fresh-start-handoff
Skill0 starsChanged 7 days agoMITView on GitHub
Writes a complete handoff so work can continue in a fresh chat or session. Use when a conversation has become long, the assistant keeps repeating the same mistake, requirements were added piecemeal over many turns, the user says the chat is "going in circles", the topic changes, or the user asks for a summary to continue elsewhere. Consolidates goals, decisions, constraints, progress and open questions into one paste-ready prompt.
In this repository11 other files it gives its agents.
- drop-in/AGENTS.mdAGENTS.md
- llms.txtllms.txt
- clarify-firstSkill
- coverage-checkSkill
- fresh-eyes-reviewSkill
- grounded-answerSkill
- honest-critiqueSkill
- learning-tutorSkill
- prompt-improverSkill
- task-specSkill
- verify-claimsSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 44
After two failed corrections, clear and start over with a better prompt: Claude Code best practices — …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Fresh-start handoff
Long, patched-up conversations degrade: models lock in early assumptions, errors compound over later turns, and they give in more to pushback as conversations grow. Restarting with one consolidated prompt is the evidence-backed fix.
When to suggest it
- The same correction has been made twice and the problem persists.
- Requirements arrived in pieces across many turns.
- The topic has changed, or the context is full of material that no longer matters.
- The conversation has turned into a back-and-forth argument about whether an answer is right.
Steps
- Collect everything that should survive, from the whole conversation (not just the last few turns).
Write the handoff as a single prompt in a code block:
## Goal ## Context (audience, background, source material to re-attach) ## Requirements (numbered, final versions only) ## Decisions made (and why) ## Constraints and preferences ## Progress so far (what is done, with file names/links) ## Known pitfalls (approaches that failed and why) ## Open questions ## Next step- Resolve, don't copy, history. Where a requirement changed over the conversation, keep only the final version, written out in full. Watch for reversals: when the user says "go back to the original", "undo that" or "the earlier one is better", find the value they mean and write it out; don't carry forward whatever was mentioned last. Treat drafts and outlines written before later changes as stale, and don't carry over the argument itself.
- Flag gaps. List anything that is genuinely undecided, so the user can confirm before pasting. If the conversation settled something, state the settled value instead of flagging it.
- Tell the user to open a new chat or run
/clear, paste the handoff, and re-attach the listed source files.
Evidence
- Requirements spread across turns cut performance by 39% on average; consolidating and restarting is the recommended fix: Laban et al., ICLR 2026 — https://arxiv.org/abs/2505.06120
- Hallucination errors compound in later turns (preprint): HalluHard, 2026 — https://arxiv.org/abs/2602.01031
- Sycophantic collapse rises from 0–57% within 5 turns to 20–97% within 25 turns (preprint): SPINE, 2026 — https://arxiv.org/abs/2609.09090
- After two failed corrections, clear and start over with a better prompt: Claude Code best practices — https://code.claude.com/docs/en/best-practices
- More in the guide: https://github.com/ji-hun-git/how-to-use-ai/blob/main/docs/06-workflows.md
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

