Find the open-source solution that fits.

Project overview

LionClaw wraps existing agent CLIs such as Codex and OpenCode in a project-local control plane. It keeps runtime profiles, durable sessions, skills, channels, scheduled jobs, private state, and audit records under a `.lionclaw/` directory while preserving the underlying agent’s own interface. Use it when you want swappable runtimes with explicit control over workspace access and credentials.

Project overview

unifusion runs an OpenCode orchestrator that sends one difficult brief to several panelists in parallel and keeps the synthesizer separate from the panel. It captures the final analysis, panel reports, session context, task events, and saved provenance, making the harness useful for comparing independent model views before a recommendation.

Project overview

claudebox launches a Claude Code project through bubblewrap on Linux or sandbox-exec on macOS. It shadows most of the home directory, mounts the project parent read-only, and blocks sensitive runtime sockets by default, with opt-in SSH, GPG, and XDG access. The project explicitly says this is not a complete security boundary; Linux is stable and macOS experimental.

Project overview

self-evolve provides a methodology and deterministic harness for improving a skill, repository, or project over repeated cycles. The loop reflects, proposes, evaluates, judges, and accepts changes, while code—not the model—decides acceptance, rejection, rollback, and which signal source to trust. It supports program checks, anchor verification, and generative evaluation for open-ended work where ground truth is limited.

Project overview

Multi-Agent Workflow Kit creates a separate Git worktree, branch, and directory for each coding agent, then places them in split panes within one tmux session for supervision and messaging. Agents are configured in .agents/agents.yaml and managed with maw commands for provisioning, syncing, navigation, and communication. The project is still an early proof of concept and requires Git 2.5+, tmux 3.2+, yq, and uvx.

Project overview

Memory Road records Claude Code events in SQLite, renders a continuity packet containing current state and open work, and injects it before each prompt through hooks. Hourly rollups, on-demand backfill, and a watchdog extend and protect the history. Installation requires Python 3.10 or later together with a background service, hooks, and scheduled jobs.

Project overview

The skill maps a repository’s technologies and attack surfaces, runs a bundled set of SAST, dependency, secret, container, and infrastructure scanners, then performs parallel deep dives into areas such as authorization, injection, supply chain, and agentic risks. It emits Markdown, SARIF, CycloneDX, and a reusable baseline, supports Claude Code only, and is best run inside an isolated container.

Project overview

oss-contribution-harness gives Claude Code a forward-only workflow for deduplicating candidate work, studying repository norms, reproducing the issue, routing security findings, drafting the change, and planning follow-up. Every network-visible action requires fresh human confirmation, and the workflow also throttles contribution volume; it expects authenticated gh plus git and jq.

Project overview

m_AI_l coordinates ongoing projects with a roster, per-thread mailboxes, and archives stored as ordinary Markdown files. Each Claude conversation can leave messages, act on its inbox, and hand work to a successor when its context fills, without a database or custom server. It is aimed mainly at non-developers seeking minimal setup and requires Claude Desktop with filesystem MCP access; the web version is not supported.

Project overview

Relay lets one orchestrator break cross-repository work into a dependency-aware ticket graph and route tasks to live Claude or Codex agents in the relevant repos. Channels retain the board, decisions, execution feed, and pull-request state, while the CLI, TUI, and desktop app share local data under `~/.relay/`. It is still pre-1.0 beta software, so interfaces and file layouts may change.

Project overview

Harness Plan keeps a long development campaign resumable across Claude Code sessions using compact state, one active feature contract, session summaries, deterministic transition scripts, and risk-gated verification. The same /harness-plan command starts or resumes work, while optional change units split larger features into reviewable stages and companion discipline skills add more structured testing and safeguards.

Project overview

harness-engineering-skill bootstraps a project harness for reliable LLM-agent work and maintains it across sessions. Its four modes cover project setup, handoff, promoting repeated mistakes into stronger guardrails, and independent evaluation. Deterministic shell scripts handle repeatable operations, while Markdown modes leave judgment to the agent; project artifacts record configuration, progress, features, lessons, and evaluation reports.

Project overview

CERA turns each conversation into a structured session map that records reasoning chains, dependencies, discoveries, and shifts in strategic posture. After those maps are added to a Claude Project knowledge base, a dedicated integrator builds a cross-session index so later conversations can resume from accumulated understanding rather than a collection of isolated facts.

Project overview

Claude North Star gives Claude Code or OpenCode a goal-oriented workflow instead of waiting for a new task after every step. It records the product vision, analyzes the repository, plans milestones, coordinates specialist sub-agents, and cycles through execution and evaluation. Progress, state, and architecture decisions persist in project files, with the user consulted mainly for strategic tradeoffs.