More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Agent Team for Claude Code

    1

    Bootstrap a 17-role software company of Claude Code subagents into any project, grounded in that project's real code.

    $19

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Agent Team for Claude Code

    Agent Team for Claude Code

    Example session with this skill installed

    /agent-team init --profile balanced --roles full --lang Thai

    • Read your context and instructions
    • Compiled the agent team

    Explored 2 repos (FastAPI backend, Expo mobile app), verified 4 check commands, then wrote:

    .claude/agents/ 17 roles (opus/high for architect, security-engineer, code-reviewer, backend-dev, e2e-tester; sonnet/high for the rest)
    .claude/commands/ 10 pipelines (/feature, /fix, /test-all, /prd, /adr, /security-audit, /release, /review, /standup, /team)
    .claude/agent-team.json config with per-repo stacks and verified check commands
    .claude/agent-memory/ per-role memory, empty
    docs/PROJECT-CONTEXT.md filled: product paragraph, repo table, architecture map, isolated test recipe, port allocation
    docs/TEAM.md org chart
    docs/LEARNINGS.md seeded with 3 bootstrap lessons
    docs/product/BACKLOG.md empty scaffold

    Hygiene red flags found: .env tracked in git (backend), node_modules vendored (mobile)
    CHECK: 27 files, frontmatter OK
    Agent types load on the next session start. Suggested first commands: /test-all then /standup

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

    About this skill

    What it does

    /agent-team init explores your actual repository — git remote, manifests, how to typecheck/lint/test/build, infra, environments, domain model, existing conventions — then asks four questions (profile, role set, report language, docs location) and generates:

    • 17 role subagents in .claude/agents/ — product-manager, architect, ux-designer, ui-designer, backend-dev, mobile-dev, web-dev, devops-engineer, test-engineer, api-tester, e2e-tester, mobile-tester, web-tester, code-reviewer, bug-triager, security-engineer, tech-writer
    • 10 pipeline commands in .claude/commands/ — /feature, /fix, /test-all, /prd, /adr, /security-audit, /release, /review, /standup, /team
    • A shared docs folder — PROJECT-CONTEXT.md, TEAM.md, LEARNINGS.md, product/BACKLOG.md
    • Per-role persistent memory in .claude/agent-memory/

    Why it is different

    The quality comes from one rich, verified context file that every agent reads first, not from long agent prompts. During init the skill runs your project's cheap checks to confirm they actually work, and records env var names, port allocations, the isolated test recipe and hygiene red flags. Agents stop guessing field names and routes.

    Every agent follows the same expert protocol: load context → plan with alternatives → verify instead of guessing → self-review as its harshest critic → leave a lesson behind in its memory file.

    Safety rails baked into every role

    • Testers run only against an isolated instance they start themselves, never dev or production data
    • No git add/commit/push/stash/checkout/reset unless you ask in that turn
    • devops-engineer prepares deploys but never deploys — it writes you a runbook
    • Never writes secrets into generated files; env var names only
    • Read-only roles (code-reviewer, security-engineer, bug-triager) cannot edit code

    Profiles

    profileheavy rolesother roles
    maxfable / maxfable / xhigh
    balancedopus / highsonnet / high
    economysonnet / mediumsonnet or haiku / medium

    Heavy roles (architect, security-engineer, code-reviewer, backend-dev, e2e-tester) always get the top effort tier. Per-role model and effort overrides go in .claude/agent-team.json.

    Reports in your language

    Set report_language once at init. Reports and documents come back in that language; code identifiers, paths, commands and HTTP details stay in English. Tested with Thai and English.

    Not a one-shot generator

    /agent-team status shows drift (agent memory sizes, context file age vs newest commit). /agent-team add <role> adds a role later. /agent-team upgrade all re-renders agents and commands from new template versions without ever touching your hand-filled docs — a .bak is kept for anything overwritten.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.

    Generates agents and docs; it does not write your product code. The init pass asks four questions and needs a few minutes to explore a large repo. PROJECT-CONTEXT.md ships with TODO sections that init fills from verified facts — if you skip that step the team is noticeably dumber. Subagents cannot spawn subagents, so the coordinating roles are slash commands run by the main session, not agents. Mobile and web testers do static and contract QA (typecheck, navigation integrity, i18n parity, API contract), not device-farm or browser-driver testing. Optional skills (impeccable, scrutinize, debug-mantra, post-mortem) are referenced only when already installed on the machine; otherwise those references degrade to no-ops.

    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

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 15 days ago

    Software
    Claude CodePython 3.8+ (stdlib only)GitimpeccableNode.js

    Permissions

    File Scopes

    project
    /.claude/agents/
    /.claude/commands/
    /.claude/agent-team.json
    /.claude/agent-memory/
    docs_dir
    /
    (default
    /docs/)

    init reads your repository to build the context file: git metadata, manifests, source files, CI and infra config. It records environment variable NAMES only, never values, and is instructed never to write secrets into any generated file. It runs your project's own cheap check commands (typecheck/lint/test) once to confirm they work. No network host is contacted by the skill itself; the WebSearch/WebFetch tools are granted to the generated architect, PM, security and writer roles for vendor docs and CVE lookups when you run them.

    Listed15 days ago

    What's inside

    Frequently Asked Questions

    No. It generates the agent team, the pipeline commands and the shared docs. The generated dev agents then write code when you run /feature or /fix — but the skill itself only sets up the team.

    No. Every role is told never to run git add/commit/push/stash/checkout/reset unless you ask in that turn, and the devops-engineer explicitly never deploys — it writes you an ordered runbook and you press the button.

    No. Testers only run against an isolated instance they start themselves using the test recipe recorded in PROJECT-CONTEXT.md, on their own allocated port, and they tear it down before finishing.

    /agent-team upgrade all re-renders only agents and commands. It never re-renders the four hand-filled docs, and anything it does overwrite gets a .bak beside it. Resetting the docs scaffolds requires an explicit --scope docs.

    No. Pick --roles core for product-manager, architect, one dev per detected stack, test-engineer, code-reviewer and bug-triager. Roles that do not apply are skipped automatically: no mobile stack means no mobile-dev or mobile-tester.

    Yes. The skill is told to reuse what exists — link and extend rather than duplicate — and render.py never overwrites an existing file without --force.

    Whatever your Claude Code account has. The profile picks the tier: max uses fable, balanced uses opus for heavy roles and sonnet for the rest, economy stays on sonnet and haiku. Any role can be overridden in .claude/agent-team.json.