X Post Generator

Generate polished X posts from user ideas/drafts with strict output formatting and brand alignment.

Workflow

1) Gather required input

Collect these fields before generating:

• Source info: user draft or user idea
• Post goal: announce / recap / invite / opinion / update
• Audience: who should care
• Media links: image/video/doc links to reference in the post

If any required part is unclear, ask concise clarifying questions first.

2) Load brand profile

Read references/brand.md.

If this is the first run (brand profile still placeholder/empty), ask user to define style. Offer quick options and allow custom:

1. Professional & concise
2. Friendly & community
3. Technical & credible
4. Visionary & bold
5. Playful & energetic
6. Custom style (user-defined)

Then continue only after style is clear.

3) Generate post options

Produce at least 3 distinct options.

Hard rules:

• Auto-generate relevant hashtags, maximum 5 unique hashtags
• Keep hashtags relevant to topic + brand
• If media links are provided, reference them explicitly in each option
• If media links are not provided, do not add a media line
• Keep tone aligned with references/brand.md

4) Output format (strict, Telegram copy-first)

Always format output so Telegram users can copy each proposal in one tap.

Rules:

• Return only option labels + fenced code blocks (no extra commentary before/after).
• Put exactly one full post per code block.
• Keep plain text inside code blocks (no markdown bullets inside).
• Use triple backticks with text language.

Use this structure:

Option 1

<post body>

#hashtag1 #hashtag2 #hashtag3
📎 Media: <link1> <link2>

Option 2

<post body>

#hashtag1 #hashtag2 #hashtag3
📎 Media: <link1> <link2>

Option 3

<post body>

#hashtag1 #hashtag2 #hashtag3
📎 Media: <link1> <link2>

If no media links are provided, omit the media line entirely.

5) Learn from accepted post (auto-update brand memory)

When user accepts one option, run:

python3 scripts/learn_from_accepted.py --post "<accepted_post_text>"

This will:

• Store accepted post into references/accepted_posts.jsonl
• Keep only last 10 accepted posts
• Auto-update the learned section in references/brand.md

Quality checks

Before final answer, verify:

• At least 3 options
• Each option in its own text code block
• No more than 5 hashtags per option
• Media line present only when media links are provided; absent otherwise
• No ambiguous placeholders left
• No extra commentary outside option labels + code blocks