run-loop — Claude Code Skill

A Claude Code skill that closes the write→observe→fix loop. After writing or editing code, Claude automatically runs it, reads the full output, diagnoses failures, fixes them, and re-runs — without waiting for you to copy-paste errors back.

The Problem It Solves

Without this skill, the cycle looks like:

Claude writes code → you run it → it breaks → you paste the error back → Claude fixes it → repeat

That's slow, manual, and puts the burden of error relay on you. With run-loop:

Claude writes code → runs it → observes the error → diagnoses root cause → fixes it → re-runs → delivers working code

Claude becomes accountable to what the code actually does at runtime, not just what it looks like on paper.

What It Does

1. Auto-detects the runtime — Node, Python, Go, Rust, Java, TypeScript, Make, and more
2. Runs and captures everything — stdout, stderr, exit code, timing
3. Classifies the failure — build error, runtime crash, test failure, missing dependency, environment issue, timeout
4. Diagnoses root cause — traces errors back to their origin, reads relevant source files before touching anything
5. Applies a targeted fix — minimum change needed, no opportunistic refactoring
6. Re-runs and verifies — confirms the fix actually worked
7. Repeats up to 5 iterations — then surfaces whatever remains with a clear diagnosis

Installation

mkdir -p ~/.claude/skills/run-loop
cp SKILL.md ~/.claude/skills/run-loop/SKILL.md

Or clone directly:

git clone https://github.com/thisidnotfound/run-loop-skill ~/.claude/skills/run-loop

Usage

Automatic

Triggers automatically after Claude writes or modifies code.

Manual

/run-loop
/run-loop npm test
/run-loop pytest tests/
/run-loop go test ./...

Supported Runtimes

Runtime	Auto-detected via	Default command
Node.js	package.json	npm test
Python	requirements.txt, pyproject.toml	pytest
Go	go.mod	go test ./...
Rust	Cargo.toml	cargo test
Java	pom.xml, build.gradle	mvn test / ./gradlew test
TypeScript	*.test.ts, tsconfig.json	npx tsc --noEmit + test runner
Any	Makefile	make test

Error Classification

Class	Meaning	Example
BUILD_ERROR	Compile/type error before execution	TypeScript type mismatch
RUNTIME_ERROR	Crash during execution	Unhandled exception, bad import
TEST_FAILURE	Tests ran but assertions failed	Wrong return value
MISSING_DEPENDENCY	Module not found	npm install needed
ENVIRONMENT_ERROR	Missing env vars, config, permissions	.env not set
TIMEOUT	Did not complete in 30s	Infinite loop, deadlock

Output Example

[Run 1] EXIT:1 — BUILD_ERROR
  → Error: Cannot find module './utils' at src/index.ts:3
  → Fix: Created missing utils.ts with expected exports

[Run 2] EXIT:1 — TEST_FAILURE
  → 2 tests failed: expected 42, got undefined
  → Fix: calculateTotal() was not returning the value

[Run 3] EXIT:0 — PASS ✓
  → All 14 tests passed in 1.2s

Why This Matters

Most AI coding tools generate code and stop there. run-loop makes Claude responsible for the output — not just the syntax. Combined with the browser-api skill (which grounds API knowledge before writing), this covers both ends: accurate knowledge going in, verified behavior coming out.

Requirements

• Claude Code CLI

License

MIT