Skip to content

cgallic/kai-cmo-harness / harness/skills/kai-write

kai-write

Verified

Write a single piece of marketing content using the marketing harness frameworks and quality gates. Supports blog posts, LinkedIn articles, emails, cold outreach, press releases, ad copy, and TikTok scripts. Automatically loads the right framework, skill contract, and persona. Use when "write a blog post", "draft an email", "LinkedIn article", "cold outreach", "write ad copy", "press release", "TikTok script", or any single content creation request. For building a complete email system, use /kai-email-system instead.

Kai root note: knowledge/, harness/, and scripts/ paths in this skill live in the Kai install, not the user's project. Resolve them against the first ancestor directory of this SKILL.md that contains a knowledge/ folder (the Kai plugin root, ~/.claude/kai, or the kai-cmo-harness repo). MARKETING.md, memory/, and any output files live in the current project. If a referenced scripts/ command is not available in this install, say so, skip it, and continue with the file-based guidance — never fabricate its output.

Write one piece of content using the marketing harness. Loads the right framework, applies quality gates, outputs ready-to-publish copy.

Phase 0: Load Product Context

Check if MARKETING.md exists in the project root (same directory as CLAUDE.md, README.md, package.json).

If it exists: Read it — skip product discovery questions. It has the product name, ICP, value prop, monetization, brand voice, current channels, and competitive landscape.

If it does NOT exist: Auto-explore the codebase to create it in the project root (next to CLAUDE.md). Do NOT ask the user what the product is. Read CLAUDE.md, README.md, PROJECT.md, package.json, landing pages, and any project files. Search for email/ad/analytics config. Then create MARKETING.md using the template from /kai-email-system. Present draft to user for confirmation.


Step 1: Identify Format and Load Context

Determine the content format from the user's request, then load the corresponding files from the harness (paths relative to the project root):

FormatFrameworkContractChecklist
Blog postknowledge/frameworks/content-copywriting/algorithmic-authorship.mdharness/skill-contracts/blog-post.yamlknowledge/checklists/content-checklist.md
LinkedInknowledge/channels/linkedin-articles.mdharness/skill-contracts/linkedin-article.yaml—
Email (lifecycle)knowledge/channels/email-lifecycle.mdharness/skill-contracts/email-lifecycle.yaml—
Email (cold)knowledge/channels/email-lifecycle.md + harness/references/cold-email-rules.mdharness/skill-contracts/cold-email.yaml—
Meta adsknowledge/channels/meta-advertising.md + harness/references/meta-ads-rules.mdharness/skill-contracts/meta-ads.yamlknowledge/checklists/meta-advertising-checklist.md
Google adsknowledge/channels/paid-acquisition.md + harness/references/google-ads-policy-reference.mdharness/skill-contracts/google-ads.yamlknowledge/checklists/paid-acquisition-checklist.md
Press releaseknowledge/channels/press-releases.md—knowledge/checklists/pr-checklist.md
TikTokknowledge/channels/tiktok-algorithm.md—knowledge/checklists/tiktok-checklist.md
SEO contentknowledge/frameworks/content-copywriting/algorithmic-authorship.md + knowledge/frameworks/aeo-ai-search/aeo-ai-search-playbook-2026.mdharness/skill-contracts/blog-post.yamlknowledge/checklists/seo-checklist.md

Read the framework file(s) and skill contract before writing. Do not guess at rules — load them.

Step 2: Brief

If the user hasn't provided a brief, create one using harness/brief-schema.md. At minimum, confirm:

  • Persona — which of the 8 personas from knowledge/personas/_persona-index.md?
  • Angle — what's the specific frame? (not just the topic)
  • CTA — what do we want the reader to do?

For ad copy, also confirm the platform so the right policy reference gets loaded.

Step 3: Write

Apply the framework rules. Key universal rules:

  • No banned words (leverage, utilize, synergy, innovative, deep dive, circle back, touch base, moving forward, at the end of the day)
  • No AI slop ("In conclusion", "It's important to note", "In today's rapidly evolving")
  • No X-not-Y binary clichés — every form: "X, not Y" / "isn't X — it's Y" / "It's a Y, not a Z" / "X — not Y". The pattern reads as LinkedIn slop and slips past subjective scoring. Use parallel-positive contrast or a single load-bearing claim instead.
  • Match the persona's language patterns and pain points
  • Single primary CTA
  • Follow word count from the skill contract

For SEO content, also apply Algorithmic Authorship rules:

  1. Conditions AFTER main clause ("Do X if Y")
  2. Instructions start with verbs
  3. Sentences under 20 words
  4. Bold the answer, not query-matching terms

Step 4: Quality Gate

Score and validate before delivering:

  1. Four U's — Unique, Useful, Ultra-specific, Urgent (each 1-4). Min 12/16 for blog/SEO, 10/16 for email/ads.

  2. Banned words — zero Tier 1 violations

  3. AI slop — zero instances

  4. Voice patterns (programmatic — DO NOT skip) — grep the draft with the Grep tool for these regexes. Any hit fails the gate.

    • , not [a-z] (catches "X, not Y")
    • — not [a-z] (catches "X — not Y")
    • \bisn'?t [a-z][^.\n]+ — it'?s\b (catches "isn't X — it's Y")
    • \bIt'?s (a\|the\|an) [^.\n]+, not (a\|the\|an)\b (catches "It's a X, not a Y")
    • \bIf you [a-z][^,.]+, [a-z] (catches "If you X, Y")
    • \bHere'?s the thing\b / \bLet that sink in\b / \bHot take\b / \bI'?ll be honest\b (LinkedIn slop)

    Skip matches inside HTML comments and code fences. Replacements: collapse to a single load-bearing claim, or use parallel-positive contrast where both halves are affirmative.

  5. Format-specific gates — from the skill contract (subject line length, word count, CTA count, etc.).

Hook integration: if the project has .claude/hooks/voice-gate.py, the hook fires PostToolUse on Edit/Write and catches voice violations at draft time. The gate step still runs the regexes — belt and suspenders.

If the piece fails, fix the specific issues and re-score. Max 2 retries. After 2 failures, deliver with failures noted.

Step 5: Deliver

Output the content with quality gate scores visible. Save to workspace/ if requested.

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 468kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 321kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 321kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 321kUpdated Claude CodeCodexVerifiedInstall