affaan-m/ECC
★ 296.3k
Python
MIT
How ECC optimizes AI coding agent performance with skills
Your AI coding agent is only using ten percent of its brain.
Unstructured LLM prompts fail when tasks span complex codebases.
First, raw coding assistants drift, repeat bad assumptions, and forget repository conventions across multi-turn sessions.
Prompt drift · Multi-turn context degradation
A structured engineering loop keeps autonomous agents grounded and verified.
Second, ECC enforces a disciplined engineering loop: plan, test, implement, review, and remember lessons.
Core loop · Plan, test, code, review, remember
Equip your coding assistant with 292 skills and AgentShield security.
Third, it bundles sixty-eight specialized agent personas, two hundred ninety-two skills, and AgentShield real-time security scanning.
ECC Substrate · 68+ Agents, 292+ Skills, AgentShield
Install the full harness into Claude Code with one command.
Fourth, you can install the plugin directly into Claude Code, Cursor, or Zed with session adapters that preserve memory across runs.
Plugin install · Works with Claude Code, Cursor, Codex, Zed
Free. MIT licensed. Over 228,000 stars on GitHub.
It's completely free and open source under the MIT license, with over two hundred twenty-eight thousand stars on GitHub.
affaan-m/ECC · MIT · 228k stars · Python & Shell