~/codigoagente
AI agents

ECC's 72 agents: which to use, when, and how to call them in Claude Code

At the end of the article you'll find the install command, a ready-to-paste prompt and a .md to download.

ECC's agents are 72 Claude Code subagents, each with a narrow job: planning (planner, architect), per-language code review (typescript-reviewer, python-reviewer…), fixing builds (go-build-resolver, rust-build-resolver…), hunting silent failures or auditing security. Five are enough to start: planner, code-reviewer, security-reviewer, the reviewer for your language and build-error-resolver.

This is part 4 of the complete ECC guide. Below you will see how the 72 are split, which model each family uses, how to call them (by command, by name or automatically) and a real test with silent-failure-hunter, which found a bug the general review missed.

What a subagent is and why ECC has so many

A subagent is a Markdown file in agents/ with frontmatter (name, description, allowed tools and model) and instructions. Claude Code runs it in a separate context: it works, returns a report, and its conversation does not clutter yours.

snippettext
---
name: code-reviewer
description: Expert code review specialist. Proactively reviews code for quality, security, and maintainability.
tools: Read, Grep, Glob, Bash
model: sonnet
---

ECC has 72 because it covers many languages and tasks, not because you should use them all. Each one only takes context when it runs; what is always loaded is its description, which Claude Code reads to decide when to delegate.

How the 72 agents are split

The 72 agents by family, counted at commit 4adeee6 · codigoagente
FamilyCountExamples
Plan and design8`planner`, `architect`, `code-explorer`, `spec-miner`
Reviewers by language and stack24`typescript-reviewer`, `python-reviewer`, `go-reviewer`, `django-reviewer`, `database-reviewer`
Quality review10`code-reviewer`, `security-reviewer`, `silent-failure-hunter`, `refactor-cleaner`
Fix builds12`build-error-resolver`, `rust-build-resolver`, `pytorch-build-resolver`
Tests and docs4`tdd-guide`, `e2e-runner`, `doc-updater`, `docs-lookup`
Operate agents5`loop-operator`, `harness-optimizer`, `chief-of-staff`
GAN generation3`gan-planner`, `gan-generator`, `gan-evaluator`
Open source, marketing and network6`opensource-sanitizer`, `seo-specialist`, `network-troubleshooter`

Models: 62 agents use sonnet, 6 use haiku (docs, comments and open source packaging) and only 4 use opus: planner, architect, spec-miner and healthcare-reviewer. On a Pro subscription, those four burn more quota.

Three ways to call an agent

1. Through a command. The easiest. Each main command has an agent behind it:

You want to…UseAgent
Plan a feature`/ecc:plan "…"``planner`
Review what you just wrote`/ecc:code-review``code-reviewer`
Fix a broken build`/build-fix``build-error-resolver`
Look for vulnerabilities`/security-scan``security-reviewer`
Remove dead code`/refactor-clean``refactor-cleaner`
Review Go or Python`/go-review`, `/python-review``go-reviewer`, `python-reviewer`
Update docs`/update-docs``doc-updater`

2. By name. Some have no command (for example typescript-reviewer, fsharp-reviewer or silent-failure-hunter). Ask Claude: "use the ecc:typescript-reviewer subagent to review src/". With the plugin they carry the ecc: prefix.

3. Automatically. Several descriptions say "Use PROACTIVELY": Claude Code may delegate to them unprompted. database-reviewer, for example, kicks in when you write SQL or migrations.

Real test: silent-failure-hunter

In part 3 we built a POST /orders endpoint with 19 green tests and ran it through /ecc:code-review, which reported 9 findings. Then we ran this agent alone on the same app.js:

snippettext
Use the ecc:silent-failure-hunter subagent to review app.js. Return its report as is, short.

It took 24 seconds and the CLI reported 0.16 USD. It found 8 issues, and the most serious one was not in the general review:

  • In the catch that returns 500, sendJson can throw again if headers were already sent. That leaves an unhandled promise rejection in an async handler, which can crash a Node process. It suggests checking res.headersSent and logging method and URL.
  • The catch {} around JSON.parse drops the original error: the client only sees invalid_json and debugging gets harder.
  • A client that aborts mid-upload ends up as a false 500; ECONNRESET and aborted should be handled separately.

It also agreed with the general review on the plain-text 404 and /products accepting any method. The lesson: specialized agents catch things the general reviewer does not prioritize. Do not run them all; run the one that matches the risk of your change.

Careful with consult: you pick, not the search

ECC ships a plain-language component finder. We asked it about code review and security for a Next.js app:

terminalsh
npx ecc-universal@2.2.3 consult "code review and security for a Next.js app" --target claude
consult recommended fsharp-reviewer for a Next.js app · codigoagente

It got capability:security, code-reviewer and security-review right, but its third pick was fsharp-reviewer for a Next.js app, only because the words "code" and "review" matched. The finder matches keywords; it does not understand your stack. Use it as a starting point and add typescript-reviewer and react-reviewer yourself.

Which agents to use for your stack

Your projectRecommended agents
Next.js / React + TypeScript`typescript-reviewer`, `react-reviewer`, `react-build-resolver`, `e2e-runner`
Python API (Django or FastAPI)`python-reviewer`, `django-reviewer` or `fastapi-reviewer`, `django-build-resolver`, `database-reviewer`
Go backend`go-reviewer`, `go-build-resolver`, `database-reviewer`
Rust`rust-reviewer`, `rust-build-resolver`
Android / Kotlin`kotlin-reviewer`, `kotlin-build-resolver`
iOS / Swift`swift-reviewer`, `swift-build-resolver`
Laravel / PHP`php-reviewer` (plus the `laravel-*` skills)
RAG or ML in production`rag-pipeline-reviewer`, `mle-reviewer`, `pytorch-build-resolver`
Anything, before production`security-reviewer`, `silent-failure-hunter`, `performance-optimizer`

If you only want a few, copy them into ~/.claude/agents/; each agent is a standalone file:

terminalsh
git clone https://github.com/affaan-m/ECC.git
mkdir -p ~/.claude/agents
cp ECC/agents/{planner,code-reviewer,security-reviewer,typescript-reviewer,build-error-resolver,silent-failure-hunter}.md ~/.claude/agents/

Agents you will not need (at first)

  • gan-planner, gan-generator, gan-evaluator: a generator-evaluator loop to build products from one line. Interesting, but the macOS shell path has an open bug (#2674).
  • loop-operator: watches long autonomous loops. Useful once you automate a lot, not on day one.
  • chief-of-staff: triages email, Slack and messaging. Unrelated to coding.
  • opensource-forker, -packager, -sanitizer: for open-sourcing an internal project without leaking secrets.

FAQ

Can I change an agent's model? Yes: edit the model: line in its frontmatter. For example, move planner from opus to sonnet if quota is a concern.

Can agents change my code? It depends on their tools: line. planner only has Read, Grep, Glob (it cannot write); build-error-resolver and tdd-guide have Write and Edit.

What is the "Prompt Defense Baseline" block? Every agent starts with rules not to change roles, reveal secrets or emit executable code without need. It is a basic defense against prompt injection, not a guarantee.

Conclusion

Of ECC's 72 agents, five cover almost all daily work: planner, code-reviewer, security-reviewer, your language's reviewer and your stack's build resolver. Add silent-failure-hunter before each deploy: in our test it found the only bug able to crash the process. Part 5 covers choosing skills and rules so you do not load context you never use.

// resource · use it now

The ECC agents you will actually use

Copy only the agents that fit your stack and let the prompt help you pick the rest. Works with Claude Code 2.1 or newer.

Repository license: MIT

1. Install

The 6 core agents, by hand
git clone https://github.com/affaan-m/ECC.git
mkdir -p ~/.claude/agents
cp ECC/agents/{planner,code-reviewer,security-reviewer,build-error-resolver,silent-failure-hunter,tdd-guide}.md ~/.claude/agents/
Every agent through the plugin
npx ecc-universal@2.2.3 setup

2. Pick your case

Paste this prompt into your agent
I have ECC's agents available. This project is <stack: e.g. Next.js + TypeScript + Postgres>.

1. Read the names and descriptions in ~/.claude/agents/ (or the ecc plugin) and tell me which 5-6 agents fit this stack and what I will use each for.
2. Add an "Agents" section to CLAUDE.md: when to use planner, which reviewer to run after each change, and that silent-failure-hunter runs on touched files before every deploy.
3. Do not copy or install anything else; only propose and write CLAUDE.md.

3. How to use it

  1. Paste the prompt at the project root.
  2. Check the list: drop any reviewer for another language.
  3. Copy only the approved ones into ~/.claude/agents/.
Prompt and short guide in one fileRepositoryaffaan-m/ECC

Help us grow

codigoagente is free, with no paywalls. If this article helped you, any of these keeps the blog going.

// newsletter

One email a week with the repos worth your time

The week's new articles, tested against real code. No spam; unsubscribe in one click.

rea-video is an Apache-2.0 skill that breaks a video down shot by shot: filmed, AI or code, voice, music, captions and how to remake it. Real test and costs.
  • Skills for your agents
  • Claude Code
read article
AI agents

guide · Marketing psychology guide for AI agents · #3

Part 3 of the guide: anchoring, decoy, paradox of choice, Hick's Law and activation energy applied to the pricing page, with examples from the codigoagente kit.
  • Skills for your agents
  • Claude Code
read article
AI agents

guide · Marketing psychology guide for AI agents · #2

Part 2 of the guide: authority, commitment, real scarcity and framing with the marketing-psychology skill, with real before/after fragments from codigoagente.
  • Skills for your agents
  • Claude Code
read article