sessionlens

See what your AI coding agent actually did.

sessionlens turns Claude Code session logs into a readable timeline: every prompt, reply, tool call, command, file edit and error, in order. A scorecard then checks how the agent worked: did it test its changes, keep you informed, own its mistakes and wrap up clearly?

It runs entirely on your machine. Your logs are never uploaded.

npx @manalisatghar/sessionlens

Try the live demo (made-up sample sessions, nothing real).

Why

Coding agents now run for minutes at a time and make dozens of tool calls. When something goes wrong, or goes surprisingly right, the only record is a large log file that nobody reads. sessionlens makes that record easy to review, so you can:

• Find the exact command or edit where a session went off track
• See which files changed and which commands ran, in one place
• Spot sessions where the agent changed code without running tests
• Compare how an agent handles errors across sessions

The scorecard

Four checks, each answered from the log with evidence you can click through to:

| Check | What it asks | | --- | --- | | Checked its own work | After the last code change, did it run tests, a build or a type check? | | Kept you in the loop | Did it explain what it was doing, or make long silent runs of tool calls? | | Owned its errors | When a tool failed, did it say so instead of quietly moving on? | | Wrapped up clearly | Did it end with a summary, or stop mid task? |

Checks that don't apply (for example, no code was changed) are marked as such and left out of the grade. These are deliberately simple, transparent heuristics: a starting point for review, not a verdict on code quality.

Privacy

• The server listens on 127.0.0.1 only and rejects requests addressed to any other host.
• It makes no outbound network requests and has no analytics.
• File paths stay on the server; the browser only sees opaque ids.
• Logs are read from disk when you open them, and nothing is written anywhere.

Usage

npx @manalisatghar/sessionlens                 # reads ~/.claude/projects and opens your browser
npx @manalisatghar/sessionlens --dir ./logs    # use a different folder
npx @manalisatghar/sessionlens --port 5000 --no-open

Requires Node 18 or newer. To install it globally instead: npm i -g @manalisatghar/sessionlens, then run sessionlens.

Development

npm install
npm run dev        # UI with sample sessions at http://localhost:5173
npm test           # parser, scorecard and server tests
npm run build && npm start

The parser (src/lib/parse.ts) is tolerant by design: unknown entry types are ignored and half-written lines are skipped and counted, so it keeps working as the log format evolves.

Roadmap

• Compare two sessions side by side
• Export a session as a shareable, redacted report
• Cost estimates per session
• Custom scorecard rules

How it was built

Built with Claude Code, which felt fitting for a tool that reviews Claude Code sessions.

License

MIT