Your coding agent stops repeating yesterday's mistakes.

failforward is a zero-dependency CLI plus Claude Code skill that records every build, test, and lint failure as plain Markdown — then recalls the relevant ones before similar work begins. Convention files extend it to Codex, Gemini, and Qwen.

GitHub ↗

Every failure recorded. Every lesson recalled.

Capturing a mistake is easy; remembering to check for it next time is the hard part. failforward automates the second half: a PostToolUse hook records failures as they happen, and a lightweight token-overlap search surfaces the relevant ones before you repeat the same class of error — scoped per-project or globally, with a built-in eval command to measure how well recall is actually working.

Capture

Automatic via PostToolUse hook

Recall

Searched before similar work

Footprint

One Python file, zero deps

GitHub ↗

Why it matters

One command away from not making the same mistake twice.

Explore other AI Tools →

← Snap Pair Next: cross-model-handoff →

Let's build a product or AI system together.プロダクトや AI システムを一緒に作りませんか。