| name | draft-anti-feed-blog |
| description | Draft a single ranked blog post for the BuildOS anti-feed / thinking-environment cluster. Reads the consolidated brief at docs/marketing/anti-feed/blog-context.md, picks the next-up T## from the WS09 dashboard (or a user-specified topic), produces a 1,200–2,000 word draft with frontmatter and cross-links, and saves it under apps/web/src/content/blogs/philosophy/. Triggers on "/draft-anti-feed-blog", "draft the next anti-feed blog", "write the next cluster blog", "draft T## blog", "write T## post". Does NOT build publish kits, TikTok scripts, or non-cluster blog posts — use the old `anti-feed` skill for those. |
Draft Anti-Feed Cluster Blog
A focused, single-purpose skill: produce one ranked cluster blog from the BuildOS anti-feed / thinking-environment cluster (WS09).
This skill follows the argument-based pattern — its entire brand, voice, audience, vocabulary, output spec, and drafting flow lives in one consolidated context file:
docs/marketing/anti-feed/blog-context.md ← always read this first.
When the brand guide, topic map, or anti-AI doctrine changes, the context file gets updated in one place and this skill picks it up next run. No skill edits needed.
When to load this skill
- User runs
/draft-anti-feed-blog
- User asks to draft the next cluster blog (T34–T43 range or beyond)
- User asks to write a specific T## post
- User asks for a free-topic blog inside the anti-feed cluster vocabulary
Do not load for:
- Publish kits (Twitter / LinkedIn / Instagram / TikTok / Reddit) → old
anti-feed skill, menu option 2
- Standalone TikTok scripts → old
anti-feed skill, menu option 3
- Founder essays, changelog posts, product launches → different mode, different voice
- Brand or positioning decisions → brand guide is the arbiter
The flow
Step 0 — Load the context file (always)
Read docs/marketing/anti-feed/blog-context.md in full. One read, end-to-end. This is the single source of truth for voice, audience, vocabulary, output format, and the drafting flow.
Do not proceed without this read. Do not paraphrase from memory; the file changes and the annealing log accumulates.
Step 1 — Identify the target post
Per §11 Step 1 of the context file:
- Open
docs/marketing/distribution/workstreams/WS09-anti-feed-cluster.md and read the live status dashboard.
- Pick the highest-priority 🟡 or ⚪ row (lowest T-number on tie). Skip ✅ and ⏸.
- Confirm with user: "Next up is T## — {title}. Draft it now?"
- If user names a different T## or a free topic, use that.
Step 2 — Outline before prose
Per §11 Step 2:
Produce a 5-bullet outline (lead-with-relief opener, mechanic/claim, reframe with term-to-own, BuildOS frame, the practice). Show to user. Wait for confirmation or edits before writing prose. Do not skip this gate.
Step 3 — Draft
Per §11 Step 3 + §10 (output format):
- Length: 1,200–2,000 words (aim 1,500)
- Body shape: hook → mechanic → reframe → BuildOS section (≤15%) → practice → closing line
- Inline checks while drafting: no AI in hook, ≥1 term-to-own appears 3+ times, ≥1 external source cited, em-dash density natural
Step 4 — Frontmatter + cross-links
Per §10:
- Fill frontmatter exactly (every field)
- ≥2 inline links to prior cluster posts (verify the files exist in
apps/web/src/content/blogs/philosophy/)
- 1 link to
docs/marketing/strategy/anti-feed-content-topic-map.md
- Pick cross-links by vocabulary continuity, not just topic adjacency
Step 5 — Voice checklist (pre-save gate)
Run every item in §11 Step 5 of the context file. Show the user the checklist with results. If any item fails, revise before saving. This is non-negotiable.
Step 6 — Save
Save to apps/web/src/content/blogs/philosophy/{slug}.md. Confirm to user: path, word count, reading time.
Step 7 — Annealing log
Ask: "Any learnings from this draft to capture in the annealing log?"
If yes — append a dated entry to §13 of docs/marketing/anti-feed/blog-context.md using the entry format in §13. If the entry includes a "Rule update" line, also edit the relevant rule section (§3, §4, §6, §10, or §11) of the context file in the same edit.
If no — skip silently.
Step 8 — Hand-off prompt
Prompt: "Draft saved at {path}. Want to build the publish kit now?"
- Yes → invoke the old
anti-feed skill (menu option 2). Don't try to build the kit yourself.
- No → end run.
Step 9 — Status update reminder
Surface this reminder before ending the run:
"Update three places before this counts as done: WS09 dashboard (T## → 🔵 or ✅), docs/marketing/distribution/README.md task quick-map, buildos-strat-tasks.md. All three or it's drift."
This skill does not write status updates itself.
Hard limits
Repeated from §12 of the context file, because prompt drift is real:
- No publish kits. Old
anti-feed skill, option 2.
- No standalone TikTok scripts. Old
anti-feed skill, option 3.
- No status-dashboard writes. Surface the reminder; user updates.
- No posting / publishing. Save to disk only.
- No non-cluster blog posts. Different mode, different skill.
- No new vocabulary coining mid-post. Flag for user; do not add to canon unilaterally.
Why this skill is so thin
The argument-based pattern: the skill is the process, the context file is the knowledge. Voice, audience, vocabulary, anti-patterns, proof assets, ranked posts, output format — all in blog-context.md. This file just enforces the flow.
When the brand guide updates, the topic map shifts, or a new anti-pattern emerges from real drafts: edit the context file once. Every future run picks it up. No skill rewrites, no drift between skill and source-of-truth.
The annealing log inside the context file is the feedback loop. Every run that produces a learning makes the next run sharper.
Source of truth
- Context (this skill consumes):
docs/marketing/anti-feed/blog-context.md
- Live status (read each run):
docs/marketing/distribution/workstreams/WS09-anti-feed-cluster.md
- Output destination:
apps/web/src/content/blogs/philosophy/{slug}.md
- Companion skill (publish kits):
.claude/skills/anti-feed/SKILL.md (menu option 2)
