Documentation

Review what Claude Code, Codex or Cursor did

Read the morning receipt, inspect the changes and checks, settle parked decisions, and find unfinished work before you accept an overnight coding-agent run.

Updated

SHORT ANSWER

How do I check what really happened overnight?

The morning receipt is a map for review, not a verdict. Start with its ending state and source locators, then inspect the actual diff or artifact, rerun the relevant checks, decide parked questions, and leave open items open.

01

Start with how the run ended and what was verified

Open .nightshift/receipts/morning-<date>-<shiftId>.md after clock-out, or render it on demand. The Shift section names the host, ending reason, open and ticked items, policy, and three verification lines: Verified, Disabled by owner, and Unavailable. A missing or disabled check is not a pass.

"$NIGHTSHIFT_PLUGIN_ROOT/runtime/ns" morning-receipt --view owner
02

Check the actual changes, not the summary

For each ticked item, open its item receipt, the named commit or output file, and the actual test result. Inspect the diff for scope, behavior, secrets, and unintended changes; rerun a focused check if the reported evidence is insufficient. A checkbox is self-reported and is not independent verification.

  • Repository mode: compare the shift branch with its base, review each commit, then inspect the current working tree.
  • Artifact mode: open the output named by the item receipt and verify it against the original request.
  • Treat a receipt's token or time figure as host-recorded usage, not a quality score.
03

Example: turn a report into a decision

Illustrative example, not a recorded Nightshift run: the report says 2 items ticked, 1 open; Verified lists npm test, Disabled says browser check by owner, and Parked records a temporary compatibility choice. Review the two diffs, run the browser path if it matters, accept or reverse the parked choice, and keep the third item on the next punch list. Do not call the shift complete because two commits exist.

  • Read any recovery or stall entry in shift-log.md and check for overlapping work.
  • Resolve every parked decision with its rollback note.
  • Use the open item and Next line to plan the next attended or scheduled shift.
04

Where the report comes from

The renderer uses the evidence ledger, resolved policy, punch list, parking lot, and shift log. Its owner, reviewer, release, and artifact views present the same records at different detail levels; presentation never changes which checks ran. The clock-out gate writes the configured view on a best-effort basis, and a failed render does not block clock-out.

Item receipts and usage ↗Public run evidence ↗
TRY IT

Start with a bounded prompt

This prompt names the outcome and preserves Nightshift’s review boundary. Paste it into the supported coding host from the project you want to change.

Render the owner morning receipt, then point me to each changed file, check, parked decision, recovery event, and open item for review.
BOUNDARIES

What this workflow does not claim

  • The renderer reports existing records and never upgrades a skipped check into a pass.
  • A tick is self-reported; review the underlying changes and evidence.
SOURCES

Evidence and sources

These links support the released behavior, public outcomes, or problem language described on this page.

Morning receipt referenceThe six sections, the four views, where the file lands, and the determinism rules.Open evidence ↗Evidence and receiptsThe ledger, comparison, renderer, artifact receipts, and their limits.Open evidence ↗Example receiptsAn index and one item receipt.Open evidence ↗
RELATED QUESTIONS

Continue from the question you have

inspect each item receiptRead the answer →see public run evidenceRead the answer →file reviewed shift historyRead the answer →