onboard
Skill0 starsChanged 24 days agoMITView on GitHub
Set up this workspace for its user by interviewing them: what they are here for, who they are, what they are working on and how the AI should work with them; then create their first domains, connect where their notes and tasks already live, choose a weekly rhythm, move in the add-ons that fit, and finish with one real task. Use when PROFILE.md is still the blank template or says onboarding is in progress, when the user wants to update their setup because something changed, or when they want to add an add-on.
168 other projects use a version of this.
Adopted by
Most recently taken up by
- jestatsio/codebase-mentortheir own version, first seen Nov 15, 2026
- chuanglibs/Claude-Code-Game-Studios-Codextheir own version, first seen Nov 15, 2026
- 5uck1ess/devkittheir own version, first seen Nov 15, 2026
- suitedaces/dorabottheir own version, first seen Nov 14, 2026
- danielrosehill/Claude-Audio-Production-Plugintheir own version, first seen Nov 13, 2026
- Donchitos/Claude-Code-Game-Studiostheir own version, first seen Nov 9, 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 9 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 9 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 19 projects that use this
CoralGameSkillproject-stage-detect
Automatically analyze project state, detect stage, identify gaps, and recommend next steps based on existing artifacts.
Also in 15 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 11 projects that use this
- +5 more from lilangMax
- +2 more from CoralGame
Other versions168 projects use a version of this.
Word for word · 1
- aminbadh/agent-workspace★ 0100% same
Edited · 0
None found.
Wrote their own version · 167
- 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
- talesofai/cohub★ 585same name
- me2resh/apexyard★ 542same name
- nexmoe/DayBee★ 424same name
- nexmoe/eve★ 424same name
157 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 repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- one-thingSkill
- upgradeSkill
- weekly-reviewSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
20 rules. 0 are common, 16 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 155 edited versions we read keep these.
16
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
Never invent: Write their words, not a plausible version of them.
- Distinctive: line 30
Ask one open question: what do they want this workspace to help with?
- Distinctive: line 38
Ask who they are (name, what they do, where they are based, languages), what they are working on (the two to five big things), and how they …
- Distinctive: line 52
Create a domain for each thing they are working on, rather than explaining the pattern and leaving it empty.
- Distinctive: line 61
Ask where their notes, documents and plans live today: a notes app, a folder of files, Notion, Google Docs, a paper notebook, their head.
- Distinctive: line 71
Whether you have a browser: Check, do not assume: web search is not a browser.
- Distinctive: line 73
Where tasks live: Ask what they use for tasks and dates: an app (Todoist, Things, TickTick, Notion, Linear, a calendar), a file, paper, …
- Distinctive: line 82
Ask about their habits, using the questions at the top of the rhythms reference: do they plan now and how, which day suits a review and how …
- Distinctive: line 93
When they are applying to something.
- Very specific: line 101
First CV: use the cv-engine skill to start the CV folder at cv/ inside the applications domain, write the main CV from experience.md, build …
- Distinctive: line 109
Ask where their code repos live.
- Very specific: line 110
Check what is already there (ls -la ~/.links).
- Very specific: line 111
Only once they agree, run the commands from BUILDING.md with their folder, and show the result with ls -la ~/.links.
- Distinctive: line 120
A domain with a clear next step: do it, or draft it (an email, a plan, a summary of the notes you read).
- Distinctive: line 121
"What should I do now?": run the one-thing skill.
- Distinctive: line 122
Building: write the first brief for their repo, as BUILDING.md describes.
- Very specific: line 126
Remove PROFILE.md's progress line: onboarding is done.
- Distinctive: line 132
Do not commit: they review and commit (AGENTS.md).
- Distinctive: line 138
The rhythm, after a few weeks: if the weekly review reports that the rhythm is not working, ask what got in the way, suggest another from …
- Distinctive: line 148
AGENTS.md: add its row or rows to the skills table.
“Common” is measured: kept by at least 6 in 10 of the 155 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
Onboarding
The first session decides whether the workspace makes sense to the user. They came for an outcome, not for a system, and they may never have used an AI this way. So: short, warm, plain words, and something real at the end.
Three modes. If PROFILE.md still says "Blank template", run the full onboarding below. If it says "Onboarding in progress", resume: read what is already saved, say in two lines what is done, and carry on from the first unfinished step without asking anything twice. Otherwise run update mode. If the user only wants an add-on, go straight to moving an add-on in.
Ground rules
- A conversation, not a form. Two or three questions per message. Follow up on vague answers ("data stuff": "analysing data, building dashboards, or something else?"). If they want to skip something, skip it and note it as unknown.
- Save after every answer, not at the end. A session can end at any moment, and the user should never have to answer twice. Write what each message told you into the right file before asking the next questions, and keep PROFILE.md's progress line current (
*Onboarding in progress: steps 1 to 3 done.*), so an interrupted setup can resume. - Never invent. Write their words, not a plausible version of them. Mark anything they were unsure of "(to confirm)".
- Search for the outside world (AGENTS.md, rule 7): whether their task app has an MCP server, how to export from their notes app. Lead with the established route.
- Get today's date from the machine (
date +%Y-%m-%d), never from memory.
Before the first question
Say, in two or three sentences, in the language they wrote in:
- This folder is theirs, on their computer. The AI reads and writes its files; nothing is uploaded.
- A few questions now, about fifteen minutes, mean they never have to explain themselves again.
- At the end you will do one real thing for them with what they told you.
1. What are you here for?
Ask one open question: what do they want this workspace to help with? The answer picks the paths that run. Everyone gets the core (steps 2 to 7). Then:
- Applying to something (internships, jobs, a Master's, scholarships, grants, competitions): the applications path.
- Writing code, for a project of their own or at work: the building path.
- Anything else: the core is enough. Offer the add-ons at step 7 only if something they said fits one.
2. Who they are
Ask who they are (name, what they do, where they are based, languages), what they are working on (the two to five big things), and how they want the AI to work and write for them:
- Replies: the language, the tone (brief or detailed, formal or casual).
- Writing: offer the list of AI writing tells and ask which to avoid, in replies and in anything written for others. Add anything else they want or never want (jargon, emoji, a phrase that annoys them).
- What the agent may do on its own: show the default ladder in AGENTS.md's "What the agent may do" (alone, draft only, never) and ask what to move. Connected email, task or calendar tools make this matter: say so.
Write it down:
- PROFILE.md: fill the fields from their answers. Add frontmatter at the top (
domain: Profile,parent: root,status: active,review_interval: 60,last_reviewed:today) and replace the "Blank template" line with the progress line, which is how AGENTS.md knows onboarding has started and not finished. - AGENTS.md, "Working with me": replace the "Not set up yet" line with their preferences, as short bold-led bullets in the second person ("Replies in French. Code and file names stay in English."). Their writing rules go here and nowhere else, and their ladder replaces the default in "What the agent may do".
- Read it back. Show them PROFILE.md and the two AGENTS.md sections as written, and ask them to approve or correct them before going on. Nothing else happens until they have seen what was written about them.
3. Their first domains
Create a domain for each thing they are working on, rather than explaining the pattern and leaving it empty. Read CONVENTIONS.md first.
- Place each one in an area of life: a top-level folder such as
studies/,work/orpersonal/, named in their words. One domain in an area can sit straight in the area folder's place (studies/as the domain itself). Two or more in one area: the area gets its own indexCONTEXT.mdwith one row per child, as CONVENTIONS.md describes, and each child gets a folder inside it. - Write each from the templates in assets/:
CONTEXT.mdandDECISIONS.md. Fill the overview and current picture from what they said, and ask for what is missing: what is being done now, what is next, any hard limits, who matters. A domain they are not working on now isdormant. - No dates in domain files. A deadline they mention goes to the task home once it exists (step 5); until then, note it and carry it over.
- Link each domain from PROFILE.md's "What I am working on".
4. Where things already live
Ask where their notes, documents and plans live today: a notes app, a folder of files, Notion, Google Docs, a paper notebook, their head.
- What you can reach (a folder on this computer, an app with a connected tool): read it, fill the domains from it, and link the source from each domain's Key Resources rather than copying everything in. Copy only what the domain needs to say what is true now.
- What you cannot reach yet: search for the way in (an MCP server or connector for the app, an export format, a sync to a folder) and offer it, with the steps. If they set it up now, read it and fill the domains; if not, note in PROFILE.md where it lives and that it is not connected.
- An organised folder or vault they want to keep using (Obsidian, PARA folders, a numbered system): map it with them folder by folder. For each, ask whether it becomes a domain, part of one, a source to link, or stays as it is. Never infer the mapping from folder names, and move nothing without a yes. A folder that becomes a domain gets a
CONTEXT.mdfrom the template, written from what is in it. - Record in PROFILE.md's Setup where their notes live.
5. Tools and the task home
- Which AI tool they use, written in PROFILE.md.
- Whether you have a browser. Check, do not assume: web search is not a browser. If a browser tool is listed, open one page with it. Write down what you actually have, so later work knows its limits.
- Git. If
git remote -vshows the upstream named on README.md'sUpstream:line (the template this workspace was made from), say so: a weekly pull would pull the template's changes into their workspace. Offer to remove that remote (git remote remove origin), or to point it at a private repository of their own. They lose nothing: template fixes arrive through the upgrade skill, which reads theUpstream:line, not the remote. If the folder is not a git repository, offergit init, so every version of every file is kept; their choice. - Where tasks live. Ask what they use for tasks and dates: an app (Todoist, Things, TickTick, Notion, Linear, a calendar), a file, paper, nothing.
- An app: search whether it has an MCP server, a connector for their AI tool, or a command-line tool. If one exists and is connected, test it by listing a few of their tasks. If it exists but is not connected, give them the setup steps and offer to wait. Once it works, write AGENTS.md's "Tasks" section: the app, how the agent reaches it, the projects or lists that matter, how "done" and "this week" are marked. If no route exists, say so and suggest a TASKS.md instead.
- Anything else: copy the TASKS.md template to the workspace root. If they already keep tasks in a file or a notebook, move the open ones into it (verb first, dates kept) with their agreement, so there is one list, not two. Then write AGENTS.md's "Tasks" section:
Tasks live in [TASKS.md](TASKS.md)., with one line on how it is used. - Either way, replace the "Not set up yet" line, then move any deadline noted in step 3 into the task home.
6. A weekly rhythm
The weekly-review skill does the same upkeep for everyone; the planning and reflection part is theirs to choose.
- Ask about their habits, using the questions at the top of the rhythms reference: do they plan now and how, which day suits a review and how long, numbers or not, one big goal or several areas.
- Suggest one or two methods from the reference that fit, with a line each on why. If they name a method that is not there, search for it and add its entry to the reference in the same shape. If they want none, leave it: the review falls back to each domain's Next Up.
- Write the choice into PROFILE.md's "Weekly rhythm" section, replacing the "Not chosen yet" text: the method's name, linked to its entry in the reference, and today's date. Then add only what is theirs: the day and length of the review, any step they skip or add, weekly steps of their own that belong to no skill. Never copy the reference's steps into PROFILE.md: the reference is their one home, and a copy would drift from it.
- The Compass also installs COMPASS.md at the root: show the life areas from the reference, let them pick, rename and merge, write one section per area, and fill the ladders with them, marking anything you suggested (proposed). Scores stay off for a first setup.
7. Add-ons
Each folder in addons/ is an add-on, with a README saying what it gives. Offer the ones their answers point to, in one line each, and move in the ones they pick (see moving an add-on in). Never move one in unasked. Then run its path below.
The applications path
When they are applying to something. Move the applications add-on in, then:
- What they want: the type (internship, end-of-studies internship, job, Master's, scholarship, grant, competition), when it starts and for how long, where, and two or three fields or role families. For a type with school or funder rules (an internship's length, a scholarship's eligibility), ask what they know, then read the opportunity-research skill's guide.
- Hard limits: anything that rules an offer out (a place they cannot go, a language they cannot work in, a minimum stipend or funding level, constraints they choose to share).
- Languages, honestly rated: native, fluent, working, basic.
- Their experience. Ask for an existing CV first: a file in the folder, a PDF, a Word document, pasted text, or a LinkedIn profile saved as PDF. Read it. Then go section by section (education, work, projects, clubs and volunteering, awards, skills) asking for what they did and what came of it, with numbers where they exist. Most people undersell: "I helped with events" often hides "ran a 200-person event". Ask once, kindly, for the result. Write every fact into experience.md under its heading, newest first.
- Where it lives: the criteria from 1 to 3 go in PROFILE.md under a new "## What I am looking for" section, where the opportunity-research skill reads them. Create an applications domain (for example
career/orstudies/applications/) in step 3's shape, whose Constraints link to that section rather than restating it. The tracker is the task home if it is an app that can hold rows with fields, otherwiseopportunities.csvin the domain folder; name it in the domain's Key Resources. - What to watch: ask which careers pages, catalogues or programme pages they want watched for new openings. If they name some, start
sources.jsonbeside the tracker from the opportunity-research skill's example, with one entry per page, and runwatchonce to record the baseline. If they decline, say plainly that the weekly watch is off until they add a page, and that they can say "watch this page" any time. - First CV: use the cv-engine skill to start the CV folder at
cv/inside the applications domain, write the main CV from experience.md, build it and look at the page image. Ask whether they want another language. - Show them experience.md and the CV, and ask them to correct anything wrong now.
The building path
When they write code. Move the building add-on in, then ask which code repositories the workspace plans for, and give each one a domain in step 3's shape (the domain holds the reasoning; the repo holds the code). Then offer BUILDING.md's "Paths across machines", since builders reach their briefs through ~/.links/workspace. It changes their home folder, so nothing is created without a yes:
- Explain in two lines what it creates: a hidden
~/.linksfolder holding two links, one to this workspace and one to the folder holding their code repos. - Ask where their code repos live. Do not guess a folder.
- Check what is already there (
ls -la ~/.links). If a link exists and points elsewhere, say where, and ask before changing it. - Only once they agree, run the commands from BUILDING.md with their folder, and show the result with
ls -la ~/.links. If they decline, say that briefs then need full paths on this machine, and move on.
Then walk through "Setting up a repo" with them for the first one.
8. One real thing
Finish with a task that matches what they came for, using what they just told you, and point out that you did not need to ask anything again. Pick from what they said:
- Applications: find two or three real offers that fit with the opportunity-research skill, record and verify them; or tailor the CV to an offer they already have in mind.
- A domain with a clear next step: do it, or draft it (an email, a plan, a summary of the notes you read).
- "What should I do now?": run the one-thing skill.
- Building: write the first brief for their repo, as BUILDING.md describes.
9. Close
Remove PROFILE.md's progress line: onboarding is done. Run uv run .agents/scripts/workspace-lint.py and fix anything it reports. Then end with a few lines:
- what was created and where (their domains, the task home, the rhythm, any add-on);
- two or three things they can say next time ("what should I do now?", "it's review time", something from their add-on);
- that they can update their setup any time by saying what changed.
Do not commit: they review and commit (AGENTS.md).
Update mode
When PROFILE.md is already filled in, do not re-run the interview. Ask what changed (a new goal, a finished project, a move, a new tool), and update only what it touches: PROFILE.md, the domains (a new one, one gone dormant, one to archive), AGENTS.md's "Tasks", "Working with me" or "What the agent may do", experience.md and the CV if their facts changed. Bump last_reviewed on what you verified.
- The rhythm, after a few weeks: if the weekly review reports that the rhythm is not working, ask what got in the way, suggest another from the reference, and replace PROFILE.md's "Weekly rhythm" section with it. Installing or removing the Compass follows step 6.
- A new add-on: offer it if what changed points to one, and move it in as below.
Moving an add-on in
An add-on's README says what it gives, where each of its files goes, and the row it adds to AGENTS.md's skills table. To move one in:
- Its skills: move each folder in the add-on's
skills/into.agents/skills/(withgit mvif the folder is a git repository). - The Claude Code links: for each skill,
ln -s ../../.agents/skills/<name> .claude/skills/<name>, thenls -la .claude/skillsto check every link resolves. On a system where links fail (some Windows setups), copy the folder instead and say so. - Its files: move each to where its README says.
- AGENTS.md: add its row or rows to the skills table.
- Remove the add-on's now-empty folder, README included, so addons/ holds only what is not yet installed. The folder itself stays, so another add-on can be added later.
- Run
uv run .agents/scripts/workspace-lint.py.
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.

