AgentTrust Blog Posting Skill

Run the whole workflow inside this skill: topic → markdown draft → cover image → HTML payload → API submission.

Core rules

• Always generate the cover image before submitting the post.
• Always use the Nano Banana Gemini image generator through the sibling skill script.
• Always attach the generated image in the same blog API request via image_base64 and image_filename.
• Never silently downgrade to “text-only blog post” unless the user explicitly asks.

Acceptable starting points

A. Topic / instructions only

If the user gives only a topic, article brief, or raw notes:

• draft the article first
• save it under blog-drafts/YYYY-MM-DD-slug.md
• include complete frontmatter
• then run the publish helper script

B. Existing markdown draft

If the user already gives a markdown file, validate it and publish from that.

Expected frontmatter shape:

---
title: "Article Title"
description: "150-160 char SEO description"
date: "YYYY-MM-DD"
author: "Gabriel Guidarelli"
tags: ["ai-security", "agenttrust"]
status: "draft"
target_keyword: "primary keyword"
---

Use frontmatter as the source of truth for:

• title
• excerpt / meta_description
• focus_keyword
• tags
• slug
• author

Default workflow

1. Write the article if needed

When starting from a topic:

• write the blog post in Gabriel’s voice
• keep it clear, direct, and useful
• save the markdown draft locally before submitting
• make sure the frontmatter is complete before moving on

2. Generate the header image with Nano Banana

Use the bundled sibling script indirectly through:

python3 skills/blog-posting-skill/scripts/publish_agenttrust_blog.py <draft-path>

That helper script already calls:

skills/nano-banana-image-generator/scripts/generate_gemini_image.py

Do not invent another image workflow unless the Nano Banana path is broken.

The image prompt is locked to the AgentTrust trust-flow composition. Read references/cover-branding.md for the exact prompt and visual rules.

3. Convert markdown to clean HTML

The helper script converts the article body into simple clean HTML.

Supported structure:

• paragraphs → <p>
• ## → <h2>
• ### → <h3>
• bullet lists → <ul><li>
• inline links → anchors
• basic bold / italic / inline code

4. Submit the post and image together

Use the production API:

POST https://agenttrust.ai/api/submitBlogPost

Auth:

Authorization: Bearer $AGENTTRUST_API_KEY
Content-Type: application/json

Required payload core:

• title
• slug
• content

Recommended payload fields used by this skill:

• excerpt
• tags
• meta_title
• meta_description
• focus_keyword
• image_alt
• author_name
• image_base64
• image_filename

When image_base64 is present, AgentTrust uploads the image and sets the blog image fields automatically. Prefer this over trying to host og_image manually.

Use the helper script

Standard end-to-end publish

python3 skills/blog-posting-skill/scripts/publish_agenttrust_blog.py \
  blog-drafts/2026-04-14-my-post.md

Publish with explicit topic context for the image

python3 skills/blog-posting-skill/scripts/publish_agenttrust_blog.py \
  blog-drafts/2026-04-14-my-post.md \
  --topic "prompt injection protection for AI email"

Dry run

python3 skills/blog-posting-skill/scripts/publish_agenttrust_blog.py \
  blog-drafts/2026-04-14-my-post.md \
  --dry-run --json-out /tmp/blog-post-preview.json

What the helper script does

publish_agenttrust_blog.py:

• parses frontmatter
• validates required metadata
• derives slug if missing
• builds the Nano Banana prompt
• generates the cover image
• base64-encodes the generated image
• converts markdown to HTML
• submits the article and image together
• returns the API response with paths and payload summary

Reporting rules

After completion, report:

• article title
• markdown draft path
• generated image path
• whether the post was actually submitted or only dry-run built
• post_id
• returned status, normally review
• any failure reason if the image step or API step broke

Practical rules

• Use the production endpoint unless the user explicitly asks for test.
• Use Bearer auth, not x-api-key.
• Posts submit in review state; do not claim they are live unless confirmed.
• Keep drafting, image generation, and submission as one workflow when the user asks for end-to-end publishing.
• If the Nano Banana image step fails, fix or retry that step before publishing.
• Prefer the helper script over manual curl because it keeps the workflow deterministic.

Output checklist

Before finishing, verify:

• draft exists locally
• frontmatter is complete
• cover image generated with Nano Banana
• image saved locally
• image attached through image_base64
• HTML payload built cleanly
• article submitted to API or explicitly dry-run only
• post_id / status captured
• Gabriel given the final paths and result