wrap-stale

name	wrap-stale
description	Back-fill a handoff for a session that ended without /wrap — reads the prior transcript and produces the handoff + calibration entry. Designed to run unattended (cron).

Back-fill a handoff for a Claude Code session that ended without /wrap.

When to use

• The kickoff skill (or a cron scan) flags a stale unwrapped session — a transcript newer than its project's last handoff, with no handoff written.
• You want the next session in that project to start oriented instead of blind.

Inputs

• Transcript path (required): the prior session's .jsonl transcript. Passed explicitly — the runner can't rely on the model inferring it.

Procedure

1. Read the transcript. Reconstruct: what the session set out to do, what actually got done, what's left unfinished, any blockers or decisions made.
2. Write a handoff to the project's per-session memory dir, same format the /wrap skill produces (Review summary → what's done → what's next → key context).
3. Write a calibration entry — grade the session honestly (predicted vs actual, misses, lessons), same as a normal wrap. Stale back-fills must not skip this, or the calibration record silently drifts toward only-the-sessions-that-wrapped.
4. Note that this handoff was back-filled (not written live), so a reader knows it's reconstructed from the transcript rather than authored at session end.

Why it runs unattended

A human-triggered wrap only happens when you remember. The sessions that most need a handoff — the ones that crashed or blew out — are exactly the ones where nobody's around to run /wrap. Putting the back-fill on a timer makes the safety net automatic. Pair it with the kickoff skill's stale-session detection, which surfaces what still needs back-filling.