agentleFS
pradeep-sankaran-07pradeep-sankaran-07/team-os-gdrive-template/.agents/skills/install-team-os/SKILL.md

install-team-os

Skill0 starsChanged 2 months agoMITView on GitHub

Set up, continue or check the installation of this Team OS in a company Google Shared Drive. Interviews the admin, writes the answers into config.yaml, shapes the folders to that company, and keeps INSTALL-CHECKLIST.md up to date so nothing is half-done. Triggers: "where do I start", "set up the team os", "install this", "what is left to set up", "check my install", "is the setup finished", "continue the installation", "add a new person to the OS". NOT for day-to-day work once the OS is running. NOT for changing access for one person, which is a Google Group membership change, not an install step.

The file

Show its rules and text

We split this skill into the rules it gives an agent

15 rules. About 0 are common, 6 are distinctive, and 9 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 9

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    One step at a time: Do not present six phases at once.

    line 43
  2. Distinctive:

    Do not treat this as paperwork to get past.

    line 76
  3. Very specific:

    Ask, in this order, and write each answer into config.yaml as you get it:

    line 82
  4. Distinctive:

    Every company has different functions.

    line 107
  5. Very specific:

    When you remove a folder, remove its rows from index.md, its routing-table rows in CLAUDE.md, its config.yaml block, and any skill that …

    line 124
  6. Very specific:

    every folder under accounts/ except _template/ and _sweep-state/, and empty accounts/_customer-list.csv down to its header and comment lines

    line 137
  7. Very specific:

    every per-person CSV in accounts/_sweep-state/, keeping the README

    line 139
  8. Very specific:

    every folder under partners/ except _template/, and empty partners/_partner-list.csv the same way

    line 140
  9. Very specific:

    every folder under outputs/ except _unassigned/

    line 145
  10. Very specific:

    Mobile: create a project and paste in the instruction block generated from references/mobile-project-template.md.

    line 223
  11. Very specific:

    Run every probe in references/verification.md.

    line 245
  12. Distinctive:

    Never mark a step done without its probe passing.

    line 253
  13. Very specific:

    Never invent a value into config.yaml.

    line 254
  14. Distinctive:

    Never grant access yourself or ask for credentials.

    line 256
  15. Distinctive:

    Do not delete anything outside this folder.

    line 258

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

Install this Team OS

You are guiding one person, the admin, through standing this up. They may be non-technical and may be working in a chat surface with no shell. Assume that.

How to run

Always read INSTALL-CHECKLIST.md first. It is the state. Never ask a question the checklist already answers, and never redo a step it records as done.

Then:

They sayYou do
"where do I start"Report the first todo step, in full, with what they physically have to do. Nothing else.
"what is left"List every todo and in progress step, grouped by phase, shortest first.
"check my install"Run the probes in references/verification.md against the folder and report what actually passed, not what is ticked.
anything elseWork out which phase it belongs to, do that step, then update the checklist.

One step at a time. Do not present six phases at once. An admin who is told everything is left does nothing; an admin told one thing does it.

Update INSTALL-CHECKLIST.md after every step, before you reply. If the checklist does not exist yet, create it from references/checklist-template.md as your first action.

Phase 1 - Get the files onto the Drive

Nothing else can happen until config.yaml exists in the Shared Drive. Create the Shared Drive, mount it with Drive for Desktop, and copy this repo's contents in from a terminal, not Finder: every behavioural part of the system lives in a dot-folder, and Finder hides those, so a drag-and-drop copies the documents and silently leaves the rulebook, the skills, the routines and the hooks behind.

Then chmod +x _ops/*.sh .claude/hooks/*.sh .codex/hooks/*.sh, because Drive does not preserve the executable bit. Confirm with ls -a that .claude, .agents, .codex and .claude-plugin all arrived.

Phase 2 - Notice, before any logging happens

This system records who opened it and when, and can publish a per-person summary of what each named colleague contributed. Before anyone's first session, the admin needs four things settled. Ask for each and record the answer in the checklist's Notes section:

  1. What is the lawful basis for recording this?
  2. Has everyone been told what is collected, who can read it, and how long it is kept? START-HERE.md has a section they can be pointed at.
  3. Does a works council or employee representative have to be consulted?
  4. Is the per-person table staying off? It is off by default in config.yaml adoption_reporting.per_person_table, and it should only go on after 2.

Do not treat this as paperwork to get past. The system only works if people use it willingly, and the quickest way to lose that is somebody discovering a productivity table with their name on it that nobody mentioned.

Phase 3 - Identity and access

Ask, in this order, and write each answer into config.yaml as you get it:

  1. Company name and email domain, into company_name and company_domain. Several generated files read these, so they come first.
  2. The Shared Drive name and id, into the drive block.
  3. The two Google Group addresses, into permissions.groups.
  4. Who runs the daily routine and who backs them up, into admin.

Then walk them through the grants, which they do in the Drive UI themselves:

  • Team group: Viewer on the drive, then Content Manager on accounts/, outputs/ and _ops/.
  • Admin group: Manager on the drive.

Explain why in one sentence: the team writes only where their work goes, and _ops/ exists as a folder rather than root files precisely so an ordinary team member's routine can close out.

Do not mark this phase done on their word. A grant that looks right in the sharing dialog and does not take effect is the most common install failure, and the symptom is "the assistant said it saved but nothing is there". Verify it per references/verification.md.

Phase 4 - Shape the folders to this company

Every company has different functions. Ask what this one actually has, and then delete what they do not need. A folder nobody fills is worse than no folder: it makes the assistant look for context that is never there.

FolderKeep?
accounts/, outputs/, _ops/Always. These are the three writable folders and the OS does not work without them.
company/Always.
product/If they build a product. A services business usually does not need it.
partners/If they have a supply, vendor or reseller side worth tracking per entity.
portfolio/If they want a cross-portfolio roll-up. Skip it and the confidentiality carve-out disappears with it, which is simpler.
integrations/knowledgebase.mdOnly if they have a second, read-only knowledge drive.

Also ask what they call things. If they say "clients" rather than "accounts", rename the folder and update index.md and the routing table to match. People do not adopt a system that uses someone else's vocabulary.

When you remove a folder, remove its rows from index.md, its routing-table rows in CLAUDE.md, its config.yaml block, and any skill that only served it. Then regenerate with ./_ops/sync-agents.sh and ./_ops/sync-skills.sh, or say that an admin with a terminal needs to run them.

Phase 5 - Remove the demo content, then seed the real thing

Delete the fiction first. Nothing else in this phase is safe until it is gone: an adopter who seeds their real customers alongside Acme's ends up with a roster nobody trusts, and a marketing skill that will happily clear Acme's claims for their publications.

5a. Delete the demo records

  • every folder under accounts/ except _template/ and _sweep-state/, and empty accounts/_customer-list.csv down to its header and comment lines
  • every per-person CSV in accounts/_sweep-state/, keeping the README
  • every folder under partners/ except _template/, and empty partners/_partner-list.csv the same way
  • the generated tables in accounts/_index.md and partners/_index.md
  • portfolio/crosswalk.csv, portfolio/revenue.csv, portfolio/revenue.meta.json
  • everything in product/prds/
  • every folder under outputs/ except _unassigned/
  • both tables in .claude/skills/acme-marketing/references/claims.md
  • the seed line in _ops/log.md, and the two demo tables in _ops/STATUS.md (## Sync status and ## Last update per account)
  • .github/, which is for this template's own repo and not for yours

5b. Replace the demo business

These describe Acme, not your company. Rewrite rather than delete, because the structure is what you want to keep:

  • company/overview.md, company/market.md, company/glossary.md, company/team.md
  • product/overview.md and product/roadmap.md, or delete product/ altogether if you do not build a product

5c. Replace every demo person

config.yaml names demo people in five places, not one. Change all of them:

  • people
  • admin.runner and admin.backup
  • customer_dashboard.schedule.runner and .backup
  • knowledgebase.refresh.runner and .backup
  • both skill_owners entries

Then the owner: field in the frontmatter of every file in .claude/skills/*/SKILL.md.

5d. Rename the two branded skills

acme-brand and acme-marketing become <company>-brand and <company>-marketing. Update the routing-table rows in CLAUDE.md, the --acme-* token names in assets/tokens/*.css, and every reference to acme-brand in other skills. Also update Shared drives/acme-os in .claude/routines/scheduled-task-prompts.md, and company_name / company_domain / repo_name in config.yaml if you have not already.

Re-run ./_ops/sync-agents.sh and ./_ops/sync-skills.sh afterwards.

5e. Check it

grep -ril acme accounts/ partners/ portfolio/ product/ company/ outputs/ \
  --exclude-dir=_template
grep -rn "sam-okafor\|maya-chen\|tobias-frank\|diego-ramirez\|priya-nair\|lena-vossen" \
  config.yaml .claude/skills/*/SKILL.md | grep -v install-team-os

Both must return nothing. The second filters out this file, which names the demo people in the line above and would otherwise match itself forever.

The first is scoped to the content folders rather than the whole repo, because after 5d the root documents and skills legitimately carry your company name, so a repo-wide grep can never come back empty and tests nothing.

Phase 5f - Seed the context

The OS is only as good as what it knows. Ask what already exists and map it in; do not ask them to retype anything.

  1. company/overview.md, market.md, glossary.md - from whatever they have: a pitch deck, a website, an internal wiki page.
  2. company/team.md and config.yaml people - the roster. role and owns are written by a human here; everything after this is derived automatically.
  3. accounts/_customer-list.csv - the account roster. This is the operative watch list, and nothing gets ingested for a company that is not on it. If they have a CRM export or a sheet, map the columns rather than retyping.
  4. product/ if kept.

Seed one account fully as the worked example, so the team can see the shape.

Phase 6 - Agent surfaces

  1. Drive for Desktop on each person's machine, signed in with their work account. Confirm the folder appears under Shared drives.
  2. Cowork: select this folder as the working folder. This is the primary surface for most of the team.
  3. Mobile: create a project and paste in the instruction block generated from references/mobile-project-template.md. It needs real file and folder ids, so generate it after Phase 3, when the files exist. See CONNECTOR-SURFACES.md.
  4. CLI (Claude Code, Codex): optional, for the admins and engineers.

Phase 7 - Automation

  1. Install the session hooks. .claude/settings.json is wired and takes effect for anyone who opens the folder in Claude Code. .codex/hooks.json is provided but repo-local hook support varies by Codex version, so confirm it actually fires before counting on it. Nothing to do for Cowork or mobile, which write session markers instead, and so do the CLIs.
  2. Register the daily admin routine as a scheduled task, using the prompt in .claude/routines/scheduled-task-prompts.md. It runs on one admin's machine and needs the local mount.
  3. Give everyone who talks to customers the single sentence that schedules their own ingest. It is in START-HERE.md and they paste it once.

Fill in config.yaml admin.runner and admin.backup, and make sure both are Drive Managers. A routine with one owner and no backup stops the week they go on holiday.

Phase 8 - Verify, then go live

Run every probe in references/verification.md. A step is done when a probe passes, not when someone says they did it. Report what actually passed.

Then send START-HERE.md to the team. Add them to the team group first, so their first session can write.

Non-negotiables

  • Never mark a step done without its probe passing.
  • Never invent a value into config.yaml. If you do not have it, leave the placeholder and leave the step todo.
  • Never grant access yourself or ask for credentials. Access changes happen in the Google Admin console, by a human, in their own browser.
  • Do not delete anything outside this folder.

Things that have gone wrong before

  • A Content Manager grant that has not been verified is not a grant. Why: the most common failure in this whole install is a sharing setting that reads correctly and has not propagated, and the person affected reports it as "the AI is broken" rather than as an access problem. (2026-09-16)
  • Remove the folders the company does not use, at install, not later. Why: an empty top-level folder makes the assistant search for context that does not exist, and every session pays for it. (2026-09-16)

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.