handoff
Skill0 starsChanged 5 months agoMITView on GitHub
Maintain a living HANDOFF.md that captures a project's full state — summary, current status, file map, decision log, change log, next steps, and blockers — so a fresh Claude instance with no prior memory (or a new human teammate) can resume the work by reading that one file. Use this skill whenever starting a new project or repository; resuming or picking up an existing one ("where did we leave off", "continue the X project", "pick up where we stopped"); setting up cross-session continuity or persistent context; or when the user asks to create, read, or update a handoff / state snapshot / session-handoff document. Also use it proactively to keep HANDOFF.md current whenever a significant change happens — a decision made or reversed, a file or resource added or removed, a milestone hit, a blocker raised or cleared, a scope or approach change — without waiting to be asked.
191 other projects use a version of this.
Adopted by
Most recently taken up by
- lhotwll217/owner-operatortheir own version, first seen Nov 16, 2026
- tanishq286/tq-forgetheir own version, first seen Nov 15, 2026
- niki914/zafirotheir own version, first seen Nov 15, 2026
- davidondrej/skillstheir own version, first seen Nov 15, 2026
- quickchatai/quickreviewtheir own version, first seen Nov 15, 2026
- HiteSit/real-scientist-skillstheir own version, first seen Nov 15, 2026
Used with
Projects with it also use decision-mapping, setup-ts-deep-modules and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
FabienDehopreSkilldecision-mapping
Turn a loose idea into a sequenced map of investigation tickets, then drive them to resolution one at a time.
Also in 8 projects that use this
liufeng1976Skillsetup-ts-deep-modules
Wire dependency-cruiser into a TypeScript repo so each package is a deep module — implementation hidden in subfolders, reachable only through its entry-point files. User-invoked.
Also in 42 projects that use this
AWSSkillto-questionnaire
Turn a decision you can't fully answer into a questionnaire for someone else to fill in.
Also in 93 projects that use this
GoogleSkillask-matt
Ask which skill or flow fits your situation. A router over the skills in this repo.
Also in 87 projects that use this
RoentekSkillloop-me
Grill me about specs for the workflows I want to build, within this workspace.
Also in 51 projects that use this
LesaniSkillwriting-beats
Shape an article as a journey of beats, choose-your-own-adventure style. The user picks a starting beat from the raw material, you write only that beat, then offer options for where to pivot next, beat by beat, until the article reaches a natural end. Use when the user has raw material and wants to assemble it as a narrative rather than an argument.
Also in 54 projects that use this
Other versions191 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 191
- mattpocock/skills★ 325ksame name
- QuivrHQ/quivr★ 40ksame name
- The-Vibe-Company/quivr★ 40ksame name
- GreyDGL/PentestGPT★ 16ksame name
- ykdojo/claude-code-tips★ 10ksame name
- ArcReel/ArcReel★ 6.6ksame name
- vinvcn/mattpocock-skills-zh-CN★ 5.8ksame name
- withcoral/coral★ 4.9ksame name
- epicenter-md/epicenter★ 4.8ksame name
- EpicenterHQ/epicenter★ 4.8ksame name
181 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.
0
Common
Most of the 172 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 60
Real facts, never placeholders: If a detail does not exist yet, write (TBC) — never invent a plausible-sounding value, and strip every …
- Distinctive: line 64
Current instruction wins: If what the user says now conflicts with the file, the new instruction takes precedence — follow it and update …
“Common” is measured: kept by at least 6 in 10 of the 172 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Handoff
What this is for
Long projects span many sessions, and context evaporates between them: auto-memory is lossy, and a fresh instance of Claude starts blind. This skill keeps a single living file, HANDOFF.md, that holds enough of the project's state for the work to resume cleanly from it alone.
The governing principle — the cold-start test. Write HANDOFF.md so that a fresh instance of Claude, with zero memory of the conversation (or a new human teammate), could pick the project back up by reading this one file. If something essential lives only in the conversation, it is not captured yet. Every decision about what to include comes back to this test.
Capture the delta — the detail the other layers drop. HANDOFF.md is not the only place a project's state is stored; it earns its keep by holding, in detail, what the others cannot. Know what each of the other layers loses, and deliberately cover the gap:
- A context-compaction summary is auto-generated and lossy — it keeps the gist and discards the specifics: exact values, commands, IDs, error messages, file/line references, and the granular path that led here.
- Auto-memory holds a handful of durable facts about the user and the project — not the project's full working state, and not reliably.
- CLAUDE.md holds standing rules and conventions — not state, not history, not the detail behind a choice.
So write down, in full, the things none of those will preserve: approaches already tried and why they failed (so they are not retried); the reasoning behind a decision, not just the decision; exact reproduction steps, commands, IDs, and values that worked or broke; subtle constraints, edge cases, and gotchas discovered along the way; and anything noticed but not yet acted on. When a detail lives only in the live conversation, it belongs here — captured in full, not summarised away. Err on the side of more detail: richer is safer than terser, as long as every line is a real, checkable fact.
The three behaviours
- At project start — if
HANDOFF.mddoes not already exist, create it fromassets/HANDOFF_template.md. - On every significant change — update
HANDOFF.mdwithout being asked. This is a standing default, not a task to wait for. Do it quietly as part of the work — do not turn every edit into an announcement or stop to ask permission. - At the start of a session, and after any context compaction — treat it as a cold start: re-read
HANDOFF.mdfirst, summarise the current state and next steps in 2–3 lines, and do a quick state-check before continuing. A compaction replaces the live conversation with a lossy summary, soHANDOFF.mdis what restores the real state and tells you where the reset conversation should resume. On surfaces with no persistent filesystem (Chat, Design), there is no automatic file to read, so look for one the user has provided — uploaded, pasted, or kept in a Project — before assuming there is none. If noHANDOFF.mdis present and this is clearly an ongoing project, offer to start one.
A significant change is: a decision made or reversed; a file created, renamed, moved or deleted; a change of approach or stack; a milestone completed; a blocker raised or cleared; a change of scope; an external resource added; next steps altered. When in doubt, if a returning reader would want to know it, log it.
Where HANDOFF.md lives
This skill applies on all four surfaces — Chat, Cowork, Code, and Design. The behaviours above are the same everywhere; only persistence differs, so put the file where it will actually survive on each:
- Claude Code —
HANDOFF.mdat the repository root. Edit the real file on each change, and commit it so teammates inherit it. It coexists withCLAUDE.md: that file holds standing rules and conventions, this one holds state and history. - Cowork —
HANDOFF.mdin the project folder, wherever it is synced (Google Drive, OneDrive, Dropbox, or local disk). If it sits in a cloud-sync client that streams files on demand (Google Drive for Desktop, OneDrive Files On-Demand, Dropbox online-only) and writes fail, the file is likely an online-only placeholder — mark the file or folder "Available offline" / "Always keep on this device" to fix it. - Chat (claude.ai) — there is no persistent file between conversations and no background process. Maintain
HANDOFF.mdwithin the session and hand it back as a file the user can save. To carry it across conversations it must live in a Project as a reference document, or be pasted/uploaded at the start of the next session. A Preferences line can make the behaviour default, but the file still has to be stored deliberately. - Claude Design — canvas- and session-oriented, the least file-based surface. Keep the state inline and give the user the file to store elsewhere.
Structure
Use assets/HANDOFF_template.md as the exact skeleton — copy it and fill it in rather than rebuilding the layout from memory, so the format stays identical across sessions. The italic convention notes under each section are meant to stay (they keep the file self-explaining); the [bracketed] bits are placeholders to replace with real content.
The sections and why each exists:
- Header — project name, last-updated timestamp, phase, surface(s). Orientation at a glance.
- 1. Summary / North Star — what the project is and the end goal.
- 2. Current state — where work left off, what is working, what is outstanding. The first thing a returning reader checks.
- 3. Map — a table of files, repos, datasets, connectors and APIs (name · where · what it does), so nothing important is invisible.
- 4. Conventions and glossary — naming, stack, terminology a newcomer would not know.
- 5. Decision log — append only, newest at the bottom: date · decision · why · alternatives rejected. Stops settled choices being silently relitigated.
- 6. Change log — append only, newest at the top: date · what changed. The running history.
- 7. Next steps — editable; keep current.
- 8. Open questions / blockers — editable.
- 9. Notes & detail — the delta (see Capture the delta): granular working knowledge that a compaction summary, auto-memory, and
CLAUDE.mdwill not preserve — gotchas, environment quirks, exact commands / IDs / values, approaches tried and why they failed, partial diagnoses, and things noticed but not yet acted on. This is where "detailed" lives; expand it freely.
Discipline
These rules are what keep the file trustworthy:
- Real facts, never placeholders. If a detail does not exist yet, write
(TBC)— never invent a plausible-sounding value, and strip every[bracketed]prompt from the template once filled. A handoff full of fiction is worse than no handoff. - Detailed, but factual. Capture as much real detail as the cold-start test and the delta need — err toward including specifics that live only in the conversation. It is a working reference, not a narrative: dense with facts, free of padding and restatement. More true detail is good; vague or invented detail is not.
- Sections 5 and 6 only grow. They are the historical record — append, do not edit or delete.
- Do not reopen logged decisions unless the user asks.
- Current instruction wins. If what the user says now conflicts with the file, the new instruction takes precedence — follow it and update the file to match.
- Australian English, ISO dates (
YYYY-MM-DD), so the file is sortable and matches the user's conventions.
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.

