Enforce anti-AI writing patterns with Vale

[benbalter]

Ports the anti-AI prose rules from the book project into a local AIPatterns Vale style, translated for a conversational blog, and makes Vale actually run in CI.

What's in here (three distinct kinds of change)

1. Lint infrastructure

• New checked-in Vale style .github/styles/AIPatterns/ (11 rules), tiered for this site's voice:
  • Error (blocks CI): hedging openers ("It's important to note that…"), reader-probing ("you might be wondering"), meta-commentary ("Let's dive into"), "This ensures…" constructions, performative enthusiasm ("Great question!")
  • Warning: formulaic transitions ("Furthermore," "In conclusion,"), setup phrases ("When it comes to"), dramatic flourishes ("Spoiler alert", "Plot twist")
  • Suggestion: negation-antithesis ("it's not X—it's Y"), rhetorical questions, weak openings, vague intensifiers—patterns this blog uses deliberately, so they never gate
• Vale now actually runs in CI. It was configured but silently skipped—ci.yml never installed the binary. Adds pinned install + vale sync + style cache to the Content Linting job.
• Unblocked Vale 3.x: the BenBalter pack crashed Vale at config load (removed link scope). Fixed upstream in vale-styles#1, released as v0.0.2, URL bumped here.
• .mdx posts (28 of them) are now linted too ([formats] mdx = md + script/vale globs).
• Enabling Vale at error level surfaced 367 errors across 162 legacy posts from existing package rules. Stylistic ones (clichés, "very", e.g.,, US date formats, quoted profanity, duplicate So/But flags) are demoted to warning in .vale.ini; the rest were genuine and are fixed below.

2. Mechanical fixes in legacy posts (typos the new error level caught)

• Doubled words ("can can", "the The"), an orphaned stray "I" line in the AI-first PM post
• TODO examples wrapped in backticks, 10AM → 10 AM, lowercase domain in CONTRIBUTING heading, lowercased a .PDF link text
• File-level Vale.Repetition off-switch for the law-review post whose legal citations (U.S.C.C.A.N., L.L.) false-positive

3. Prose edits in legacy posts — review these

Eleven small rewrites remove now-error patterns from 2011–2018 posts. They're all filler-removal ("It's worth noting that…", "As noted above,", "as mentioned earlier,", "Let's take a look at" → "Here are", "This creates a" → "The result is"), but they do alter published text—worth a skim in the diff.

Dramatic flourishes ("Spoiler alert: it didn't.") were deliberately kept in the vintage posts and demoted to warning—they're authentic human voice there; new writing still gets flagged visibly.

Verification

• script/vale (error level): 0 errors across 204 files
• Smoke-tested every rule tier fires (including a fix for punctuation-terminated tokens breaking Vale's \b wrapping)
• script/cibuild-content passes locally (remark, textlint, vale; harper/languagetool skip without local installs)
• npm run build passes
• Vale release asset URL verified (HTTP 200)

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Updated (UTC)
✅ Deployment successful! View logs	benbalter-github-com	a7d3abb	Jun 08 2026, 02:46 AM