Run Governed AI Workflows from Claude Code — Plugin v0.2.0

You are in Claude Code debugging a deployment. A container keeps crashing, you are three stack traces deep, and you suddenly remember: this week’s content plan was supposed to go out an hour ago. Normally you would context-switch to the JieGou console, find the workflow, fill in the parameters, and hit run. Not anymore.

With the JieGou Claude Code Plugin v0.2.0, you type: “Run the weekly content plan workflow.” Claude Code calls JieGou’s API, the workflow executes with full governance, and you get the status back in your terminal. You never leave your editor.

What Ships in v0.2.0

The plugin exposes 11 API tools and 6 skills that cover the core surface area of the JieGou platform.

The 11 Tools

These are standalone API tools that work without a WebSocket connection. Claude Code calls them directly over HTTPS using your API key:

• list-recipes — Browse your account’s recipe catalog with optional filters
• run-recipe — Execute a recipe with parameters and get results back
• recipe-status — Check the execution status of a running or completed recipe
• list-workflows — Browse available workflows across departments
• run-workflow — Trigger a multi-step workflow with input parameters
• workflow-status — Poll or check the current state of a workflow run (including approval gates)
• search-knowledge-base — Query your organization’s knowledge bases using RAG search
• get-analytics — Pull execution analytics: success rates, token usage, cost breakdowns
• create-schedule — Set up recurring execution schedules for recipes or workflows
• publish-social — Publish content to connected social channels (LINE, Instagram, Facebook, and more)
• get-account-info — Retrieve account metadata, quotas, and feature flags

The 6 Skills

Skills are higher-level capabilities that compose multiple tools into natural conversation patterns:

1. Status Check — “How are my workflows doing?” summarizes recent executions, failures, and pending approvals
2. Dispatch Workflows — “Run the weekly sales report” triggers a workflow with smart parameter inference
3. Browse Recipes — “Show me marketing recipes” lists and describes available recipes with usage stats
4. Run Recipes — “Summarize this document using the executive brief recipe” executes with inline results
5. Manage Workflows — “Pause the onboarding workflow” or “Approve the pending content review” handles lifecycle operations
6. Trigger Automations — “Schedule the competitor analysis to run every Monday at 9 AM” creates and manages recurring schedules

Why Governance Matters Here

This is the part that separates the plugin from a raw API wrapper.

Every execution triggered through the Claude Code plugin passes through the same governance stack as executions from the JieGou console. That means:

Audit trails. Every recipe run, workflow trigger, and knowledge base query is logged with the requesting user, timestamp, parameters, token usage, and cost. Your compliance team can see exactly what was executed, by whom, and when — even if it was triggered from a terminal.

RBAC enforcement. The plugin authenticates using your JieGou API key, which is scoped to your user role. If you are an Editor, you can run recipes but not delete workflows. If you are a Viewer, you can check status but not trigger executions. The same 5-role permission model (Owner, Admin, Manager, Editor, Viewer) applies regardless of the client.

Approval gates. When a workflow hits an approval step, it pauses — even when triggered from Claude Code. The plugin’s workflow-status tool reports the pending approval, and the designated approver can act from the console or the plugin itself. No approvals are bypassed just because the execution started from a developer’s terminal.

Token budgets and rate limits. Department-level token budgets and per-user rate limits apply. The plugin cannot circumvent cost controls that your account administrator has configured.

Compare this to calling an LLM API directly from a script. You get the output, but no audit log, no role check, no approval gate, no cost tracking. The plugin adds the governance layer that turns ad-hoc API calls into accountable operations.

BYOK: Your Keys, Your Models

The plugin uses the Bring Your Own Key model. It authenticates against your JieGou account using your API key, and all LLM calls use the model keys your organization has configured. No API keys are stored or proxied through additional intermediaries. If your team uses Claude on Anthropic’s API, that is the model your recipes execute against. If you have switched a department to GPT-4o for cost reasons, that configuration is respected.

Context: Anthropic Partnership

JieGou is a member of the Anthropic Claude Partner Network. The Claude Code plugin has been submitted to both the Anthropic marketplace and the Claude Marketplace (the enterprise app store). In an ecosystem of 834 plugins, the JieGou plugin is one of the most feature-rich — most plugins expose 2-4 tools, while ours ships 11 tools and 6 skills covering execution, analytics, scheduling, and content publishing.

The plugin works standalone. No WebSocket connection is needed — all 11 tools operate via direct API calls. This makes it lightweight to install and reliable in CI/CD environments, remote containers, and SSH sessions where persistent connections are impractical.

What This Looks Like in Practice

A developer working in Claude Code might have a session like this:

“Show me the recipes in the marketing department.”

Claude Code calls list-recipes and returns a formatted list with descriptions and last-run timestamps.

“Run the competitor analysis recipe for Q1 2026.”

Claude Code calls run-recipe, passes the parameters, and streams back the result — a structured analysis with sections, data points, and recommended actions.

“Schedule that to run every Monday at 9 AM and publish the summary to our LINE channel.”

Claude Code calls create-schedule to set up the recurrence, then notes that the publish-social tool can handle the distribution step.

All of this happens within the governed context of the developer’s JieGou account. Every action is logged, every permission is checked, every token is counted.

Getting Started

Install the plugin from the Claude Code plugin registry or the Anthropic marketplace. You will need a JieGou account and an API key (generated from Settings > API Keys in the console). The plugin auto-discovers your available recipes, workflows, and knowledge bases on first connection.

For teams already using JieGou, this is a natural extension of your existing setup. For developers evaluating the platform, the plugin offers a terminal-native way to explore what governed AI execution feels like — without leaving the environment where you already work.

The plugin is open for feedback. If you hit rough edges or want additional tools, file an issue or reach out on our Discord. v0.3.0 is already in planning.