/blog @path/to/source.md          # markdown file
/blog @path/to/notes.txt          # text file
/blog <paste plain text here>     # plain text directly

$ARGUMENTS

• If it looks like a file path (contains

  /

  or

  .md

  or

  .txt

  ), read the file.
• Otherwise, treat

  $ARGUMENTS

  itself as the source content.

digraph blog_post {
    "Read source file" [shape=box];
    "Analyze content" [shape=box];
    "Ask category" [shape=box];
    "Gap analysis" [shape=box];
    "Gaps found?" [shape=diamond];
    "Ask user to fill gaps" [shape=box];
    "Sufficient?" [shape=diamond];
    "Write draft" [shape=box];
    "Self-review" [shape=box];
    "All rules pass?" [shape=diamond];
    "Fix violations" [shape=box];
    "Save .md file" [shape=box];

    "Read source file" -> "Analyze content";
    "Analyze content" -> "Ask category";
    "Ask category" -> "Gap analysis";
    "Gap analysis" -> "Gaps found?";
    "Gaps found?" -> "Write draft" [label="no"];
    "Gaps found?" -> "Ask user to fill gaps" [label="yes"];
    "Ask user to fill gaps" -> "Sufficient?";
    "Sufficient?" -> "Ask user to fill gaps" [label="no, ask again"];
    "Sufficient?" -> "Write draft" [label="yes"];
    "Write draft" -> "Self-review";
    "Self-review" -> "All rules pass?" ;
    "All rules pass?" -> "Fix violations" [label="no"];
    "Fix violations" -> "Self-review";
    "All rules pass?" -> "Save .md file" [label="yes"];
}

1. Read the source content from

   $ARGUMENTS

   (file path → read file; plain text → use directly)
2. Analyze content: identify language, key topics, available data
3. Ask user to confirm category (case / study / guide) via AskUserQuestion — suggest your best guess as the first option
4. Gap analysis — compare source content against the chosen category template. Identify which sections lack sufficient information to write a useful post.
5. Fill gaps — if gaps exist, ask the user one question at a time via AskUserQuestion until every template section has enough material. Use multiple choice when possible. Do NOT proceed to writing until the content is sufficient.
6. Write draft — write the post following all rules in the Iron Rules section, incorporating both the source file and the user's answers
7. Self-review — review the draft against the Self-Review Checklist below. Do NOT show the draft to the user yet.
8. Fix & repeat — if any checklist item fails, fix the violation and re-review. Repeat until every item passes.
9. Save the final post to user-specified path, or default:

   blog/posts/YYYY-MM-DD-<slug>.md

• No fabricated content — every claim traces back to the source or user's answers
• Code blocks are identical to the source (except redacted secrets)
• No secrets in output — API keys, tokens, passwords, connection strings replaced with

  <REDACTED>

• Post length is proportional to source length (no padding)

• Active voice throughout — no "was implemented", "was observed"
• Positive form — no "don't forget to", "it is not uncommon"
• Specific and concrete — no "significantly improved", "much faster"
• No needless words — every sentence earns its place
• Emphatic word at end of each key sentence
• One paragraph, one topic — no paragraph covers two subjects
• Topic sentence leads each paragraph
• Parallel structure in all lists — items share the same grammatical form
• Related words kept together — modifiers next to what they modify
• No loose sentence chains — no run-on compound sentences
• Consistent tense within each section

• Zero AI puffery (groundbreaking, seamless, pivotal, vital, testament)
• Zero overused AI words (delve, leverage, multifaceted, foster, realm, tapestry)
• Zero empty -ing phrases (ensuring reliability, showcasing features)
• No filler intros ("In recent years...", "As we all know...")
• No restating conclusion
• No puffery-as-emphasis (non-negotiable, crucial, indispensable)
• No promotional adjectives (robust, powerful, state-of-the-art)
• No formatting overuse (excessive bullets, bold on every line, emoji)

• Frontmatter matches the spec exactly (no extra fields)
• Template sections are all present (or intentionally skipped by user)
• Hook is specific and pulls the reader in
• TL;DR is 3 bullets max
• Section lengths vary (not all the same)

• No data at all — template section has zero corresponding content in source
• Too vague — source mentions something but lacks specifics (e.g., "performance improved" without numbers)
• Missing context — reader would ask "why?" or "compared to what?" and the source doesn't answer

"The Results section needs numbers. Do you have benchmark data?"

• Yes (I'll provide it)
• Not yet measured (skip Results section)
• Only rough estimates

• Never fabricate. Use only what exists in the source. If the source says "benchmark: p50 3ms", write about that result. Do not invent context, backstory, or details.
• Preserve code blocks exactly. No renaming variables, no adding lines, no removing lines. Format/comment cleanup only.
• Do not over-expand. A 30-line source does not become a 120-line post. Add structure and connective tissue, not padding.

• Active voice. "We reduced latency" not "Latency was reduced."
• Omit needless words. Cut every sentence that repeats what was already said.
• Specific over vague. "p99 dropped from 45ms to 8ms" not "performance improved significantly."
• Positive form. "Use sync.Pool" not "Don't forget to consider sync.Pool."
• Place emphasis at end of sentence. The important word goes last.
• One paragraph, one topic. A paragraph that covers both "why we chose Go" and "how batching works" must be split.
• Topic sentence first. Start each paragraph with its main point. Supporting details follow.
• Parallel structure. List items must share the same grammatical form. All nouns, all verbs, or all imperative — pick one and hold it.
• Keep related words together. Place modifiers next to what they modify. "A server written in Go that handles 4200 req/s" not "A server that handles 4200 req/s written in Go."
• No loose sentence chains. Avoid run-on compound sentences. Break into shorter sentences with varied structure.
• One tense per section. Pick past or present and stick with it within a section. Do not mix.

• Redact secrets before output. If the source content contains patterns that look like API keys, tokens, passwords, or connection strings, replace them with

  <REDACTED>

  in the output and warn the user.
• Never reproduce

  .env

  variables, Bearer tokens, or private keys in the final blog post.

---
title: "<concise, descriptive title>"
date: YYYY-MM-DD
description: "<one sentence summary — what this post delivers to the reader>"
category: case | study | guide
tags:
  - tag1
  - tag2
  - tag3
draft: true
featured: false
---

• If the source has frontmatter, preserve and fill gaps
• If not, generate from content
• Do not add fields not listed above (no

  author

  , no

  slug

  , no

  image

  )
• Match the language of the source content

• 3 bullets max. Busy reader stops here if not relevant.

• The specific problem being solved
• Real constraints (latency budget, team size, existing system, etc.)

• What approach was chosen and what alternatives were considered
• Why this one won (concrete trade-off reasoning)

• Key code with brief explanation
• Design decisions that matter

• Benchmarks, metrics, before/after (use tables when comparing)

• What actually surprised you
• What you would do differently

• Actionable insights the reader can apply to their own work

undefined

• 3 bullets max.

• The central concept, explained clearly

• Technical details, code/math/diagrams

• Why this matters for working engineers

• Strengths, weaknesses, open questions

• Key points + recommended resources (if any)

undefined

• 3 bullets max. What you'll be able to do after reading this.

• What the reader needs before starting (tools, versions, knowledge)

• Step-by-step with code blocks
• Each step: what to do, why, and expected result
• Group related steps under subheadings if needed

• Common errors and how to fix them
• Edge cases worth knowing

• Summary of what was covered
• Next steps or further reading

undefined

• Adding a "Conclusion" section that restates what was already said
• Writing 3 paragraphs where 3 sentences would do
• Inventing quotes, scenarios, or backstory not in the source
• Starting with "In the rapidly evolving world of..."
• Using more than 2 adjectives in a single sentence
• Every section having the same length (vary rhythm)