent tmux sessions and streams the real terminal to a web UI, so a session survives a dropped connection, a closed laptop or a reboot.… Read more →
Pros
- Runs nine coding CLIs in persistent tmux sessions streamed to any browser
- Touch-optimised mobile UI with QR login and push notifications
- Idle respawn and limit-reset resume for multi-hour unattended runs
Cons
- Installs via a curl-to-shell script; read it before running
- Network binding and tunnels need care, though a password is required
Pricing: Open source · orchestration self-hosted mobile tmux remote-access
38
Audio and voice feedback on every Claude Code hook, with a fully documented hook reference
★ 551 · +2 this week · MIT · updated 2026-06-04
6.5/10
What it does Claude Code Hooks is a reference project that wires a small script to every Claude Code hook event so the agent gives you audible and visual feedback as it works. In the author's default setup you hear a sound on session start, a mouse-click sound on PreToolUse, a keyboard sound on PostToolUse, and a spoken message on… Read more →
Pros
- Documents all hook events with their trigger and payload fields
- Ready-to-copy scripts and a settings fragment for macOS, Linux and Windows
- Useful template for writing your own hooks
Cons
- Hook list is pinned to a Claude Code version and can lag newer releases
- Only adds sound and voice feedback, not deeper automation
Pricing: Open source · hooks claude-code notifications reference python
39
Index past coding-agent sessions into local SQLite, queryable by your agent and browsable by you
★ 551 · +11 this week · AGPL-3.0 · updated 2026-10-02
6.5/10
What it does Obelisk indexes your past coding-agent sessions into one local SQLite database and makes them queryable by your agent and browsable by you. It reads transcripts from Claude Code, Codex, DeepSeek Harness, Kimi Code, OMP and Pi, and folds them into a single schema with a source field on every row. It has two sides that share the… Read more →
Pros
- Indexes multiple agents' transcripts into one searchable schema
- Agent skill answers history questions in natural language via JS queries
- Companion desktop app for browsing sessions and usage
Cons
- Reads and writes a real ~/.obelisk index; back it up before destructive rebuilds
- Prebuilt desktop app is macOS-only for now
Pricing: Open source · session-history memory sqlite search multi-provider
40
A senior-engineer layer for Claude Code: explore and clarify before the native build loop
★ 397 · MIT · updated 2026-06-25
6.5/10
What it does SWE-ATLAS (Software Engineer AI Agent Atlas) is a template that scaffolds a curated set of skills, subagents, slash commands and engineering conventions into a Claude Code project. Its bet is that Claude Code already handles the build loop (plan mode, goal, auto mode, dynamic workflows), so ATLAS focuses on the front of the loop: deciding what to… Read more →
Pros
- Wireframe and prototype commands validate the shape before generating code
- Three CLAUDE.md modes from collaborative review to autonomous runs
- free-will skill logs high-stakes decisions with rejected alternatives
Cons
- Autonomous mode removes the approval gate and is riskier
- Adds ceremony some may prefer to skip
Pricing: Open source · claude-code template wireframes decision-logs engineering-conventions
41
Claude Code skill and CLI to run tasks as subagents in parallel Git worktrees
★ 339 · +-1 this week · MIT · updated 2026-04-27
6.5/10
What it does Subtask, by zippoxer, gives Claude Code a skill and a command-line tool for running work in parallel. Claude drafts tasks, spawns subagents to do them, tracks their status, reviews the code and asks whether to merge or request changes. Each task gets its own Git worktree, so several can run at once without stepping on each other,… Read more →
Pros
- Each task isolated in its own worktree
- TUI shows progress, diffs and conversations
- Subagents can be Claude, Codex or OpenCode
Cons
- Self-described early development with known bugs
- Default installer is a curl-to-shell script
Pricing: Open source · subagents git-worktrees parallel-agents tui claude-code
42
Open Python agent harness with parallel tools, MCP, memory, subagents, background tasks and REST/SSE serving
★ 250 · +3 this week · MIT · updated 2026-10-01
6.5/10
What it does OmniCoreAgent is an open-source Python framework for building AI agents that hold together in production. It wraps a language model with the parts an application needs around the model loop: a prompt contract, local Python tools, MCP server tools, memory, context management, a workspace for files, guardrails, and events. It can run several independent tool calls in… Read more →
Pros
- Parallel tool batches and structured observations
- Signature-based loop detection beyond step counts
- Optional serving and durable background tasks
Cons
- A framework to build on, not a drop-in coding-agent extension
- Large API with a learning curve
Pricing: Open source · python agent-harness mcp background-tasks fastapi
43
Claude Code subagents that route work to Gemini or GPT by scope, then review before writing
★ 152 · MIT · updated 2026-07-26
6.5/10
What it does Gemini-GPT Hybrid is a pair of Claude Code subagents that route a request to a second model based on its scope, then bring the result back through Claude. The idea is that no single model fits every task: a whole-codebase audit wants a very large context window, while a tight bug fix wants fast, focused iteration. The… Read more →
Pros
- Routes big-context work to Gemini and focused work to GPT automatically
- Soft mode keeps Claude as the only writer behind a review pass
- Explicit, auditable routing logic
Cons
- Depends on external Gemini and Cursor/Codex CLIs you must install and authenticate
- Hard mode lets external models overwrite files, so commit first
Pricing: Open source · multi-model subagents orchestration gemini code-review
44
Cross-agent skills for disciplined AI development, plus a Bash safety hook for Claude Code
★ 151 · +2 this week · MIT · updated 2026-10-02
6.5/10
What it does AI-Driven Development is CodeAlive-AI's umbrella collection of Agent Skills plus one Bash safety hook. The skills aim to make coding agents follow repeatable engineering habits and to let an agent manage its own setup, such as MCP servers, hooks, settings and plugins, so you do not have to hand-edit JSON, YAML or TOML config files. What is… Read more →
Pros
- Skills follow the open Agent Skills standard, so they work across many coding agents
- Covers practical workflows: bug-fix protocol, plan gate, agent self-configuration
- bash-guard hook asks before destructive shell, DB and infra commands
Cons
- Broad grab-bag; some skills (OS health, C# rename) are niche
- The hook installs via a curl-to-shell script you should read first
Pricing: Open source · agent-skills engineering-practices safety-hook cross-agent prompt-engineering
45
Yume Freemium
Native desktop UI for the official Claude Code CLI with orchestration and background agents
★ 150 · +1 this week · updated 2026-09-26
6.5/10
What it does Yume is a native desktop application that puts a graphical front end on the official Claude Code CLI. It spawns the real claude binary as a subprocess and uses your existing Claude subscription, so subagents, MCP, hooks, skills and CLAUDE.md all keep working. It aims to fix the rough edges of a terminal, flickering, input lag in… Read more →
Pros
- Spawns the real claude binary, so subagents, MCP, hooks and skills all work
- Tabs, panes, visual diffs and always-visible 5h/7d usage meters
- Background agents with git branch and worktree isolation
Cons
- Proprietary and closed source with paid tiers; free demo is capped
- Licence re-verifies weekly over the network
Pricing: Freemium · desktop-app claude-code orchestration background-agents gui
46
Port of Cursor's pstack playbooks for verified bug fixes and features
★ 750 · +159 this week · MIT · updated 2026-10-02
6.3/10
What it does pstack-claude, by Michael Denyer, ports Lauren Tan's pstack to agents other than Cursor. pstack is an opinionated stack of skills for getting concise, verified work out of a coding agent. You tell its poteto-mode entry point your goal, and it picks a playbook: for a bug it reproduces the failure, investigates, delegates the fix, and reruns the… Read more →
Pros
- Playbooks demand reproduction and passing evidence for fixes
- Plugin installs for Claude Code and Codex, skills-only for others
Cons
- A port of an existing Cursor plugin, so it depends on tracking upstream
- Brief README with most detail in a reference doc
Pricing: Open source · agent-skills workflows debugging claude-code-plugin codex
47
An engineering team in a box for Claude Code: role-based agents plus safety and quality hooks
★ 269 · MIT · updated 2026-04-18
6.3/10
What it does My Claude Devteam is a Claude Code plugin that sets up a team of role-based subagents plus a set of automation hooks. The idea is to split work the way an engineering team does. A planner breaks larger tasks into written task prompts. Engineers implement the changes, a critic reviews them, and specialists handle debugging, database work,… Read more →
Pros
- Clear role split with per-agent tool permissions
- Useful safety hooks like secret and force-push blocking
- Editable Markdown agents you can fork
Cons
- No commits since April 2026
- Optional curl overwrites your ~/.claude/CLAUDE.md, and prompts use pressure language
Pricing: Open source · claude-code subagents hooks code-review branch-protection
48
Codex config where a stronger model orchestrates and a cheaper one executes, with an independent reviewer
★ 1.7k · +53 this week · Apache-2.0 · updated 2026-10-01
6.2/10
What it does This project is a ready-made configuration for OpenAI Codex that splits work between a stronger model and a cheaper one. A root orchestrator plans the task and hands bounded pieces to named subagents. A separate reviewer role checks the result at the end. The installer asks which Codex plan you are on and copies the matching profile… Read more →
Pros
- Splits planning, execution and review across models to control cost
- Guided installer that never overwrites blindly
- Includes a token-usage reporter
Cons
- Tied to specific OpenAI model names that will need updating
- Multi-agent runs consume much more quota
Pricing: Open source · codex orchestration subagents config token-usage
49
Non-blocking subagents for the Pi coding agent in tmux, zellij, cmux or WezTerm panes
★ 716 · +6 this week · MIT · updated 2026-05-12
6.2/10
What it does pi-interactive-subagents is an extension for the Pi coding agent that lets the main session spawn subagents in separate terminal multiplexer panes without blocking. A call to start a subagent returns immediately, the child runs in its own pane, and when it finishes its result is fed back into the main conversation as a notification that triggers a… Read more →
Pros
- Subagents run in visible panes and report back without blocking the main session
- Bundled planner, scout, worker, reviewer and visual tester agents
Cons
- Works only with Pi, not with any app in the directory
- Commit activity has slowed since May 2026
Pricing: Open source · pi subagents tmux orchestration async
50
Claude Code output style that makes the agent write clear, natural Korean
★ 1.4k · +26 this week · MIT · updated 2026-08-23
6.1/10
What it does Fluent Korean is a small Claude Code plugin that ships output styles for writing clear Korean. Coding agents are often tuned to be terse, and in Korean that tends to produce dropped particles and verb endings, strings of bare nouns and odd vocabulary that are hard to read and easy to misread. This plugin adds instructions at… Read more →
Pros
- Coding and non-coding variants of the output style
- Instruction text can be reused in other tools
- No scripts, just Markdown instructions
Cons
- Only useful for Korean-language work; README mostly in Korean
- Adds some tokens and adherence fades in long sessions
Pricing: Open source · korean output-style claude-code-plugin localization