agentleFS
EnzoSacinRuizEnzoSacinRuiz/aios-kit/.claude/skills/onboard/SKILL.md

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

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.
  • lilangMaxSkill

    soak-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

  • lilangMaxSkill

    team-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-hubSkill

    teach-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

  • CoralGameSkill

    team-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

  • CoralGameSkill

    playtest-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-hubSkill

    arrange

    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

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
  1. Very specific:

    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 …

    line 18
  2. Very specific:

    Ask: "What language do you want this workspace to work in?" Everything you write from this point forward — every interview question, every …

    line 25
  3. Distinctive:

    Don't archive by path list — the demo grows over time and a list goes stale the moment it does.

    line 32
  4. Distinctive:

    Ask each of these eight in its own message and wait for the answer before asking the next:

    line 73
  5. Distinctive:

    When the agent writes as you, how should it sound — and what should it never sound like?

    line 81
  6. Very specific:

    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 …

    line 130
  7. Distinctive:

    Don't recap what you wrote.

    line 138

“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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

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

  1. Check whether this is a first run or a re-run. 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 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.
  2. 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 translate CLAUDE.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.
  3. 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 — /onboard replaces 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 under archives/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 underneath archives/demo/ — knowledge/playbook/index.md becomes archives/demo/knowledge/playbook/index.md, not a flat dump — creating parent directories as needed. Use git mv <path> <dest> || mv <path> <dest> for each: git mv fails 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 plain mv — and the mv fallback still lands it in archives/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 empty projects/<name>/ reads as a live engagement that isn't one. Run this only under projects/: 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 in CLAUDE.md promises 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.md Directories are not enough. The banner predicate is right for demo content, but it also catches files that are machinery — an index.md is the entry point the routing map promises, and references/voice.md is a routing row of its own. Step 5 writes both back; an empty knowledge/ folder or a missing voice profile is the kit failing its own Routing check, produced by its own installer. (The two knowledge/*/CLAUDE.md ingestion 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 in archives/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 into archives/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.
  4. 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:

    1. Who are you, and what do you do?
    2. What do you sell, and to whom?
    3. What's your goal for this quarter?
    4. Who's on your team, and when should the agent pull each of them in?
    5. What tool holds the actual truth about your tasks, and about your money?
    6. What eats your time over and over, that you wish ran on its own?
    7. What should the agent never do, no matter what?
    8. 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.md in 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.

  5. 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 /leverage reads 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.md and knowledge/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 own index.md. Write both back as empty stubs, not as content: a # knowledge/<folder> — index title, one line naming what this wiki holds and pointing at the CLAUDE.md beside it for the node format, then a ## Nodes heading with nothing under it. Invent no nodes — the list stays empty until /distill writes the first one, and /distill needs 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 in references/, 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 is poisoning in the one file whose entire job is to sound like them.
  6. Turn answer 7 into a standing rule. 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 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, not rule-1.md or never-do.md). Write it into .claude/rules/.
  7. 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-fix if 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. /onboard sets up who the operator is; /distill is the ongoing feed of what they learn afterward.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.