# Prompt #1: Turn Your AI Agent into Your Hashnode Publishing Assistant

Hello All,

Have you noticed how we consume tech content these days? Nobody reads a 10-minute blog post anymore — we paste a prompt into Claude or ChatGPT and say "do this for me." 🤖

So I'm trying something new on this blog. Welcome to the **Prompts** series — instead of long walkthroughs, each post here gives you **one battle-tested prompt** you can copy, paste into your AI assistant, and get the job done.

And this first one is a bit meta: the prompt below is the exact one I used to build this very series. My AI agent read my blog, verified my API access, created the draft, and published it — all through the Hashnode GraphQL API. I didn't open the dashboard once. It was a beautiful thing to see!

Sounds interesting? Let's dive in.

---

## What Does This Prompt Do?

It turns any LLM agent that can run `curl` — Claude Code, Cursor, Cline, you name it — into your personal **Hashnode blog assistant**. Once loaded, it can:

* List all your posts with views and stats
    
* Create drafts and publish posts into a series
    
* Update existing posts (fix that typo without leaving the terminal!)
    
* Do all of this **safely** — your token stays in an environment variable, never in the chat
    

> The key idea: your PAT (Personal Access Token) is a password. The prompt explicitly tells the LLM to never ask for it, never print it, and never write it to a file.

---

## When Would You Use This?

* You publish on Hashnode and live in the terminal like me
    
* You want to automate your publishing workflow
    
* You want quick answers like *"which of my posts has the most views?"* or *"which posts are missing tags?"* without clicking through the dashboard
    

---

## The Setup is Super Simple

1. Grab a Personal Access Token: Hashnode dashboard → **Account Settings** → **Developer** → Generate token.
    
2. Export it where your agent's shell can actually see it — add it to your shell profile:
    
    ```bash
    echo 'export HASHNODE_PAT=<your-token>' >> ~/.zshrc
    ```
    

That's it. Now hand your AI assistant the prompt below.

---

## The Prompt 📋

```plaintext
You are my Hashnode blog assistant. Talk to the Hashnode GraphQL API directly with curl.

Endpoint: POST https://gql-beta.hashnode.com
Body format: {"query": "...", "variables": {...}}

Auth: my Personal Access Token is in the HASHNODE_PAT environment variable.
Send it as the header "Authorization: Bearer $HASHNODE_PAT", always
shell-interpolated. Never ask me to paste the token, never print or echo it,
and never write it into a file.

Rules:
- Public reads (post, feed, user, tag) need no token. `me`, drafts, and every
  mutation require it.
- Pagination is cursor-based: read pageInfo.endCursor and hasNextPage, pass
  endCursor as the next "after". Page size caps at 100 (drafts: 50).
- Tags are referenced by slug, max 15 per post.
- Introspection is enabled — if you're unsure about a field name, introspect
  the schema instead of guessing.
- If a call returns FORBIDDEN with a "Pro plan" message, stop and tell me.
  Do not retry.
- Before any mutation (publish, update, delete), show me exactly what you're
  about to change and wait for my confirmation.

Start by querying `me` to confirm auth works and find my publication id.
Then wait for my instruction — for example: "list my posts with view counts",
"draft a post titled X in series Y", or "fix the typo in post Z".
```

---

## The Catch (Gotchas I Hit Today)

As with any workflow, there are a few things you should be aware of before jumping in:

* **Exporting in your terminal is not enough.** 🤦‍♂️ Agent tools spawn fresh shells — a variable exported in your open terminal window never reaches them. Put it in `~/.zshrc` (or `~/.bashrc`) so every new shell picks it up. This one cost me three tries.
    
* **Navbar items and series are read-only via the API.** You can query them, but creating them is dashboard-only. Everything else — drafts, posts, series assignment — is fully scriptable.
    
* **Queries are cached for ~25 seconds.** If you just published and a read looks stale, wait a moment before assuming something went wrong.
    

---

## Key Takeaways

* Your blog is now an API away — one prompt turns any LLM agent into a publishing assistant
    
* Keep the token in an env var, and let the prompt enforce the safety rules
    
* Ask for confirmation before every mutation — trust, but verify
    

Paste the prompt, say *"list my posts"*, and enjoy never opening the dashboard again. 😉

More prompts coming soon — keep an eye on the **Prompts** tab in the navbar. Happy prompting!