Regressionledger

RegressionLedger v0.1.0 — first release 🎉

Stop your coding agent from re-applying fixes that already failed.

A zero-dependency Claude Code hook + CLI that fingerprints every edit, links it to the next test/build outcome, and hard-blocks the agent from re-applying a patch that previously failed — across sessions, surviving context compaction.

Highlights

• 🚫 Hard block (PreToolUse deny) when a previously-failed fix is re-applied, with the reason attached
• 🧬 Semantic fingerprint — normalized tokens, string/number literals abstracted, but true ≠ false
• 🔗 Outcome linkage from real test/build runs (npm, jest/vitest/mocha, pytest, go, cargo, tsc, eslint…)
• 💾 Durable JSON ledger that survives session restarts and context compaction
• 🔒 Concurrency-safe writes (file-locked); hooks fail open so they can never break the agent
• 📦 Zero runtime dependencies — offline, no API keys, no native build

Install

# inside your project
npx regressionledger init        # once published to npm
# for now, from source:
git clone https://github.com/anlor1002-alt/regressionledger
cd regressionledger
node bin/regressionledger.js init

Try it (no agent needed)

npm run demo     # watch the doom-loop get blocked
npm test         # 35 tests, zero dependencies

Quality

• 35/35 tests passing
• Independently code-reviewed; outcome-classification false positives, fail-open gaps, and a ledger lost-update race were all fixed
• Verified: concurrent writes lose nothing; hostile hook input always exits cleanly

See the README for full usage, configuration, and how it differs from existing tools.

Note: not yet on npm — npx regressionledger will work after the first npm publish.