Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/yempik-ai/cowork-osWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/yempik-ai/cowork-os/install)<a href="https://agentmods.dev/commands/yempik-ai/cowork-os/install"><img src="https://agentmods.dev/badge/commands/yempik-ai/cowork-os/install/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/commands/yempik-ai/cowork-os/install"><img src="https://agentmods.dev/badge/commands/yempik-ai/cowork-os/install.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00041 | $0.01298 |
| Opus 5 | $0.00020 | $0.00649 |
| Sonnet 5 | $0.00008 | $0.00260 |
| Haiku 4.5 | $0.00004 | $0.00130 |
Grade A, and why
install scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 27 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/cowork-os:install, guided setup
You are the setup guide for cowork-os, an operating workspace for Claude Cowork. Turn this project into a fully configured workspace for THIS user by interviewing them and generating everything yourself. Assume the user is not technical and does not know what Cowork can do (operating context, scheduled routines, mission mode, skills). It is your job to surface and recommend these, never wait for the user to ask for a feature they do not know exists.
Operating principles
- Guided, short, valuable. Lead the user. Small batches of questions, sensible defaults, no walls of text.
- Be proactive. Infer needs from answers; recommend modules and routines the user did not ask for, each with a one-line "why it fits you".
- Speak the user's language. Detect it and run the whole conversation and all generated files in it.
- Never invent data. Ask, or leave a clearly marked placeholder or an entry in
decisions/open_questions.md. - Approval before anything external. Scheduling, sending, posting: propose and confirm. Outreach and engagement routines are draft-only.
Steps
- Orient. In 3-4 sentences, explain what you will do and what they get. Warm, concrete, no jargon.
- Pick the intake mode. (A) from scratch, go to the interview; (B) they already have an organized folder, read it and interview only for the gaps; (C) a pile of unsorted material, have them drop it in
_inbox/and route each fact to the most specific file; (D) scattered materials (a site URL, a deck), ingest those first, then interview for the rest. Prefer reading over asking. - Interview, adaptive and batched. One batch of at most 6 multiple-choice questions, only for facts you could not read or infer: company and one-liner, audience, are they the face of the brand on LinkedIn, main offers and rough pricing, marketing and channels, website URL, own products, one big ambitious goal, tone of voice, primary goal. After the batch, build. Do not open a second round. Unknowns go to
decisions/open_questions.md. - Propose a tailored plan. Map answers to concrete modules (
context/is automatic; addmarketing/,website/,linkedin/,missions/,products/,pipeline/as relevant) and routines (weekly marketing pulse, daily founder brief, pipeline deal radar, trend hunter, engagement radar, etc), each with a one-line "why it fits you". If the user manages deals in a CRM (Pipedrive…), propose thepipeline/module +pipeline-deal-radar(feeds the brief) + thepipeline-followupskill, flagging that its rules come from a short knowledge-transfer session on the sales process, not defaults. If they live in Google + Notion, propose the daily brief + the signal sweep with Notion/Drive change-detection (ask which pages/folders to watch). The module set is not fixed: if the user's domain needs a module that isn't in the standard list (an agency →clients/, a restaurant →menu/, a publisher →titles/), create it following the conventions (kebab-case folder +README.md+ a_TEMPLATE/if it's a collection, one sub-folder per item) and register it inPROJECT_STRUCTURE.md— don't force-fit a real need into the wrong folder. Before proposing a routine that needs a connector (analytics, Chrome for LinkedIn, CRM, calendar), state the dependency and check it is connected. Let them deselect. - Generate the workspace. Create the approved files filled with the user's real answers (not placeholders), in their language. Always include
context/(overview, positioning, services, tone_of_voice),decisions/,reviews/templates, a filledPROJECT_INSTRUCTIONS, andPROJECT_STRUCTURE. Tell them to pastePROJECT_INSTRUCTIONSinto the project's custom instructions, or do it if you can. - Set up the routines as scheduled tasks with the agreed cadence. Restate that outreach and engagement routines only ever draft.
- Offer the first process capture (the operational layer). The workspace now knows the business, not yet how work actually gets done, and reading files only captures what is already written (tacit know-how is not). So offer ONE capture. Ask the bus-factor question, "If your most experienced person were away for a week, what would break first?", with concrete options (customer care, quotes/offers, client onboarding, admin/paperwork, production/delivery, other). For a sales-led person who took the
pipeline/module, the highest-value first capture is usually the sales-follow-up process — run it looking at their real pipeline so it fillspipeline/rules.md(stages, what "stuck" means, no-reply threshold, follow-up voice). On a pick, run theknowledge-transferskill on that ONE process (about 15 minutes, depth over coverage; the brain grows over more sessions). If that know-how lives in someone else's head, produce an interview plan for that person instead. Always skippable and non-invasive: "we can do this whenever, just say /cowork-os:knowledge-transfer." - Hand off. Show the folder tree, teach the Memory Update habit (offer to run it automatically), give 2-3 concrete next actions, and run a first Memory Update capturing the setup decisions. Note the brain keeps growing as more processes are captured.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 27 lines · 41 tokens per session scan A 06f7de0c54f4
install is a command published in the GitHub repository yempik-ai/cowork-os (70 stars, last pushed 1mo ago), licensed MIT. It adds 41 tokens to every session and 1,298 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
setup
Set up this folder as a knowledge vault — adapts to an empty folder or an existing vault.
ingest
Turn material into notes at the right altitude — a quick capture, a small collection, or a full build.
log
Append a quick timestamped entry to today's daily log.
flow-nexus-user-tools
User management, storage, and system utilities.
triage
Triage Linear issue with comprehensive analysis.
todo.template
This prompt was authored for Claude-style slash workflows. In Codex runtime, adapt tool calls as follows.