Analyze a PostHog Funnel With Codex and Soku CLI

Give Codex allowlisted, read-only PostHog access through Soku CLI, inspect the live project schema, measure a conversion funnel, and produce a drop-off diagnosis without exposing provider keys.

Best for

Product-led growth teams that want Codex to analyze activation data while keeping PostHog credentials server-side and the command surface read-only.

Access mode

Read-only analysis

No connected account changes.

The prompt

Read https://soku.ai/cli/skill.md, connect Codex to my Soku workspace, and perform a read-only PostHog funnel analysis for signup to activation over the last 30 days. List projects, inspect the allowlisted tools and live data schema before querying, then quantify each funnel step and the most relevant segments. Use only event names found in the live schema. Safety boundary: do not change connected accounts or publish content. Use read commands only, and return proposed actions for a human to review separately.

Paste this into Claude Code, Codex, Cursor, or another shell-based agent. It will pause when browser approval or a human review is required.

What you will get

Codex returns a funnel table, segment comparison, and evidence-backed drop-off hypotheses after discovering the live PostHog schema instead of guessing event names.

A reproducible funnel table with the exact date window and query
Segment-level drop-off findings separated from unverified hypotheses

Before you start

A PostHog project granted to the active Soku brand
A defined business question for the funnel window and activation event

Works with

PostHog

Run it with Soku CLI

1
Select the PostHog project
List only the projects granted to the active Soku brand and confirm the intended project.
soku posthog list-projects
2
Inspect the live schema
Discover allowlisted read tools and event definitions before composing a query.
soku posthog list-tools --project-id <project_id>
3
Run the funnel query
Use the discovered schema and preserve the exact query in the report for reproducibility.
soku posthog query --project-id <project_id> --tool execute-sql --arguments '<json>'
4
Explain drop-offs
Separate measured drop-offs from hypotheses that require qualitative or implementation evidence.

Safety boundary

Read only

Soku exposes only allowlisted PostHog read tools here. The agent must inspect the schema, avoid guessed identifiers, keep credentials server-side, and make no PostHog mutations.

Related Soku skills: soku-posthog-growth-analytics

More Agent Workflows use cases

🔎

Audit Google Ads With Claude Code and Soku CLI

Give Claude Code read-only access to Google Ads through Soku CLI, inspect tracking, search terms, creative coverage, budgets, and Quality Score, then save an evidence-backed account audit.

📊

Review Meta Ads Performance With Codex and Soku CLI

Connect Codex to Meta Ads reporting through Soku CLI, compare campaign and creative performance, identify fatigue and CPA changes, and return a read-only optimization brief with evidence.

🗓️

Build a Weekly Cross-Channel Ads Report With an AI Agent

Use a local AI agent and Soku CLI to combine Google Ads, Meta Ads, and GA4 into one weekly report that explains changes, flags anomalies, and preserves the reporting window.

📈

Analyze GA4 Landing Pages With Claude Code and Soku CLI

Let Claude Code query GA4 through Soku CLI, rank landing pages by traffic and conversion behavior, find paid-traffic leaks, and save an evidence-backed CRO shortlist.