ecc-loopECC loop for Claude

What you get

• A plan you approve first: goal, files, tests, risks and what done means, saved to .ecc/plans/ before any code is written.

• Tests before code: Claude writes the test, shows it failing, then builds until it passes.

• A review from a fresh context: a separate Claude that never saw the build conversation checks the diff against the plan.

• A short report: plan, tests added, review verdict and build, lint and type results, instead of a transcript.

• Session memory: a summary in .ecc/sessions/ with rules learned, so "resume where we left off" works.

How it works

1. 1

   Ask for coding work as usual, or say "do this the ECC way".

2. 2

   Claude researches the repo and writes a short plan. You say yes or redirect.

3. 3

   It writes the failing tests, then the code, staying inside the plan.

4. 4

   A fresh reviewer checks the work. Blocking issues go back for fixes.

5. 5

   It runs the tests, lint, type check and build and records the results.

6. 6

   It saves a session summary and suggests a rule, hook or skill for anything done twice.

Works with or without ECC installed

• With the ECC plugin, it conducts ECC's own planner, TDD, review and security agents, and includes an index of all 71 agents, 314 skills and 98 commands.
• Without it, it runs the same roles with Claude's own subagents and built in prompts.
• In plain chat with no file tools, it gives you the plan and checklist as text.

Privacy and safety

• Instructions only: no scripts, nothing runs on install and nothing is sent anywhere.
• It writes plan and session files only inside your project's .ecc/ folder.
• Starter hooks and the full ECC install are optional and only added if you choose to.

Requirements

• Claude Code or Cowork for the full loop with files and subagents. In plain chat you get the plan and checklist as text.
• Optional: jq for the starter hooks, Node.js 18+ for the full ECC install.