Skip to content

CLAUDE.md

Latest commit

 

History

History
60 lines (55 loc) · 3.41 KB

CLAUDE.md

File metadata and controls

60 lines (55 loc) · 3.41 KB
Raw

Claude preferences

Writing style

  • No em dashes (U+2014, the character that looks like a long dash: —). Use a comma, period, or restructure the sentence instead.
  • No other AI writing artifacts: no "delve", no "it is worth noting", no "in conclusion", no filler throat-clearing.
  • Keep prose direct, short, and conversational. Plain language over formal.
  • Avoid hedging phrases like "I cannot", "please note that", "as an AI".

Blog posts

  • Use BLOG_POST_TEMPLATE.md at repo root as the skeleton for new posts.
  • Audience is mixed (leaders + engineers). TL;DR plus a callout table near the top gives leaders a stop point; deeper sections serve engineers.
  • Standard section order: TL;DR → Why this matters → What I did → How I did it → What I would do differently → Takeaways.
  • STAR is implicit in the structure. Situation = "Why this matters", Task = the **Goal**: line at the top of "What I did", Action = "What I did" + "How I did it", Result = TL;DR + callout table. Recruiters can lift each letter in a 60-second skim.
  • Voice is first-person, retrospective.
  • Fully generic content. No employer name, no commercial vendor names unless the post is explicitly about a public OSS tool. Use category descriptors (e.g. "CNAPP", "policy-as-code engine", "database access proxy").
  • Numbers as orders of magnitude (tens, hundreds, a petabyte). No dollar figures tied to an employer.
  • Length target: 5-10 minute read (~1000-2500 words). Each line must earn its place. Cut filler ruthlessly. "Every line justified" is the bar.
  • Executive readability: a non-technical executive should be able to read the TL;DR, callout table, and "Why this matters" sections and walk away with the headline and the business value. Technical depth belongs in "How I did it."
  • No redundancies across the suite. Before writing a new post, scan existing _drafts/ and _posts/ content; if the lesson overlaps meaningfully with an existing post, either cut the redundancy or fold the new content into the existing post.
  • Section budget (devil's-advocate rules):
    • "What I would do differently": 2-4 bullets, forward-looking framing ("Ship X first" not "I should have shipped X first"). Cut filler. If you cannot get to 2 genuine lessons, the section should not exist.
    • "Takeaways": 3-6 bullets, each unique to the post. Cut anything that restates another section or reads as generic industry advice.
  • Self-score and grade at the end of every draft, in an HTML comment block (<!-- ... -->) so the rubric survives in markdown source but does not render. Rubric (out of 100):
    • Tone (15) : matches voice, no AI-isms, no em dashes
    • Clarity (20) : STAR is cleanly identifiable
    • Brevity (15) : hits the 5-10 minute target
    • Evidence (20) : concrete numbers, patterns, outcomes
    • Uniqueness (15) : does not overlap with other posts in the suite
    • Narrative fit (15) : advances the professional story arc
    • Grades: A 90+, B 80-89, C 70-79, D <70.
  • Address gaps identified in the self-score before promoting to _posts/. List the gaps explicitly in the comment block and either fix them or note why they were left.
  • Workflow: draft into _drafts/ (Jekyll ignores it for deploy), review, then mv to _posts/ once approved. Filename for _posts/ must be YYYY-MM-DD-slug.md. Permalink convention is /blog/<slug>/.
  • Preview locally with bundle exec jekyll serve --drafts.