AgentWorks
An open-source platform for goal-driven AI agents that plan, run and measure work on a schedule against a metric you set, running on coding-agent CLIs such as Claude Code, Codex, Cursor and Pi that a developer already uses.
AgentWorks lets a developer describe an outcome, choose a metric that proves it and set a target, then has agents plan the work, run it on a schedule, measure every run against the goal, and revise their own plan when the metric does not move. An autonomy setting per workflow - ask first, run steps, edit workflow, or full - controls how much the agent can do without approval, and anything outward-facing can always be gated behind manual sign-off.
Always-on 'crew' teammates each keep their own files, tools, memory, browser and schedule, and can be reached through the app, Slack or WhatsApp. Playbooks are ready-made Markdown skill packages - 23 shipped at the time captured, covering browser QA, reliability, security, performance, FinOps and growth analytics - that install into a workflow and can be authored by hand.
It runs on top of existing coding-agent CLIs (Claude Code, Codex, Cursor, Pi) rather than replacing them, installing as a macOS app with a bundled MCP bridge.