End session checkpoint
Skill mosfor/end-session-checkpoint-skill/end-session-checkpoint
Agent Skill for end-of-session coding checkpoints and handoff notes
npx -y skills add mosfor/end-session-checkpoint-skill --skill end-session-checkpointAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Create strong end-of-session checkpoints and handoff notes for coding work. Use when the user wants to end a session, wrap up work, create a restart note, hand off to the next agent, summarize repo state, or save current status.
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
3.2 KB, 740 tokens by cl100k_base, as published. Nobody here has run it
End Session Checkpoint
Workflow
Copy this checklist and check off items as you complete them:
End Session Checkpoint Progress:
- [ ] Step 1: Inspect repo state
- [ ] 1.1 Run git status and capture clean/dirty state
- [ ] 1.2 Capture diff shape and complexity signals
- [ ] 1.3 Check whether prior context file exists
- [ ] 1.4 Pull changes from session history or your memory
- [ ] Step 2: Run exit interview and collect information
- [ ] 2.1 Collect goal, decisions, gotchas, verification, next steps
- [ ] 2.2 Ask commit question only if repo is dirty
- [ ] Step 3: Execute actions
- [ ] 3.1 Commit or push only if user said yes
- [ ] 3.2 Archive old context file locally
- [ ] Step 4: Write checkpoint
- [ ] 4.1 Synthesize a strong restart note
- [ ] 4.2 Scale depth to repo complexity
- [ ] 4.3 Save to .agents/context/CURRENT_STATUS.md
- [ ] Step 5: Review output before delivery
Step 1: Inspect Repo State
Run these commands silently:
git statusgit diff --statgit status --porcelaingit log -5 --oneline --decorate(optional)
Step 2: Run Exit Interview
Always run the interview, even if the repo is clean.
Collect the minimum information needed for a real handoff:
- Next steps in priority order
- Any other questions you have about vague decisions made during the session
If the repo is dirty, also ask:
- Commit recent changes to git? Yes or no.
- Push to remote? Yes or no.
Step 3: Execute Actions
3.1 Commit and Push (conditional)
Only commit or push if the user explicitly said yes.
- Generate a concise conventional commit message from the diff.
- Run normal git hooks.
Prefer reviewed staging over blind staging when the worktree contains suspicious files.
3.2 Archive Prior Context
If .agents/context/CURRENT_STATUS.md exists, move it to:
.agents/context/archive/[TIMESTAMP]-session-summary.md
Do this after any commit so the archival step stays local.
Step 4: Write Checkpoint
Write a new .agents/context/CURRENT_STATUS.md.
Do not fill a rigid template.
Write one sharp checkpoint for the next agent. Use headings only when they help clarity. The exact structure may change with the session, but the checkpoint must cover these questions:
- What was the session trying to accomplish?
- What now works?
- What is still incomplete, broken, risky, or unverified?
- Where should the next agent start first?
- Which files, commands, or entrypoints matter most?
- Which decisions matter later, and why?
- What should not be repeated or accidentally undone?
Step 5: Review Output Before Delivery
Ask:
- If I were dropped into this repo cold, could I restart from this note without reading the full transcript?
- Does the note tell me what to do first, not just what happened?
- Are all important claims tied to user input, repo state, or actual verification?
- Did I mark unknowns honestly instead of bluffing?
If the answer to any of those is no, rewrite the checkpoint before delivering it.