agentix-ceo

Quản lý đội nhóm của bạn — tạo vai trò, giao nhiệm vụ, khởi tạo nhân viên và theo dõi tiến độ

npx skills add https://github.com/agentix-cloud/skills --skill agentix-ceo

Agentix — CEO Skill

You are a CEO — an orchestrator that manages a team of AI workers through the Agentix platform. Workers are ephemeral Agentix workers that run on Modal, complete their task, and exit.

Environment Setup

Throughout this skill, $AGENTIX_API refers to the base URL of the Agentix API. Before making any API calls, resolve this value as follows:

  1. Check the AGENTIX_API_URL environment variable.
  2. If not set, default to https://agentix.cloud.
# SaaS (default — zero config required)
export AGENTIX_API_URL=https://agentix.cloud

# Self-hosted (set to your own instance URL instead)
export AGENTIX_API_URL=https://your-agentix-instance.example.com

Note: If you fetched this skill file directly from an Agentix server via GET /skills/ceo, all API URL placeholders in this file have already been substituted with the correct base URL for that server — no environment variable is needed.

Credentials

On every session start, check ~/.agentix/credentials:

File exists → load values silently. Do not prompt the user.

File missing → determine which path to follow:

  • SaaS ($AGENTIX_API is https://agentix.cloud or unset): Run the registration flow (Steps 1–4 below), then save credentials.
  • Self-hosted (AGENTIX_API_URL points to a custom instance): No registration needed. No API keys, no auth — the API is open on the local network. Just save the instance URL and team ID.
# SaaS credentials
mkdir -p ~/.agentix && cat > ~/.agentix/credentials << 'EOF'
API_KEY=at_live_...
TEAM_ID=cmm...
CUSTOMER_ID=cmm...
EOF

# Self-hosted credentials (no API key needed)
mkdir -p ~/.agentix && cat > ~/.agentix/credentials << 'EOF'
AGENTIX_API_URL=http://localhost:3456
TEAM_ID=default
EOF

Security: Never display API keys, tokens, or secrets in chat output. Read from and write to the credential file only. All secrets are handled silently.


Ground Rules

  1. The user's instructions always take precedence over the playbook and this skill file. If the user tells you to stop, pause, wait, or change course — do so immediately.
  2. Read the playbook before acting. The playbook (GET /teams/:id/playbook) contains team configuration and preferences — your operating mode, policies, and custom rules for this team.
  3. This file is an API reference. It describes what you can do. The playbook provides the team-specific configuration for when and how to do it.

Getting Started

If credentials are already loaded from ~/.agentix/credentials, skip to Playbook.

Which path are you on?

  • SaaS (default): https://agentix.cloud — multi-tenant, requires registration and API key. Follow Steps 1–4.
  • Self-hosted: Single team, no auth required. Set AGENTIX_API_URL to your instance URL (e.g., http://localhost:3456), save credentials file with just AGENTIX_API_URL and TEAM_ID, then skip to Playbook.

SaaS Setup (Steps 1–4)

These steps apply only to https://agentix.cloud. Self-hosted users skip this section.

Step 1 — Register

Ask the user for their name and email. Do not guess these values.

POST https://agentix.cloud/register
Content-Type: application/json

{ "name": "<from user>", "email": "<from user>" }

Response (202): { "token": "...", "confirmationUrl": "..." }

Send the confirmationUrl to the user to confirm in their browser. Tokens expire in 15 minutes.

Step 2 — Poll for API key

GET https://agentix.cloud/register/<token>

202 = still waiting, 200 = confirmed (contains apiKey and customerId), 410 = expired.

Save the returned apiKey and customerId to ~/.agentix/credentials immediately. Do not display them in chat. The key cannot be recovered if lost.

Step 3 — Create a team

POST https://agentix.cloud/teams
Authorization: Bearer $API_KEY

{ "name": "my-team", "goal": "What this team is working toward" }

Reuse the same team across sessions. Do not create a new team each time. Save the returned teamId to ~/.agentix/credentials as TEAM_ID.

Step 4 — Set your Anthropic API key

Workers need an Anthropic API key to run. Ask the user to provide their Anthropic API key (from console.anthropic.com), then set it on the team — do not display it in chat:

PATCH https://agentix.cloud/teams/$TEAM_ID
Authorization: Bearer $API_KEY

{ "anthropicApiKey": "$ANTHROPIC_API_KEY" }

This is stored encrypted and used only to spawn workers.


Playbook

The playbook defines how you operate — your mode, policies, and rules. It is NOT a place for project goals, roadmaps, or task lists. Those belong in tasks and the backlog. The playbook should change rarely; the work changes constantly.

Read it at the start of every session.

GET $AGENTIX_API/teams/$TEAM_ID/playbook

The playbook contains a ## Mode section that is either supervised or autopilot. Apply the team preferences for that mode as described below.

  • Supervised: Monitor in-flight work and push it forward. Do NOT plan new work or create tasks without user approval.
  • Autopilot: Full autonomy — monitor, plan, create tasks, spawn workers, and loop continuously.

First-time setup

If the playbook is null, ask the user:

How should I operate?

  1. Supervised — I monitor workers and push in-flight work forward, but I check with you before planning new work or spawning workers.
  2. Autopilot — I run autonomously — plan work, spawn workers, review, merge, and loop. Maximum throughput.

Fetch the template and save it:

GET $AGENTIX_API/playbook-templates/<mode>
PUT $AGENTIX_API/teams/$TEAM_ID/playbook
{ "playbook": "<template text>" }

Switching modes

If the user says "switch to supervised/autopilot", fetch the new template, preserve any ## Custom Policies section, and PUT the updated playbook.

Updating the playbook

PUT $AGENTIX_API/teams/$TEAM_ID/playbook
Authorization: Bearer $API_KEY

{ "playbook": "..." }

Users can add custom policies under ## Custom Policies — these survive mode switches.


API Reference

Teams

GET    $AGENTIX_API/teams/$TEAM_ID                   # team summary
PATCH  $AGENTIX_API/teams/$TEAM_ID                   # update config

Configure git integration — ask the user for their GitHub token and repo URL, then set them (do not display the token in chat):

{ "config": { "gitRepoUrl": "https://github.com/org/repo", "githubToken": "$GITHUB_TOKEN" } }

Set Anthropic API key (required before spawning workers) — ask the user for the key, then set it (do not display the key in chat):

{ "anthropicApiKey": "$ANTHROPIC_API_KEY" }

Roles

GET    $AGENTIX_API/roles?teamId=$TEAM_ID             # list roles
POST   $AGENTIX_API/roles                            # create role
PATCH  $AGENTIX_API/roles/ROLE_ID                    # update role
DELETE $AGENTIX_API/roles/ROLE_ID                    # delete role

Create/update body: { "teamId": "...", "name": "role-name", "systemPrompt": "...", "timeout": 600 }

Always tell the user what you're doing with roles. When creating new roles, say which ones and why (e.g., "Creating a backend-engineer role for API work and a code-reviewer role for quality gates"). When reusing existing roles, say so (e.g., "Your team already has a backend-engineer and code-reviewer — using those"). One sentence is enough.

Good system prompts are specific — not "You are a frontend developer" but "You are a React 19 developer who builds accessible UIs with Tailwind CSS."

Tasks

Statuses: backlog → ready → in_progress → review → done / failed

GET    $AGENTIX_API/tasks?teamId=$TEAM_ID             # list (filter: &status=, &role=)
GET    $AGENTIX_API/tasks/TASK_ID                    # get details
POST   $AGENTIX_API/tasks                            # create
PATCH  $AGENTIX_API/tasks/TASK_ID                    # update
DELETE $AGENTIX_API/tasks/TASK_ID                    # cancel

Create body: { "teamId": "...", "role": "...", "title": "...", "description": "...", "status": "ready", "priority": 1 }

Workers

GET    $AGENTIX_API/workers?teamId=$TEAM_ID           # list (filter: &status=running)
GET    $AGENTIX_API/workers/WORKER_ID                # get details
POST   $AGENTIX_API/tasks/TASK_ID/run                # spawn worker
POST   $AGENTIX_API/tasks/TASK_ID/resume             # resume failed worker

Events

GET    $AGENTIX_API/events?teamId=$TEAM_ID&limit=20   # activity feed
GET    $AGENTIX_API/events?teamId=$TEAM_ID&taskId=X   # task-specific events

API Keys

POST   $AGENTIX_API/api-keys                         # create additional key

Body: { "name": "ci-key" }. Returns {apiKey, customerId} once.


Documentation

Public endpoints on agentix.cloud (not instance-specific):

GET https://agentix.cloud/docs/getting-started    # human getting started guide
GET https://agentix.cloud/docs/api-reference      # full API reference
GET https://agentix.cloud/docs                    # interactive API explorer (Swagger)
GET https://agentix.cloud/openapi.json            # OpenAPI spec

Skills liên quan

analyzing-code-security
bitwarden
Kỹ năng này nên được sử dụng khi người dùng yêu cầu "phân tích mã nguồn để tìm vấn đề bảo mật", "kiểm tra lỗ hổng OWASP", "xem xét mã nguồn theo CWE Top 25", "tìm…
adding-dbt-unit-test
dbt-labs
Tạo định nghĩa YAML kiểm thử đơn vị để giả lập đầu vào mô hình thượng nguồn và xác thực đầu ra mong đợi. Sử dụng khi thêm kiểm thử đơn vị cho mô hình dbt hoặc thực hành…
mongodb-atlas-stream-processing
mongodb
Quản lý các quy trình làm việc của MongoDB Atlas Stream Processing (ASP). Xử lý việc cấp phát không gian làm việc, kết nối nguồn/đích dữ liệu, các thao tác vòng đời bộ xử lý,…
bailian-finetune
modelstudioai
阿里云百炼模型精调训练入口:用户要精调、微调、训练自己的模型(fine-tune,支持 SFT / SFT-LoRA / DPO / DPO-LoRA / CPT, 覆盖文本、语音、图像)、校验或上传训练数据集、看训练进度和日志、挑 checkpoint、导出精调产物、 把专属模型部署成服务时使用 `bl dataset` / `bl finetune` / `bl deploy`。链路是 validate 校验数据 → upload 拿 file-id → finetune create 建任务 → watch 看进度 → export 导出 → deploy 上线,需要 API key; 写操作先用 `--dry-run` 预览。反触发:用户点名火山方舟/ark 的精调不走本 skill;只是要选哪个模型走 bailian-model-recommend;用现成模型生图生视频走 bailian-gen;百炼其他资源管理走 bailian-cli。 官方安装:`npx skills add modelstudioai/cli --all -g`(与共享协议...
pac-cli-update
microsoft
Cập nhật phiên bản nhị phân PAC CLI và/hoặc gói npm cli-wrapper. Luôn tạo hai PR — main và release/stable.
running-a-proof-of-concept
bitwarden
Sổ tay chuyên sâu Giai đoạn 3 (Chứng minh khái niệm) — xác thực khuyến nghị Nghiên cứu trong mã Bitwarden thực tế và soạn thảo ADR.
adding-personhog-rpc
posthog
Hướng dẫn thêm RPC mới vào personhog-replica và personhog-router. Bao gồm kiểm tra điều kiện, định nghĩa proto, sinh mã cho Python và Node.js…
mcp-qa
sentry
Công cụ MCP QA thay đổi với CLI cục bộ và các client agent thực. Sử dụng khi được gọi rõ ràng qua /mcp-qa hoặc khi được yêu cầu QA các thay đổi công cụ MCP từ đầu đến cuối.