ECC

Why pick it

You want the most complete agent operating system — not a plugin or a skill pack, but a full harness with memory, hooks, security scanning, continuous learning, and a dashboard. ECC is 261 skills, 66 agents, and 84 legacy commands evolved over 10+ months of daily production use.

What you get

• Memory persistence — hooks that save and load context across sessions automatically.
• Continuous learning — auto-extracts patterns from sessions into reusable skills.
• Token optimization — model routing, system-prompt slimming, background processes.
• Verification loops — checkpoint vs continuous evals, grader types, pass@k metrics.
• Parallelization — git worktrees, cascade method, when to scale instances.
• AgentShield — security scanning against prompt injection and agent attack vectors.
• Dashboard — Tkinter desktop GUI with dark/light theme, or ecc status --markdown for CLI.

Why it over the others

vs wshobson/agents — wshobson is a menu of specialists you compose yourself. ECC is a system: memory persists across sessions, skills self-improve, hooks run automatically. More setup, more power.

vs Compound Engineering — CE is a workflow loop (plan → work → review → compound). ECC is the substrate underneath any workflow: memory, hooks, model routing, security. You can run CE's loop on top of ECC's infrastructure.

vs Ponytail — Ponytail is one focused idea (write less code). ECC is everything: 261 skills across 12 language ecosystems, plus the runtime that manages them. Different scale entirely.

Works with

Claude Code, Codex, Cursor, OpenCode, Gemini, Zed, GitHub Copilot, and more. MIT-licensed; ECC Pro adds a hosted GitHub App for private repos ($19/seat/mo).

Quick start

In Claude Code: /plugin marketplace add affaan-m/ECC, then /plugin install ecc@ecc. Read the Shorthand Guide first — there's a lot here.

233k+ stars, 230+ contributors. Not just configs — a complete system built from real-world multi-harness engineering workflows.