onboard
Skill0 starsChanged 3 months agoMITView on GitHub
Use this on day one, right after someone clones this kit, or whenever they say "set me up," "onboard me," "let's get started," "fill in my info," or "get this ready for my business." Also fires on complaints like "this still thinks I'm the demo company," "why does the agent keep talking about a business that isn't mine," or "I never actually told it who I am." Runs a one-question-at-a-time interview, archives the demo content, and writes the operator's real context files, wiki indexes, voice profile, and first rule. Safe to re-run any time — it updates in place, will not archive the demo twice, and will not overwrite a wiki index that already holds real nodes.
171 other projects use a version of this.
Adopted by
Most recently taken up by
- TsukiNi22/skillstheir own version, first seen Nov 16, 2026
- TsukiNi22/ai-utilstheir own version, first seen Nov 16, 2026
- akovfoj-lang/plainframetheir own version, first seen Nov 16, 2026
- chuckplayer/claude-agent-packtheir own version, first seen Nov 16, 2026
- jestatsio/codebase-mentortheir own version, first seen Nov 15, 2026
- chuanglibs/Claude-Code-Game-Studios-Codextheir own version, first seen Nov 15, 2026
Used with
Projects with it also use soak-test, team-live-ops 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.
lilangMaxSkillsoak-test
Generate a soak test protocol for extended play sessions. Defines what to observe, measure, and log during long play sessions to surface slow leaks, fatigue effects, and edge cases that only appear after sustained play. Primarily used in Polish and Release phases.
Also in 10 projects that use this
lilangMaxSkillteam-live-ops
Orchestrate the live-ops team for post-launch content planning: coordinates live-ops-designer, economy-designer, analytics-engineer, community-manager, writer, and narrative-director to design and plan a season, event, or live content update.
Also in 10 projects that use this
HK-hubSkillteach-impeccable
One-time setup that gathers design context for your project and saves it to your AI config file. Run once to establish persistent design guidelines.
Also in 18 projects that use this
CoralGameSkillteam-level
Orchestrate level design team: level-designer + narrative-director + world-builder + art-director + systems-designer + qa-tester for complete area/level creation.
Also in 12 projects that use this
CoralGameSkillplaytest-report
Generates a structured playtest report template or analyzes existing playtest notes into a structured format. Use this to standardize playtest feedback collection and analysis.
Also in 12 projects that use this
HK-hubSkillarrange
Improve layout, spacing, and visual rhythm. Fixes monotonous grids, inconsistent spacing, and weak visual hierarchy. Use when the user mentions layout feeling off, spacing issues, visual hierarchy, crowded UI, alignment problems, or wanting better composition.
Also in 13 projects that use this
- +5 more from lilangMax
- +1 more from CoralGame
Other versions171 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 171
- Donchitos/Claude-Code-Game-Studios★ 39ksame name
- parcadei/Continuous-Claude★ 3.9ksame name
- parcadei/Continuous-Claude-v2★ 3.9ksame name
- parcadei/Continuous-Claude-v3★ 3.9ksame name
- nateherkai/AIS-OS★ 1.4ksame name
- poshan0126/dotclaude★ 871same name
- me2resh/apexyard★ 542same name
- nexmoe/DayBee★ 424same name
- nexmoe/eve★ 424same name
- kousen/claude-code-training★ 352same name
161 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
7 rules. 0 are common, 4 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 159 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Read context/me.md. If it is missing, empty, or opens with > 🟡 DEMO — /onboard replaces this file with yours., this is a first run — the …
- Very specific: line 25
Ask: "What language do you want this workspace to work in?" Everything you write from this point forward — every interview question, every …
- Distinctive: line 32
Don't archive by path list — the demo grows over time and a list goes stale the moment it does.
- Distinctive: line 73
Ask each of these eight in its own message and wait for the answer before asking the next:
- Distinctive: line 81
When the agent writes as you, how should it sound — and what should it never sound like?
- Very specific: line 130
Read .claude/rules/README.md first — it is the format contract every rule in that folder follows, and this step must match it exactly: the …
- Distinctive: line 138
Don't recap what you wrote.
“Common” is measured: kept by at least 6 in 10 of the 159 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
Onboard
A kit that ships with a demo company works perfectly in the demo and lies in every session
after that, because nobody replaced the example data with the truth. Every file the agent
reads at the start of a conversation — who you are, what you sell, who's on the team, what
the quarter's goal is — still describes a business that does not exist. That is poisoning,
and it is silent: nothing errors, the agent just confidently gets the operator wrong forever.
This skill replaces the demo with the operator's real context, on record, once.
Procedure
- Check whether this is a first run or a re-run.
Read
context/me.md. If it is missing, empty, or opens with> 🟡 DEMO —/onboardreplaces this file with yours., this is a first run — the demo is still installed. If the demo banner is gone and the file already describes a real operator, this is a re-run: skip step 3 (the demo is already archived) and go straight to the interview to refresh the files in place. - Ask the working language before anything else.
Ask: "What language do you want this workspace to work in?" Everything you write from this
point forward — every interview question, every file in
context/, the new rule, the decision log entry — goes in that language. Do not translateCLAUDE.md, this skill, or anything else under.claude/; those stay in English so the kit behaves the same for every install regardless of who's running it. - Archive the demo before writing anything new — first run only.
Don't archive by path list — the demo grows over time and a list goes stale the moment it
does. Instead, find every file by its first line: the demo banner is exactly
> 🟡 DEMO —/onboardreplaces this file with yours., and every demo file in this kit carries it as a global constraint. Run, in order:mkdir -p archives/demo- Walk every tracked and untracked file in the repo, excluding
.git/and anything already underarchives/demo/(so a second run can't re-archive its own output), and test whether line 1 matches the banner exactly. For each match, preserve its relative path underneatharchives/demo/—knowledge/playbook/index.mdbecomesarchives/demo/knowledge/playbook/index.md, not a flat dump — creating parent directories as needed. Usegit mv <path> <dest> || mv <path> <dest>for each:git mvfails whenever the path isn't tracked by git — a fresh clone with nothing committed yet, or a path a previous partial run already moved with plainmv— and themvfallback still lands it inarchives/demo/either way. Skip a path that's already gone; a previous run (or this loop, on a second pass) may already have moved it. find projects -mindepth 1 -type d -empty -delete— a project folder that existed only to hold the demo's files is now an empty shell, and an emptyprojects/<name>/reads as a live engagement that isn't one. Run this only underprojects/: the routing map names no individual project, so nothing there is promised to exist.mkdir -p context knowledge/external knowledge/playbook decisions references/sops— recreate every directory the routing map inCLAUDE.mdpromises exists, even the ones now emptied by the move above, so the map never points at nothing. These five are the opposite case from the line above: the map declares them, so they come back empty rather than being swept away.touch decisions/log.mdDirectories are not enough. The banner predicate is right for demo content, but it also catches files that are machinery — anindex.mdis the entry point the routing map promises, andreferences/voice.mdis a routing row of its own. Step 5 writes both back; an emptyknowledge/folder or a missing voice profile is the kit failing its own Routing check, produced by its own installer. (The twoknowledge/*/CLAUDE.mdingestion manuals carry no banner on purpose — they are format contracts, not demo data, and survive this step untouched.) Tell the operator plainly, in your own words: nothing was deleted, the original demo content now lives inarchives/demo/under the same paths it came from, and they can open it any time. Doing this before you write the new files matters — write first and archive after, and the operator's own answers get swept intoarchives/demo/along with the fictional company. Because the check is "does line 1 match," not "is this on a list," a second run finds no banners left to match and does nothing — archiving is idempotent by construction, not by a flag you have to remember to check.
Run the interview. One question per message — never a form, never batched. Ask each of these eight in its own message and wait for the answer before asking the next:
- Who are you, and what do you do?
- What do you sell, and to whom?
- What's your goal for this quarter?
- Who's on your team, and when should the agent pull each of them in?
- What tool holds the actual truth about your tasks, and about your money?
- What eats your time over and over, that you wish ran on its own?
- What should the agent never do, no matter what?
- When the agent writes as you, how should it sound — and what should it never sound
like? Ask for the register too: does anything change between an internal note and
something a client reads?
If the answer to question 4 is "nobody, I work alone," don't accept it and move on.
Taken literally it produces a two-line
context/team.mdin the always-loaded layer — a file that costs budget in every session and answers no question anyone will ask. Ask the follow-up instead: who do you rely on who isn't on your payroll? A subcontractor you call when the work overflows, a bookkeeper or accountant, a designer or editor you always use, a lawyer, a peer you check hard decisions with. Record each one exactly as you would an employee — name, what they cover, and the trigger that pulls them in — and say plainly in the file that these are outside the company. Someone always has a team; it is just not payroll. If they genuinely rely on no one at all, write that as the one line it is and move on — a short true file beats an invented roster.
On a re-run, show the current content of each file first and ask only what changed — don't re-run the full interview against someone who already answered it.
- Write the operator's files —
context/, both wiki indexes, and the voice profile.context/me.md— answer 1, in plain first-person-about-the-operator terms.context/work.md— answer 2 (what's sold, to whom), plus answer 5 written as a pointer, not a copy: name the tool that holds task and money status, and say explicitly that its numbers are never duplicated here. Add answer 6 under a short "recurring friction" note — this is what/leveragereads when it looks for something worth automating.context/team.md— answer 4: names, roles, and the trigger for looping each person in.context/priorities.md— answer 3. State it as the one place quarter goals get edited, and warn against copying it anywhere else — a copied goal is correct today and wrong the day it changes, because nobody remembers to update the copy.decisions/log.md— append one entry:[YYYY-MM-DD] DECISION: workspace initialized via /onboard | REASONING: replaced demo content with the operator's real context | CONTEXT: operator said the agent must never <answer 7, verbatim>.knowledge/external/index.mdandknowledge/playbook/index.md— the demo's indexes were archived in step 3, and the routing map promises each of these folders is a[[wiki-link]]graph with its ownindex.md. Write both back as empty stubs, not as content: a# knowledge/<folder> — indextitle, one line naming what this wiki holds and pointing at theCLAUDE.mdbeside it for the node format, then a## Nodesheading with nothing under it. Invent no nodes — the list stays empty until/distillwrites the first one, and/distillneeds the file to exist so it has something to append to. On a re-run, only create these if they are missing. An index that already lists real nodes is months of accumulated knowledge — overwriting it with a stub would be the worst damage this skill could do.references/voice.md— answer 8, as a short profile with four parts: tone, what the operator sounds like, what they avoid, and how internal writing differs from client-facing. The routing map promises a voice profile inreferences/, and the demo's was archived — leaving this unwritten points that row at nothing. Write only what the operator actually said; if answer 8 was thin, say so in the file and leave the sections short rather than filling them in. A guessed voice ispoisoningin the one file whose entire job is to sound like them.
- Turn answer 7 into a standing rule.
Read
.claude/rules/README.mdfirst — it is the format contract every rule in that folder follows, and this step must match it exactly: the rule stated plainly, then the reasoning behind it. If the operator's answer to question 7 didn't include a reason, ask one follow-up — "why does that matter?" — before writing the file. Name the file after the topic, not after the word "rule" (client-email-review.md, notrule-1.mdornever-do.md). Write it into.claude/rules/. - Close with proof, not a summary. Don't recap what you wrote. Tell the operator to ask this exact question next: "based on what you now know about me, what should I focus on this week?" That question only lands if the files you just wrote were actually read — it's the test, not a courtesy.
Idempotent by construction. Step 1 is the only gate that matters: the demo banner's
presence in context/me.md is the single source of truth for "has this already run." A
re-run always updates files in place through steps 2, 4, 5, and 6; it only skips step 3, and
only when the banner is already gone. Never check archives/demo/ for this — a manually
copied backup there would produce a false positive and leave the demo installed forever.
When NOT to run this
context/already describes the operator, and one fact in it is now wrong. Don't run the interview again for a single correction — edit the file directly, or run/route-fixif the agent acted on the wrong fact and you need to trace why.- Raw material — a transcript, a document, a pile of notes — needs to become knowledge.
That's
/distill, not this./onboardsets up who the operator is;/distillis the ongoing feed of what they learn afterward.
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.

