What are JTBD-style trajectory examples?

These are 'Jobs-to-be-Done' examples provided in the toolkit that show the agent exactly how to use available tools to achieve typical user goals within a specific mode.

What is a PostHog Agent Mode?

A mode is a specific configuration that limits the tools, prompts, and prompt injections available to the PostHog AI assistant based on the user's current task or product area.

Why should I use specialized modes instead of a general agent?

Specialized modes improve agent accuracy and reduce latency by narrowing the context window and providing only the tools necessary for a specific job, such as SQL analysis or error tracking.

How does the agent switch between different modes?

The agent uses a built-in 'switch_mode' tool that allows it to transition to a more relevant mode while preserving the current conversation context.

Are new AI modes safe to release to production?

Yes, this skill requires all new modes and tools to be implemented behind PostHog feature flags, allowing for controlled rollouts and testing.

PostHog Agent Mode Architect

Name: PostHog Agent Mode Architect
Author: PostHog

byPostHog

•

30,957

•

分析与监控

Guides developers through the end-to-end process of creating and configuring specialized context-aware modes for the PostHog AI assistant.

This skill provides a structured framework for extending PostHog's AI capabilities by implementing specialized 'modes' that scope tools, prompts, and behaviors to specific product areas. It streamlines the complex workflow of defining AgentToolkits, updating schemas, and managing backend-first tool integrations. By following these patterns, developers can create highly focused AI assistants for specific Jobs-to-be-Done (JTBD), ensuring the agent always has the relevant context and permissions for tasks like product analytics, SQL querying, or error tracking while maintaining safe deployment via feature flags.

主要功能

01Automated schema regeneration and scaffolding workflows

02Backend-first tool integration patterns for better AI context

03JTBD-style trajectory example implementation for agent training

04Standardized feature flagging for safe mode deployment

0530,957 GitHub stars

06Contextual tool management and scoping by product area

使用场景

01Developing a specialized AI agent mode for a new PostHog feature like Web Analytics

02Refactoring frontend-only tools into backend-first implementations to improve AI reliability

03Optimizing agent performance by restricting available tools to relevant domains

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add posthog/posthog implementing-agent-modes

For use in Claude.ai and ChatGPT

Download Skill