handoff
Skill75 starsChanged 2 months agoMITView on GitHub
Compacts the current conversation into one self-contained, tool-agnostic handoff document at docs/handoffs/<slug>.md so any fresh agent or teammate can resume the work from that file alone. References pushed artifacts by path or URL, inlines anything not on the remote, and redacts secrets and PII. Re-running on an existing handoff offers enhance or overwrite. Writes one project artifact and, when needed, a private scratchpad backup; never commits.
- Reads credentials
191 other projects use a version of this.
Adopted by
Most recently taken up by
- xiaoxiaosrm/dsh-mattpocock-skillstheir own version, first seen Nov 8, 2026
- ethosagent/ethostheir own version, first seen Nov 8, 2026
- dsifry/metaswarmtheir own version, first seen Nov 8, 2026
- glebis/claude-skillstheir own version, first seen Nov 8, 2026
- domengabrovsek/agent-configtheir own version, first seen Nov 8, 2026
- domengabrovsek/claudetheir own version, first seen Nov 8, 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★ 309ksame name
- QuivrHQ/quivr★ 40ksame name
- The-Vibe-Company/quivr★ 40ksame name
- GreyDGL/PentestGPT★ 16ksame name
- ykdojo/claude-code-tips★ 10ksame name
- ArcReel/ArcReel★ 6.3ksame name
- vinvcn/mattpocock-skills-zh-CN★ 5.5ksame 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.
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .claude/CLAUDE.mdCLAUDE.md
- brainstormSkill
- code-reviewSkill
- commitSkill
- deepSkill
- dreamSkill
- gauntletSkill
- how-to-runSkill
- initSkill
- jiraSkill
- paper-initSkill
- permissionsSkill
- promptSkill
- prSkill
- refactorSkill
- resetSkill
- tddSkill
- unit-testSkill
- worktreeSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 172 edited versions we read keep these.
2
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
- Distinctive: line 39
Never duplicate tracked-artifact content — reference by path plus a one-line summary.
- Distinctive: line 41
Keep the document tool-agnostic: "a fresh agent" / "a new session" — never a named AI product.
- Very specific: line 47
Write the document to docs/handoffs/<slug>.md, creating the folder if missing.
“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
Step 1: Slug and focus
Derive a kebab-case <slug> from the user's arguments if given (e.g., "finish the migration tests" → migration-tests), otherwise from the conversation's most recent active thread; with no arguments, state the inferred topic and slug in one line before continuing. Record a Focus for next session only on a clear signal — verbatim user arguments or an explicitly stated next objective. Never manufacture one: with no signal, omit it and leave direction to the resumer.
Step 2: Locate the doc; create, enhance, or overwrite
Resolve the root: when git rev-parse --is-inside-work-tree returns true, use git rev-parse --show-toplevel, including in a linked worktree or subdirectory. Otherwise read $CLAUDE_PLUGIN_ROOT/skills/init/references/multi-repo-detection.md and apply it, then cover both of its non-workspace answers: when it finds exactly one child repo, treat that child repo as the root — writing above it would put the handoff outside version control, where the Step 5 /optimus:commit recommendation cannot reach it; when it finds no recognized structure, use the current directory. A multi-repo workspace uses its workspace root. The handoff folder is docs/handoffs/ under that root. Then branch:
<slug>.mdexists → read it, thenAskUserQuestion: Enhance (merge new context; keep still-valid content and append a History line) or Overwrite (fresh rewrite). Follow an already explicit choice without asking again. Before replacing existing content, confirm its exact current bytes are recoverable in Git; otherwise copy them to the session scratchpad and report the backup path. If no private scratchpad is available, preserve the file and write a new slug instead. Never claim uncommitted content is in Git history.- No slug match but other handoffs exist → list them (filename · title · Last updated), then
AskUserQuestion: Continue one (pick via a follow-up question, adopt that file's slug as<slug>, treat as Enhance) or Create new. - Folder empty or absent → create new.
Step 3: Classify artifacts against version control
Per repo (per child repo in a multi-repo workspace, where paths and SHAs are also repo-qualified): find the branch and short HEAD SHA for Origin, each path's tracked / modified / staged / untracked status, and unpushed commits via git log --oneline HEAD --not --remotes (commits no remote-tracking branch contains). Not a git repo → note "not a git repo" in Origin and inline everything. Sort every artifact into three buckets:
- Tracked and pushed → reference by repo-relative path / SHA / URL, never inline.
- Tracked-but-modified, staged-not-committed, or committed-not-pushed → inline the relevant content (diff, file body, or commit message) — another clone will not have it.
- Untracked → inline.
Step 4: Draft or reconcile
If the conversation has not already established the codebase's current state, briefly verify it in the repo before asserting it. Fill the Handoff document template. New or overwrite → write fresh. Enhance → keep still-valid content; update Current state (and Goal/Next steps if present) to current reality, dropping completed steps; preserve and extend decisions and open questions, promoting any the new work resolved; add new artifacts; refresh Last updated; append one History line.
- Prioritize knowledge a fresh agent could not re-derive from code or git history — decisions and why, rejected alternatives, constraints, gotchas, open questions.
- Never duplicate tracked-artifact content — reference by path plus a one-line summary. Inline only bucket-2 and bucket-3 content.
- Trust git state over chat memory when they conflict; if the current state is ambiguous, say so in one line rather than guessing.
- Keep the document tool-agnostic: "a fresh agent" / "a new session" — never a named AI product.
Step 5: Redact, write, verify, report
Redact as you write: every line of the document — references, authored prose, and inlined content — must be checked against the Redaction patterns table before it goes to disk, with matches replaced by the exact marker [REDACTED: <kind>] and the structure preserved (e.g. DATABASE_URL=postgres://app:[REDACTED: password]@db:5432/app). Inspect URL userinfo, query parameters, and fragments for credentials or signed-access tokens; retain the public resource location after removing sensitive access data, or use a non-clickable redacted reference when no public URL is usable. Ordinary paths and SHAs need no prose rewrite, but are not exempt from secret scanning. A file whose name looks like a secret is never inlined with its values, regardless of tracked state — see the table's last two rows. This file is destined for /optimus:commit, and removing a leaked credential from Git does not revoke it.
Write the document to docs/handoffs/<slug>.md, creating the folder if missing.
Then verify: read the file back from disk and re-scan the full document, including every reference and URL, against the Redaction patterns table, fixing any hit before you report. This pass checks what actually landed; it reduces exposure risk but is not a guarantee that every secret has been detected.
Report the written path and any scratchpad backup. If the resolved root is not itself a git repo — a multi-repo workspace root, or a directory with no recognized structure — note the file is not under version control; suggest moving it into a child repo or initializing version control at the root before /optimus:commit. Recommend /optimus:commit so the handoff reaches the remote — staying in this conversation, so the context being handed off is captured — and that the resumer point a fresh session at the written file. This skill writes one project artifact (docs/handoffs/<slug>.md, creating the folder if missing), plus a private scratchpad backup when needed; it never stages, commits, or pushes.
Handoff document template
Omit any section with nothing to say. Square-bracketed lines are author instructions — never emit them into the document.
# Handoff: <short title of the work>
- **Created:** <YYYY-MM-DD> · **Last updated:** <YYYY-MM-DD>
- **Origin:** <repo> @ `<branch>` · commit `<short-sha>` (<pushed | NOT pushed>)
- **Focus for next session:** <one line — include only if the user passed args or the conversation states the next objective; otherwise omit this bullet>
- **How to use this doc:** Resume cold; read top to bottom — everything is here or linked by repo-relative path.
## Goal
<Include only if the conversation establishes a clear objective. 1-3 sentences: the outcome and why. Omit if not clear.>
## Current state
<2-6 bullets: what is done, in progress, blocked. Prioritize what a fresh agent could not re-derive — decisions made and why.>
**Decisions, constraints & gotchas** [omit if none]:
- <A decision made and why — and any alternative considered and rejected, with the reason.>
- <A non-obvious gotcha, or a constraint that gates the work.>
**Open questions** [omit if none]:
- <Something the conversation left unresolved — what is undecided and what depends on it.>
## Next steps
<Include only if the conversation points to concrete follow-up actions; otherwise omit and leave direction to the resumer.>
1. <Concrete first action on resume.>
## Relevant files & artifacts
> Committed/tracked → referenced by repo-relative path / SHA / URL.
> Uncommitted or unpushed → content inlined below (won't exist on another machine).
[Anchor each reference on a durable identifier (a type / function / config name the resumer can search for) and say why it matters; never cite line numbers — a bare path can move under a refactor.]
- `path/to/file.ts` — `SessionStore`: <why it matters> (tracked)
- Issue/PR: <URL> — <relevance>
### Inlined (not yet on remote)
[Omit this section if nothing is inlined.]
```diff
<the fragment that carries the decision — schema, type shape, state machine, or the changed lines that matter, not a full diff dump; or new-file body / unpushed commit message; secrets redacted>
```
## History (most recent first)
- <YYYY-MM-DD>: <one-line summary of what this session changed>
---
**Resume here:** open this file in a fresh session with any AI agent in this repo. If it has a Next steps section, begin at item 1; otherwise pick up from Current state and the decisions above.Redaction patterns
| Class | Catch | Marker |
| API keys / tokens | sk-…, ghp_…, xox[bap]-…, AWS AKIA…, Google AIza…, JWTs eyJ…, bearer | [REDACTED: API key] / [REDACTED: token] |
| Passwords / secrets | password=, creds in connection strings, client_secret, -----BEGIN … PRIVATE KEY----- | [REDACTED: password] / [REDACTED: private key] |
| URL access data | userinfo credentials; query/fragment secrets such as token, access_token, api_key, sig, and signed-download parameters | remove access data from a usable public URL, otherwise [REDACTED: URL credentials] in a non-clickable reference |
| Connection strings | mongodb+srv://…:…@, mssql://…;Password=… | [REDACTED: connection string] |
| PII | personal emails (keep role addresses like support@), phones, addresses, national IDs | [REDACTED: PII] |
| Env/secret files (text) | inlined .env / *.key / *.pem / credentials.* / secrets.* bodies | inline variable names only, never values |
| Binary/opaque secret files | *.pfx / *.sqlite / *.db bodies | never inline — reference by name with a "recreate from your secure source" note |
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.

