The diff is only part of the story.

Give reviewers a redacted record of what your AI coding agent ran, what failed and what changed afterward. Published to the pull request automatically.

Read the setup guide View source on GitHub

Requires Node 22 or later and an authenticated GitHub CLI. Also available as a Claude Code plugin.

A published pull request log

## Decision log

**Checks (last run):** npm build ✓ · npm test ✓

### Decisions you confirmed
- Open to work: Open, remote or local
- Experience: Pull from my resume
- Blog post: Draft it, I review first

### Scope
+330 −79 lines · 16 files (1 added) · src · (root) · scripts  
Tests: 1 file changed · no matching test change: 3 source files

Checks and confirmed decisions, attached to the pull request.

See what actually ran

See test, build and lint results in order, alongside failures and the edits that followed.

Know where to look

A test edited after it failed, code changed after the final test run and unrecorded commits give reviewers specific places to inspect.

Keep raw sessions local

Raw command output, full local paths and the agent's own explanations stay out of the published log.

Install once.
Keep working as usual.

The setup guide covers the plugin route, configuration and troubleshooting.

Read the full guide

npm install -g pr-decision-log
pdl init --user
pdl doctor

Project details

607: tests passing in CI, on Node 22 and 24View test results
0.1.10: published on npmView on npm

As of 25 Oct 2026. View this version on GitHub.

Why I built it this way.

Before building anything, I measured my own sessions. The agent's hidden reasoning turned out to be stored empty, so the tool reports what actually ran and what happened instead of what the agent says it was thinking.

Read the engineering case study