Add post on learning MCP with an LLM as reviewer

[ultimatecoder]

Adds I wrote the code and let an LLM review it — building the Transcript MCP server by hand while an LLM planned the work, reviewed each step, answered questions and tracked progress, without writing any code.

Formatting

Prompts and LLM output sit in collapsible <details> blocks. The prompt is open by default; the 260-line generated plan is collapsed so it does not dominate the post.

This uses kramdown's per-element markdown="1" rather than the global parse_block_html setting, so markdown parsing inside raw HTML is opted into here and nowhere else on the site.

Assets

The hero image ships as JPEG. As a photograph it needed 1.5 MB as PNG and still missed the 300 KB budget after scripts/optimize-images ran its full ladder (716 KB at the narrowest rung). JPEG holds 1600px at 220 KB.

Each of the four diagrams gets a lead-in sentence, descriptive alt text and a numbered caption.

Verification

• ./scripts/test passes — HTML-Proofer clean across 42 files, 186 external and 80 internal links
• ./scripts/optimize-images --check — all images within budget
• Rendering checked in the browser at 1280px and at narrow widths; the collapsible blocks wrap rather than scroll horizontally

One fix was needed for the link checker: the quoted plan contains a literal <link> placeholder that kramdown was parsing as an HTML tag, so it is now in backticks.

🤖 Generated with Claude Code