For AI agents

Your agent writes the fix. We check its work.

Every fix PocketBees can't apply itself comes as a scoped prompt: the change to make and a clear done condition. When your agent finishes, PocketBees re-checks the live site and tells you whether it worked.

fix-01.prompt · Claude Code

# Task (from PocketBees)
Add Organization and WebSite schema to the homepage.

# What to do
Add Organization and WebSite JSON-LD to the homepage head, using the real name, logo URL and social profiles.

# Why it matters
Structured data tells search engines and AI assistants who you are.

# Done when
Homepage JSON-LD includes Organization and WebSite.

How the hand-off works

01 · COPY

Copy the prompt

Each fix has its own prompt: what to change, why, and how to check it.

02 · BUILD

Your agent makes the change

Paste it into Claude Code, Cursor, Codex or Hermes. You review and merge as usual.

03 · VERIFY

PocketBees checks it

Once the change is live, PocketBees re-checks against the done condition and marks the fix proven, or tells you what's still missing.

What's in every prompt

Agents do their best work with a narrow task and a clear finish line.

# Task

One change, described in a sentence.

# What to do

The exact change, and the commands to verify it.

# Why it matters

So the agent doesn't trade it away for something else.

# Done when

A condition PocketBees can check on the live site.

Or connect your agent directly

MCP server

Agents that speak MCP can list your sites and jobs, fetch each prompt and ask for a re-check. Create a token in Settings → AI agents.

What agents can't do

Mark a fix done (PocketBees checks it itself), change Cloudflare or change your Autopilot settings.

Read the docs

llms.txt describes PocketBees for AI tools in plain text.