Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Claude Inc

    1

    Roles that judge someone else's work (QA, code and security reviewers, a skeptic, a safety officer, a fact-checker) run as separate subagents, so the work never grades itself.

    Free

    5 usesSecurity scanned
    Claude Inc

    Claude Inc

    Example session with this skill installed

    /team add a "Download CSV" button to the tasks page

    • Read your context and instructions
    • Compiled the claude inc
    • Generated the app files

    Summary

    Sized M (one feature, two tasks, no auth or data-model change). Ran PM -> tech lead -> one combined gate -> build -> verify -> docs. 5/5 acceptance criteria pass; nothing pushed.

    How the run went

    1. Product manager -> .team/SPEC.md: 5 numbered acceptance criteria. Open question: include completed tasks? Default: follow the current filter.
    2. Tech lead -> .team/PLAN.md: T1 export_tasks(filters) in app/export.py, T2 route GET tasks/export.csv plus the button. Each task owns its files, so they build in parallel.
    3. Gate: you approve spec and plan in one click and keep the default for the open question.
    4. Engineers: T1 and T2 built side by side, logged in .team/LOG.md. The button uses a relative URL because your project rules say the app runs under a path prefix.
    5. QA engineer (independent subagent) -> .team/QA.md: 4/5 pass. Bug Q1: a title with a comma and a quote breaks the row, at app/export.py:20 (a hand-built join). A failing test reproduces it.
    6. Fix round: T1 switches to the csv module; the test passes. QA re-runs: 5/5 pass.
    7. Senior reviewer (read-only) -> .team/REVIEW.md: no blockers, one nit (stream large exports later).
    8. Tech writer: README gains an "Exporting tasks" section.

    Report

    ItemResult
    Acceptance criteria5/5 pass
    Bugs found / fixed1 / 1
    ReviewNo blockers, 1 nit
    Memo.company/ memo written, ledger updated
    Next stepCommit and push wait for your go-ahead

    export.csv

    ZIP · project files

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    A single agent session does whatever you ask, in the order you ask it. That works for small tasks and breaks down for big ones: it changes servers before anyone knows what's broken, designs an enclosure before checking the physics, writes code before anyone agreed what "done" means, and grades its own work.

    Claude Inc puts the right roles in the right order, with gates between phases. Each department is a skill with defined roles. Each role has one job and a written output, and the next role works from that output instead of from chat history. Roles that judge someone else's work (QA, code and security reviewers, a skeptic, a safety officer, a fact-checker) run as separate subagents, so the work never grades itself.

    The departments

    CommandUse it whenYou get
    /companyYou don't want to pickRoutes to the right department and says why
    /war-roomSomething is broken right nowSeverity, ranked hypotheses, smallest fix with rollback, postmortem
    /skunkworks"Can this physical build even work?"First-principles math, an independent skeptic, the cheapest test, a safety review with veto power
    /studioDesigning a game or mod systemDesign doc, a real balance simulation, a paper playtest before any code
    /board-meetingToo many projectsEvery project scored, challenged, and Focus / Maintain / Pause / Kill resolutions for you to approve
    /research-desk"Is this worth doing?"A brief where every claim has a source and a fact-checker checked it
    /ops-centerWeekly server checkUptime, TLS expiry, disk and RAM trends, days until full, follow-ups
    /teamBuild, fix or ship softwarePM, tech lead, engineers in parallel, UI designer, QA, senior and security review, docs, release

    How it works

    • One rule book. Every department follows the same protocol: verdicts (PASS / REVISE / STOP), at most two revise loops before it asks you, and one memo format, so any department can read another's output.
    • One project ledger. Each project has an entry with its stage, open decisions and memo history. /board-meeting reads it to tell you where your time should go, and /company reads it to route new requests.
    • Private config. ~/.claude/company/config.md holds your hosts, house rules and deploy recipe. Departments never copy its values into code, commits or committed memos, and secrets are never stored in it, only the name of the env var that holds them.
    • Gates, not surprises. Restarts, deploys, deletes, DNS changes and purchases each need your approval, even mid-outage, and every proposed change comes with a rollback.

    Where it stops

    It doesn't replace your monitoring (/ops-center is a weekly check, not an alerting system), and it won't act on servers that aren't in your config. /skunkworks checks feasibility; it doesn't design the finished build.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.

    • Works best in Claude Code. Agents without subagents run reviews inline, which the skill tells you, so those reviews aren't independent.
    • /ops-center covers HTTP, TLS, disk, RAM and Proxmox. Anything else needs a read-only command in your config's extra_checks.
    • /team deploys only with a recipe you configure (Docker over SSH, Fly.io, static rsync, or your own steps).
    • Departments propose; you approve. Nothing restarts, deploys or deletes on its own.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    5 uses

    Times used to date

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 6 days ago

    • Free with an Agensi account

    Software
    Git

    Permissions

    Terminal / Shell
    Browser
    Network Access
    Read Files
    Write Files
    Environment Variables
    Listed6 days ago

    What's inside

    Frequently Asked Questions

    No. The first department you run creates ~/.claude/company/ with a config template. Fill in only the sections you use: Homelab for /ops-center, Deploy for /team, and so on.

    Only with read-only commands, unless you approve a specific change, with its rollback, in a question dialog.

    Yes. Every department works directly (/war-room ...) without going through /company.

    It stays on your machine, in ~/.claude/company/ and your project folders. Memos in projects are kept out of git by default.

    It's built for Claude Code, which has subagents and question dialogs. Other agents that can run subagents work too. Without subagents, reviews run inline and the skill tells you they weren't independent.

    New in Agents & Orchestration

    Free