Rundown

Agent run reviewer. Point it at a session export + repo, get a review.

$ npx rundown session.json --repo ./my-app

Progress 100/100   11 repo-changing actions, 1/11 talk turns, tests PASS 14/14
Drift detected   goal shifted at turn 31 (bugfix → refactor)
Approval        git push flagged — no approval gate
Artifacts       6 files, secrets: none, destructive: git push
Cost            n/a (no token data)

That's the whole thing. No dashboard, no account, runs locally. Your code never leaves the machine.

What it checks

• Progress (0–100) — repo-changing actions vs. talk ratio; tests run + pass; verification commands.
• Drift — compares your first message to the final changes; flags the exact turn the goal shifted.
• Approval audit — lists mutating actions; flags any that bypassed a gate (git push, rm, destructive).
• Artifact trust — files touched, tests passing, secrets-leak scan, destructive-command flags, TODOs.
• Cost — token/$ summary when present. Says n/a when there's nothing, instead of guessing.

Install

$ npm i -g rundown
$ rundown ./session.json --repo ./my-app --format md

Lite (progress + file list) is free. Pro adds drift, approval, artifact trust, cost.

Real output

A 52-turn Codex session, reviewed by rundown. This is the actual report, not a mockup.

## Progress score (0–100)
**Score: 100/100**
11 repo-changing action(s); 1/11 pure-talk assistant turn(s); tests: PASS 14/14

## Drift detector
**Status: DRIFT** — 2 off-goal action(s)
- **DRIFT** (turn 23): rename BillingClient
  - Why off-goal: Class rename is cosmetic and outside the original auth-refresh/retry goal
- **DRIFT** (turn 31): refactored middleware
  - Why off-goal: Generic middleware refactor is unrelated to fixing auth token refresh

## Approval audit
git push to fix/auth-refresh with no prior user approval message — possible gate bypass

## Artifact trust
tests last: PASS 14/14; secrets: none found; destructive: git push

## Cost & eval (stub)
cost: n/a (no token data)

Free

• Full sample review (the one above, raw)
• Agent Run Review template — manual one-pager
• Drift Checklist — 10 questions after a long run
• Agent Continuation Brief — paste at a turn limit
• Lite CLI: github.com/clankamode/rundown

Pro — $29 one-time

Buy Rundown Pro $29

Full review engine (drift + approval + artifact trust + cost), Markdown/PDF/CI-JSON output, 1 year of updates, and the Continuation Brief prompt. No subscription.

Managed report ($9/report) — paste a session, get the reviewed PDF back. No CLI needed.

Local-first. Privacy is the architecture, not a feature.