Draft Post
Generate a single content asset tailored to one channel. Strategic layer lives in build-in-public — this is the execution layer for "I need to ship one good post right now."
When to Use
- User says "draft a tweet/thread/post about X"
- A cron job from
build-in-public is triggered and needs to produce the actual draft
- User wants 2-3 variations of the same idea for different channels
When NOT to Use
- Multi-asset campaign with a calendar → use
campaign-plan (TBD) or just iterate this skill
- Quality check on existing draft → use
brand-review-content
- Engagement reply to someone else's post → use
reply-research
Inputs to Gather
If any of these are unclear, ASK before drafting. Do not invent context.
- Channel — X post, X thread, LinkedIn, blog, email. Default to X if unspecified.
- Topic/Trigger — what shipped, what failed, what you're thinking about. Specific > abstract.
- Goal — engagement, build-in-public transparency, lead-gen, education, or community signal.
- Length constraint — auto-derived from channel but override if user asks.
- Voice — if user has a configured brand voice (see
brand-review-content skill for the framework), apply it. Otherwise default to: first-person, specific, no AI-isms, no corporate jargon, dry humor OK, never "leverage" / "game-changer" / "unlock".
Channel-Specific Structure
X Post (single tweet, ≤280 chars)
Pattern:
- Hook in first 8 words — concrete detail, surprising number, contrarian take, or sharp question
- One specific point — never two. Threads exist for that reason.
- Optional CTA — only if there's a real ask (link, reply prompt, follow). Don't manufacture one.
Avoid:
- Opening with "Just" or "So"
- Generic engagement bait ("agree?", "thoughts?")
- More than 1 emoji unless writing in a tone where 2-3 fit
- Hashtags inside body text (max 1-2 at the end if needed)
Output 2-3 variants when possible. Different hook angles, same point.
X Thread (5-10 tweets, optional 25k mode for power users)
Structure:
- Hook tweet — earn the click. Promise something specific.
- Setup — 1 tweet of context for people who don't know you
- Body — 3-7 tweets, one idea per tweet. Use line breaks aggressively.
- Reversal or insight — the moment readers screenshot
- CTA — soft. Follow, reply, link out. Match the goal.
Pitfalls:
- Don't start tweet 2 with "A thread 🧵" — the hook should make people want it
- Don't number tweets unless the content is genuinely a list
- End on substance, not "follow for more"
LinkedIn Post (150-300 words, paragraph breaks every 1-2 lines)
Structure:
- Hook line, then a blank line
- Setup paragraph (2-3 short lines)
- Substance (3-5 short paragraphs, one idea each)
- Takeaway / insight
- Optional question to drive comments
Tone shift from X: more professional framing, less in-jokes, no shitposting. Same authenticity, dressed differently.
Blog Post (500-2000 words)
Structure:
- Headline — provide 2-3 options. Specific beats clever.
- Lede — first paragraph promises what the reader gets and proves the writer earned the read
- 3-5 H2 sections with descriptive headings
- Code blocks, screenshots, or specific numbers wherever possible
- Conclusion — what changed for you, what the reader should try
- CTA — relevant link, follow, newsletter
SEO is secondary for build-in-public blogs. Voice and specificity win.
Short Email (newsletter, ≤300 words)
- Subject line — provide 2-3 options. Curiosity > clickbait.
- Preview text — extends the subject, doesn't repeat it
- One main idea, treated like a long-form tweet
- Sign-off consistent with brand voice
Voice Defaults (when no brand voice configured)
If the user hasn't defined their voice, apply these defaults — they match indie-hacker / build-in-public conventions:
- First person, present tense for what's happening now
- Past tense for stories — "I shipped X and Y broke" beats "shipping X"
- Specific numbers over vague claims: "47 users" not "some users"
- Name the thing — products, tools, libraries by name
- Failure stories outperform success stories — write the honest version
- Dry humor lands; forced humor doesn't. Skip if unsure.
- Never use these words: leverage, unlock, game-changer, revolutionize, seamless, holistic, robust, cutting-edge, synergy
- Never start a post with "Just" or "So" (filler openers)
Workflow
- Confirm channel and topic. Ask if either is unclear. Do not guess.
- Pull context. Check recent git activity, recent posts (via
xurl if relevant), notes the user references.
- Pick the hook angle. State which hook pattern you're using and why (1 line, not a paragraph).
- Draft. Output 2-3 variants for short content, one full draft for long content.
- Annotate. Brief note on which voice rules were applied, any guesses you made.
- Offer next step. "Want me to run this through brand-review-content?" or "Adapt for LinkedIn?" or "Tighten one of these variants?"
Output Format
## Draft: [Channel] — [Topic]
**Hook angle:** [which pattern, 1 line]
---
[Variant 1 or full draft]
---
[Variant 2 if short content]
---
[Variant 3 if short content]
---
**Voice notes:** [what was applied, e.g. "specific numbers, failure framing, no corporate words"]
**Open questions:** [anything you guessed and want the user to confirm, e.g. "I assumed your goal was engagement — if it's lead-gen, the CTA should change"]
Pitfalls
Don't draft without a trigger. Generic "thoughts on building in public" posts are filler. If the user can't name what happened today, push back: "What specifically shipped or broke? That's the post."
Don't auto-add hashtags. Most indie-hacker accounts under 10K followers get nothing from hashtags. Only add if user asks or the post genuinely fits a tagged community.
Don't manufacture stakes. "This changed everything" when it didn't is the loudest tell of AI-generated content. If the impact was small, say it was small.
Don't write threads when a single tweet works. Most build-in-public updates are single tweets. Reserve threads for genuine deep-dives.
Don't write in someone else's voice. If the user has 50+ existing posts, reference 2-3 of them in your draft logic. If they don't, draft conservatively and ask for feedback to calibrate.
Don't end with "DM me if interested" unless that's the real ask. Most CTAs are filler. Cut them.
Source
Adapted from Anthropic's knowledge-work-plugins (Apache-2.0). See references/anthropic-source-notes.md for origin repo, adaptation decisions, and further mining candidates.
Related Skills
build-in-public — strategy layer (cadence, X Premium, VPS, cron setup)
brand-review-content — run drafts through voice/quality check before posting
reply-research — drafting replies to other accounts (different workflow)
xurl — actually posting to X
1---
2name: draft-post
3description: Draft a single piece of social content — X post, X thread, LinkedIn post, blog post, or short email — with channel-specific structure, hook patterns, and brand-voice application. Use when the user asks to "write a tweet about X", "draft a thread on Y", "post about today's progress", or any other single-asset content request. Pairs with brand-review (check before posting) and build-in-public (strategy layer).
4version: 1.0.0
5author: senna
6triggers:
7 - "draft a post"
8 - "write a tweet"
9 - "draft a thread"
10 - "write a LinkedIn post"
11 - "post about"
12 - "blog post about"
13metadata:
14 source: https://skillmd.com/skills/theheavenlyd3mon/draft-post
15 hermes:
16 tags: [social-media, x, twitter, linkedin, content, writing, build-in-public]
17 related_skills: [build-in-public, brand-review-content, reply-research, xurl]
18---
19
20# Draft Post
21
22Generate a single content asset tailored to one channel. Strategic layer lives in `build-in-public` — this is the execution layer for "I need to ship one good post right now."
23
24## When to Use
25
26- User says "draft a tweet/thread/post about X"
27- A cron job from `build-in-public` is triggered and needs to produce the actual draft
28- User wants 2-3 variations of the same idea for different channels
29
30## When NOT to Use
31
32- Multi-asset campaign with a calendar → use `campaign-plan` (TBD) or just iterate this skill
33- Quality check on existing draft → use `brand-review-content`
34- Engagement reply to someone else's post → use `reply-research`
35
36## Inputs to Gather
37
38If any of these are unclear, ASK before drafting. Do not invent context.
39
401. **Channel** — X post, X thread, LinkedIn, blog, email. Default to X if unspecified.
412. **Topic/Trigger** — what shipped, what failed, what you're thinking about. Specific > abstract.
423. **Goal** — engagement, build-in-public transparency, lead-gen, education, or community signal.
434. **Length constraint** — auto-derived from channel but override if user asks.
445. **Voice** — if user has a configured brand voice (see `brand-review-content` skill for the framework), apply it. Otherwise default to: first-person, specific, no AI-isms, no corporate jargon, dry humor OK, never "leverage" / "game-changer" / "unlock".
45
46## Channel-Specific Structure
47
48### X Post (single tweet, ≤280 chars)
49
50Pattern:
51- **Hook in first 8 words** — concrete detail, surprising number, contrarian take, or sharp question
52- **One specific point** — never two. Threads exist for that reason.
53- **Optional CTA** — only if there's a real ask (link, reply prompt, follow). Don't manufacture one.
54
55Avoid:
56- Opening with "Just" or "So"
57- Generic engagement bait ("agree?", "thoughts?")
58- More than 1 emoji unless writing in a tone where 2-3 fit
59- Hashtags inside body text (max 1-2 at the end if needed)
60
61Output 2-3 variants when possible. Different hook angles, same point.
62
63### X Thread (5-10 tweets, optional 25k mode for power users)
64
65Structure:
661. **Hook tweet** — earn the click. Promise something specific.
672. **Setup** — 1 tweet of context for people who don't know you
683. **Body** — 3-7 tweets, one idea per tweet. Use line breaks aggressively.
694. **Reversal or insight** — the moment readers screenshot
705. **CTA** — soft. Follow, reply, link out. Match the goal.
71
72Pitfalls:
73- Don't start tweet 2 with "A thread 🧵" — the hook should make people want it
74- Don't number tweets unless the content is genuinely a list
75- End on substance, not "follow for more"
76
77### LinkedIn Post (150-300 words, paragraph breaks every 1-2 lines)
78
79Structure:
80- Hook line, then a blank line
81- Setup paragraph (2-3 short lines)
82- Substance (3-5 short paragraphs, one idea each)
83- Takeaway / insight
84- Optional question to drive comments
85
86Tone shift from X: more professional framing, less in-jokes, no shitposting. Same authenticity, dressed differently.
87
88### Blog Post (500-2000 words)
89
90Structure:
91- **Headline** — provide 2-3 options. Specific beats clever.
92- **Lede** — first paragraph promises what the reader gets and proves the writer earned the read
93- **3-5 H2 sections** with descriptive headings
94- **Code blocks, screenshots, or specific numbers** wherever possible
95- **Conclusion** — what changed for you, what the reader should try
96- **CTA** — relevant link, follow, newsletter
97
98SEO is secondary for build-in-public blogs. Voice and specificity win.
99
100### Short Email (newsletter, ≤300 words)
101
102- **Subject line** — provide 2-3 options. Curiosity > clickbait.
103- **Preview text** — extends the subject, doesn't repeat it
104- **One main idea**, treated like a long-form tweet
105- **Sign-off** consistent with brand voice
106
107## Voice Defaults (when no brand voice configured)
108
109If the user hasn't defined their voice, apply these defaults — they match indie-hacker / build-in-public conventions:
110
111- First person, present tense for what's happening now
112- Past tense for stories — "I shipped X and Y broke" beats "shipping X"
113- Specific numbers over vague claims: "47 users" not "some users"
114- Name the thing — products, tools, libraries by name
115- Failure stories outperform success stories — write the honest version
116- Dry humor lands; forced humor doesn't. Skip if unsure.
117- Never use these words: leverage, unlock, game-changer, revolutionize, seamless, holistic, robust, cutting-edge, synergy
118- Never start a post with "Just" or "So" (filler openers)
119
120## Workflow
121
1221. **Confirm channel and topic.** Ask if either is unclear. Do not guess.
1232. **Pull context.** Check recent git activity, recent posts (via `xurl` if relevant), notes the user references.
1243. **Pick the hook angle.** State which hook pattern you're using and why (1 line, not a paragraph).
1254. **Draft.** Output 2-3 variants for short content, one full draft for long content.
1265. **Annotate.** Brief note on which voice rules were applied, any guesses you made.
1276. **Offer next step.** "Want me to run this through brand-review-content?" or "Adapt for LinkedIn?" or "Tighten one of these variants?"
128
129## Output Format
130
131```
132## Draft: [Channel] — [Topic]
133
134**Hook angle:** [which pattern, 1 line]
135
136---
137
138[Variant 1 or full draft]
139
140---
141
142[Variant 2 if short content]
143
144---
145
146[Variant 3 if short content]
147
148---
149
150**Voice notes:** [what was applied, e.g. "specific numbers, failure framing, no corporate words"]
151**Open questions:** [anything you guessed and want the user to confirm, e.g. "I assumed your goal was engagement — if it's lead-gen, the CTA should change"]
152```
153
154## Pitfalls
155
1561. **Don't draft without a trigger.** Generic "thoughts on building in public" posts are filler. If the user can't name what happened today, push back: "What specifically shipped or broke? That's the post."
157
1582. **Don't auto-add hashtags.** Most indie-hacker accounts under 10K followers get nothing from hashtags. Only add if user asks or the post genuinely fits a tagged community.
159
1603. **Don't manufacture stakes.** "This changed everything" when it didn't is the loudest tell of AI-generated content. If the impact was small, say it was small.
161
1624. **Don't write threads when a single tweet works.** Most build-in-public updates are single tweets. Reserve threads for genuine deep-dives.
163
1645. **Don't write in someone else's voice.** If the user has 50+ existing posts, reference 2-3 of them in your draft logic. If they don't, draft conservatively and ask for feedback to calibrate.
165
1666. **Don't end with "DM me if interested" unless that's the real ask.** Most CTAs are filler. Cut them.
167
168## Source
169
170Adapted from Anthropic's knowledge-work-plugins (Apache-2.0). See `references/anthropic-source-notes.md` for origin repo, adaptation decisions, and further mining candidates.
171
172## Related Skills
173
174- `build-in-public` — strategy layer (cadence, X Premium, VPS, cron setup)
175- `brand-review-content` — run drafts through voice/quality check before posting
176- `reply-research` — drafting replies to other accounts (different workflow)
177- `xurl` — actually posting to X