# Agent Run Review — one-page template

Fill this in after any long agent session (Codex / Claude Code / Cursor agent / Hermes / local).
Mirrors the paid Rundown report, manual version. Copy the block, answer each line.

```
# Agent Run Review — <session id / date>

**Goal (what I asked for):** __________________________________________________
**Agent:** __________  **Turns:** ____  **Branch:** ____________

## Progress (0–100)
- Repo-changing actions: ____   (edits / writes / file moves)
- Pure-talk turns: ____
- Tests run? ____  Result: PASS __/__  (or n/a)
- Verification commands run? (build/lint/test): ______________
Score gut-check: __/100  (lots of talk + no tests = low)

## Drift check
- Final changes vs. original goal — ON GOAL / DRIFTED
- Off-goal actions (list turn + what):
  1. turn __ : ________________________________
  2. turn __ : ________________________________
- Did the goal shift mid-run? From "____" to "____"?

## Approval audit
- Mutating actions (edit/write/terminal): ____
- Any that bypassed an approval gate? (git push, rm, destructive):
  __________________________________________________
- Did anything run that I didn't explicitly approve? ____

## Artifact trust
- Files touched: ________________________________
- Tests passing? ____   Secrets leak scan: none / <flag>
- Destructive commands: none / <list>
- Unresolved TODOs left in code: ____

## Cost (if known)
- Tokens: ____   $: ____   (or n/a)

## Verdict
Ship / Review again / Reject — because: ____________________________
```

Rule of thumb: if "Drift check" has more than one off-goal line, don't trust the branch blind.
Get the automated version: `npx rundown <session.json>`.