Works with the AI Agents you already use
Session Handoff: Stop Claude Code Reporting Finished Work as Undone
A new session starts from copies: a notes file, yesterday's summary, a follow-up list.
$5
Session Handoff: Stop Claude Code Reporting Finished Work as Undone
Example session with this skill installed
Start of session: what did the last sessions on this project actually do? Then, before I close: is it safe to /exit?
- Read your context and instructions
- Compiled the session handoff
SESSION HANDOFF: last 24h, newest session first. sessions examined 3 of 3; fully read 3, partial 0, could not read 0.
== session 2b1f9c0e 11-07 13:56Z -> 11-07 17:01Z [READ]
person's words (4): ... "no, do it now" ...
world-changing actions (3): [GIT] git push origin main ...
repos clean and pushed: 1 of 2
NOT READY: ./api: 2 commit(s) not pushed
verdict: NOT READY (exit 2)
Connects securely to your tools. The creator never sees your data.
About this skill
A new session starts from copies: a notes file, yesterday's summary, a follow-up list. Whatever the last session did after those were written is missing, so the new session reports the old state as current. The transcript on disk is the only complete record. Session Handoff reads it.
What it does:
- digest: for the last N hours, newest session first: every message you typed or queued (tool results and harness noise dropped, long pastes shortened), the world-changing actions the agent ran (git commit/push/merge/reset, GitHub merges, installs, deploys, container and service restarts, scheduling, sends, recursive deletes, file writes outside temp), and the agent's last message. Always "sessions examined N of N", three states per session: read, partial, could not read.
- hook: the same, compact (under 9,000 characters), for a Claude Code SessionStart hook. It never blocks a session.
- exit-check: refuses to say READY while a repo has uncommitted or unpushed work, a branch has no upstream, or your notes file has no fresh NEXT SESSION line. Each reason printed, exit 2.
It never says "nothing happened" from a partial or unreadable transcript, and a missing transcript folder is "could not read", not "no sessions".
One Python 3 script, standard library only, no network, writes nothing. Windows, macOS and Linux (the transcript-folder naming rule for Windows paths is in the selftest). Ships with a 40-case selftest you run first, including real git cases in a temporary repo.
Built from a real pattern: a fresh session reported six services as unpatched from an overnight report, hours after the previous session had patched them at the owner's own words, typed while the agent was busy. No note had those words. The transcript did.
It reads the Claude Code transcript format as it is today; if Anthropic changes it, sessions show as partial or empty and the script says it could not interpret them. The action list is keyword rules (editable with --keywords): it can miss an action it has no rule for and cannot see whether a command succeeded. exit-check covers git and one notes line only.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 9 days ago
Requires
Permissions
File Scopes
No network calls. Writes nothing. Runs read-only git commands (status, rev-parse, rev-list) for exit-check.