content-draft

Generate platform-specific content drafts from backlog items, vault notes, or free topics.

Arguments

Five input modes — modes 1–3 run the full Steps 1–7 flow; modes 4–5 are subcommands that bypass it.

1. No args — /content-draft — presents the backlog for selection
2. Note path — /content-draft 04_Knowledge/AI-Weekly/2026-W12 - AI Weekly Digest.md — drafts from that note
3. Free topic — /content-draft "Why generation is not creation" — drafts from a topic string
4. Revise — /content-draft revise 2026-04-21-engramai-launch — generates the next version of an existing draft (see Subcommand: revise)
5. Archive — /content-draft archive 2026-04-21-engramai-launch — moves a stale draft to 05_Content/Archived/ and flips the backlog state (see Subcommand: archive)

Backlog state conventions

Each line in 05_Content/Backlog.md uses one of three checkbox states:

• [ ] — queued, not yet drafted (or drafted but neither published nor archived)
• [x] — published. Set manually by the user after moving Drafts/<slug>/ → Published/<slug>/.
• [-] — archived. Set automatically by /content-draft archive <slug> (which also moves the folder).

/content-draft itself never sets [x] — publication is a deliberate human action.

Subcommand: revise

Iterate on an existing draft. Skips Steps 1–3 and the backlog update; reuses Step 4's format generators with the existing version + user feedback as input.

1. Resolve folder — Match <slug> against subdirectories of 05_Content/Drafts/. Accept either the bare slug (engramai-launch) or the full folder name (2026-04-21 - engramai-launch). If 0 matches, error with the available folders listed. If >1 match, ask which one.

2. Read existing files — Read every .md file in the resolved folder (e.g., linkedin.md, linkedin-v2.md, twitter.md). These are the inputs.

3. Pick format(s) — If multiple formats are present, ask which to revise (single, multi, or "all"). Default to the format with the highest existing version count.

4. Determine next version — For each chosen format, scan filenames matching <format>(-v\d+)?\.md. Treat <format>.md as v1. Next file is <format>-v<N+1>.md where N is the max existing version. Files like linkedin-final.md are user-renamed picks — ignore them for version counting.

5. Ask for feedback — Prompt: "What would you like to change in the next version?" Accept free-form input (tone shift, new angle, tighter hook, fix a fact, etc.).

6. Generate — Pass the latest existing version of that format + the user's feedback to the format-specific generator from Step 4 of the main flow (LinkedIn → linkedin-content skill; Twitter/article/newsletter → built-in templates; technical blog → technical-blog-writing skill). Pillar is inherited from the existing file's frontmatter — do not re-ask.

7. Write output(s) — Save to Drafts/<folder>/<format>-v<N+1>.md. Frontmatter mirrors v1 but date: is the revision date.

8. No backlog update — The → [[…/]] link from the v1 run is still valid. Skip Step 6.

9. Present — Show the new file path(s) inline (same format as Step 7).

Subcommand: archive

Move a stale draft from Drafts/ to Archived/ and flip its backlog state to [-].

1. Resolve folder — Same matching rules as revise.

2. Confirm — List the files inside Drafts/<folder>/, then ask: "Archive Drafts/<folder>/ → Archived/<folder>/ and mark backlog [-]? (y/n)". Wait for explicit confirmation.

3. Move — On confirmation, mv 05_Content/Drafts/<folder>/ 05_Content/Archived/<folder>/. If Archived/<folder>/ already exists, ask whether to skip, overwrite, or rename (suffix -2).

4. Update backlog — Locate the line in 05_Content/Backlog.md containing → [[<folder>/]] (the link appended at v1-creation in Step 6 of the main flow). Change its checkbox from [ ] to [-]. If no matching line is found, surface a warning and skip the backlog edit — the move already succeeded, don't fail the whole operation.

5. Report — Tell the user what moved and whether the backlog was updated. Mention that [x] on Backlog items remains a manual signal for publication.

Step 1 — Resolve source material

Mode 1: From backlog (no args)

Read 05_Content/Backlog.md. Collect all [ ] items from ## Queue and sort them by priority tier based on their from: origin tag:

Priority	Label	Origin tags
P1	Your thinking & reflections	from:thinking, from:daily
P2	Your summaries	from:meeting
P3	External content	from:newsletter, from:daily-github, from:ai-weekly-digest

Present items grouped by tier with continuous numbering:

**P1 — Your thinking & reflections**
1. **Founder narrative** · The question every VC asked... · [[2026-04-02]] · `from:daily`
2. **Personal reflection** · Why I stopped optimizing... · [[Deep Work Note]] · `from:thinking`

**P2 — Your summaries**
3. **AI observer** · Claude's context window changes... · [[Meeting Knowledge/...]] · `from:meeting`

**P3 — External content**
4. **Builder workflow** · Four agent frameworks... · [[GitHub Trending]] · `from:daily-github`

Within each tier, maintain original backlog order (FIFO). Skip empty tiers. Ask which item to draft. Read the selected item's [[source note]] link.

Mode 2: Direct note (path arg)

Read the specified note directly.

Mode 3: Free topic (quoted string)

Search the vault (Grep across 03_Thinking/, 04_Knowledge/, 01_Execution/) for notes related to the topic. If matches found, read the top 2-3 most relevant. If no matches, proceed with the topic string alone — the draft will be generated from the user's knowledge and the topic.

Expand context

After reading the primary source, look for linked context:

• If the source mentions a project → read the project note (02_Projects/[Project]/[Project].md)
• If the source is a meeting note → check for related decision notes or knowledge notes
• If the source has [[wiki-links]] → read up to 3 linked notes for context

Keep total context under ~2000 lines. Prioritize depth on the primary source over breadth.

Step 2 — Identify pillar

If the source came from a backlog item, use its pillar tag. Otherwise, infer:

Signal	Pillar
Startup decisions, fundraising, team, product pivots	Founder narrative
AI papers, model releases, industry trends, research	AI observer
Tools, automation, workflows, systems thinking	Builder workflow
Career, learning, identity, philosophy	Personal reflection

If unclear, ask the user.

Step 3 — Select output format(s)

Ask the user which format(s) to generate:

Format	How it's generated
LinkedIn post	Delegates to linkedin-content skill
Twitter/X thread	Built-in (see Thread Format below)
Non-technical article	Built-in, uses Voice Guide (see Article Format below)
Technical blog	Delegates to technical-blog-writing skill
Newsletter edition	Built-in (see Newsletter Format below)
All	Generates all applicable formats

Step 4 — Generate drafts

LinkedIn Post

Invoke the linkedin-content skill. Pass it:

• The assembled source material and context
• The pillar and content angle
• Let the skill handle hook formula, formatting, hashtags, and CTA

Twitter/X Thread Format

Generate a thread of 3-7 tweets:

🧵 1/ [Hook — the provocative claim or question that stops the scroll]

2/ [Setup — the context or conventional wisdom you're challenging]

3/ [Core insight — the non-obvious thing you learned/observed]

4/ [Evidence or example — concrete, specific, from your experience]

5/ [Implication — "here's what this means for..."]

6/ [Open question or CTA — invite engagement, don't close the loop]

Rules:

• Each tweet ≤280 characters
• Tweet 1 is everything — it must create tension or curiosity
• No hashtags in the thread body (optional in final tweet only)
• Use line breaks for readability within tweets
• Concrete > abstract. "We lost 40% of signups" > "we had retention issues"

Non-Technical Article Format

Read 05_Content/Voice Guide.md and load the ## Non-Technical Article profile.

If the profile exists, follow its patterns. If empty/TBD, use this default structure:

# [Title — specific, not clickbait]

[Opening: concrete external event or personal moment]

[Pivot: the deeper question this raises]

---

[Argument section 1: the conventional view and why it's incomplete]

[Argument section 2: the insight, grounded in experience or evidence]

---

[Honest self-audit: where your argument might be wrong]

[Closing: an unresolved question, not a neat conclusion]

Target: 1000-2500 words. Use bold for key distinctions, blockquotes for pivotal questions.

After generating, suggest the user add the final published version to 05_Content/Voice Guide.md to calibrate future drafts.

Technical Blog

Invoke the technical-blog-writing skill. Pass it:

• The source material and technical context
• The target audience level
• Let the skill handle structure, code examples, and formatting

Newsletter Format

# [Newsletter title — issue number or date]

> [1-2 sentence hook — why this edition matters]

## [Section 1 title]
[Key insight, 150-300 words]

## [Section 2 title]
[Key insight, 150-300 words]

## [Section 3 title]
[Key insight, 150-300 words]

---

**What I'm thinking about:** [Personal aside — 2-3 sentences connecting themes]

**Worth reading:** [1-2 external links with one-line commentary]

Target: 800-1500 words total. Designed for email readability — short paragraphs, clear headers, scannable.

Step 5 — Write outputs

Create the draft folder and write files:

05_Content/Drafts/YYYY-MM-DD - <slug>/
  linkedin.md      (if LinkedIn was selected)
  twitter.md       (if Twitter was selected)
  article.md       (if article or technical blog was selected)
  newsletter.md    (if newsletter was selected)

The <slug> is a URL-friendly version of the content hook (lowercase, hyphens, max 50 chars).

Each draft file includes frontmatter:

---
pillar: <pillar name>
source: <source note path>
format: <linkedin|twitter|article|newsletter>
status: draft
date: YYYY-MM-DD
---

Step 6 — Update backlog

If the source was a backlog item, update it in 05_Content/Backlog.md:

• Append · → [[YYYY-MM-DD - <slug>/]] to the item line

Do not change the checkbox. See Backlog state conventions above for the full lifecycle ([ ] → [x] is manual on publish; [ ] → [-] is automatic via /content-draft archive).

Step 7 — Present to user

Show the user what was generated:

• List the files created with their paths
• For short formats (LinkedIn, Twitter), show the full draft inline
• For long formats (article, newsletter), show the first ~500 chars and the file path
• Remind: "Edit these drafts to your satisfaction. To iterate: /content-draft revise <slug>. To give up on this draft: /content-draft archive <slug>. To publish: move the folder to Published/ and mark the backlog item [x]."