Publishing Content Strategy

Create original posts, voice, strategy

Publishing Flow

Content in agent/outputs/{platform}/ is automatically posted by process-outputs.yml:

agent/outputs/x/tweet-20260203-001.txt  →  posted  →  agent/outputs/x/posted/tweet-20260203-001.txt

File Naming Convention

Use date-based naming: {type}-{YYYYMMDD}-{NNN}.txt

• •tweet-20260203-001.txt - first tweet created on Feb 3, 2026
• •tweet-20260203-002.txt - second tweet created same day

❌ Don't use: tweet-week2-001.txt, tweet-monday.txt (becomes meaningless after posting)

Posting Cadence

Current approach: One post per session/PR

Rationale:

• •Distributed posting = better engagement
• •Each post gets its own algorithm window
• •Avoids rate limits

Note: Research suggests 3-5 posts/day optimal. Test and adjust based on data.

Queue Management (Week 1 Learning)

Max 3 pending tweets per PR. The posting workflow processes ALL pending files at once.

Why: On Day 3, creating 16 tweets that all posted in one burst hit the 17-tweet rate limit and wasted algorithm engagement windows. Keeping the pending queue small ensures distributed posting.

Evidence: agent/memory/learnings/posting-cadence-strategy.md, agent/memory/learnings/2026-02-03-x-rate-limits.md

File Rules

Agent creates files. Workflow handles posting.

• •✅ Create new files in agent/outputs/{platform}/
• •✅ Read files in posted/ to check what was published
• •✅ Modify/rename own files before they're posted
• •❌ NEVER remove/modify files in posted/
• •❌ NEVER delete files (no rm permission)
• •❌ NEVER try to post (workflow does this)

Workflow responsibility: posting, moving to posted/, handling failures, cleanup.

Supported Formats

• •✅ Single tweets - tweet-YYYYMMDD-NNN.txt (limit set by X_MAX_TWEET_LENGTH var, default 25000 for Premium)
• •✅ Threads (--- separated) - thread-YYYYMMDD-NNN.txt

Thread format:

First tweet text here
---
Second tweet (reply to first)
---
Third tweet (reply to second)

Thread quota warning: 10-part thread = 10 tweets against rate limit.

Know Your Permissions

Read .github/workflows/agent-work.yml to see --allowed-tools. Don't attempt commands not in the list (e.g., rm is not allowed).

Value Rule: Never Mix Value Types

A post delivers value in one of two ways. Pick one per post. Never both.

Why not both? Mixing them dilutes each. The insight gets cut short to fit the promo. The promo feels forced because it's wedged into someone else's topic. The reader gets half an insight and half an ad.

❌ "Big news just dropped. [1 sentence]. Anyway here's my repo → [link]" (Insight cut short. Promo doesn't follow from the hook. Reader gets neither.)

✅ Content value: "Big news just dropped. Here's what it means, why it matters, and what most people miss." ✅ Outcome value: "I built X that solves Y. Here's the repo → [link]"

Growth Strategies

Build in Public (BIP)

Evaluate if current repo/project is BIP-worthy. If yes, determine cadence.

BIP evaluation criteria:

• •Is repo public?
• •Is work interesting/novel?
• •Are there learnings worth sharing?
• •Would audience find it valuable?

If BIP-worthy → post often. BIP thrives on frequency and consistency.

If BIP-worthy, posts can include:

• •What was done/learned this session
• •Conclusions or findings
• •Repo/PR link for proof
• •Metrics updates
• •Explain the idea/concept (for new followers)
• •How it started, where it is now
• •Repo promotion with context

What to share:

• •Progress and metrics (followers, engagement)
• •Learnings (what worked, what didn't)
• •Behind-the-scenes (how it works)
• •Failures and pivots (authenticity builds trust)
• •Skill development journey (what you're reading, studying, mastering)
• •"Today I learned" moments from reading top voices
• •Surprising findings from research deep-dives

Why BIP works:

• •Public repo = built-in proof
• •People root for underdogs
• •Vulnerability creates connection
• •The journey IS the content

3-Bucket Content Strategy

Balance three content types for maximum reach:

Missing any bucket limits audience growth.

Hook Engineering

First line determines if anyone reads. Engineer hooks, don't just write them.

Hook formula: Bold Statement + Tension + Credibility

❌ Weak: "I want to share my thoughts on growing on Twitter." ✅ Strong: "I went from 0 to 10,000 followers in 57 days without posting a single thread. Here's exactly how."

Hook patterns that work:

• •Specific numbers: "5 things I learned..."
• •Contrarian: "Most people are wrong about..."
• •Story opener: "Last week I failed publicly..."
• •Question: "Why do 90% of accounts never grow?"

Thread Strategy

Threads get 63% more impressions than single tweets.

Structure:

• •Hook (tweet 1) - Must stop the scroll
• •Value (tweets 2-4) - Deliver on the promise
• •CTA (final tweet) - Follow, share, or link

Rules:

• •3-5 tweets optimal. HARD MAX: 5 tweets per thread. (Week 1 evidence: a 10-part thread consumed 10/17 daily quota in one post)
• •Under 250 chars per tweet (under 200 better)
• •Cliffhanger every 1-2 tweets
• •Zero hashtags in main content
• •Use 1x/week for deeper content

Learning Journey as Content

The process of building expertise IS content. Share what you're reading and learning.

Why it works:

• •Shows curiosity and growth (people follow learners, not just experts)
• •Positions you as a curator - filtering signal from noise for your audience
• •Creates natural engagement: people share their own takes
• •Builds authority over time through visible knowledge accumulation

Content from reading/learning:

• •"Just read [author]'s take on [topic]. Key insight: [takeaway]. Here's why it matters..."
• •"3 things I learned this week about [domain]" (thread)
• •"I used to think X. After reading [source], I now think Y. Here's what changed..."
• •"[Author] nailed this: [paraphrased insight]. But I'd add..."
• •"Been deep-diving [topic] all week. The biggest surprise: [finding]"

Rules:

• •Always add your own angle - don't just summarize
• •Credit the source (builds goodwill + networking)
• •Connect learnings back to your domain/project
• •Show evolution: "I used to think... now I think..."

Questions as Content

Questions drive replies. Replies drive reach. Ask genuinely.

Why questions work:

• •Algorithm rewards replies heavily (reply-to-reply = 75x)
• •Low friction for audience to engage (everyone has an opinion)
• •Surfaces insights you can learn from
• •Builds community (people feel heard)

Question formats:

• •Opinion poll: "What's the biggest bottleneck in [domain] right now?"
• •This or that: "[Tool A] or [Tool B] for [use case]? And why?"
• •Genuine curiosity: "Has anyone solved [specific problem]? Here's what I've tried..."
• •Contrarian probe: "Hot take: [bold claim]. Change my mind."
• •Experience ask: "What's one thing you wish you knew before [doing X]?"
• •Prediction: "Where does [domain/technology] go in the next 12 months?"

Rules:

• •Ask questions you actually want answers to (authenticity shows)
• •Keep questions specific, not vague ("What do you think about AI?" = bad)
• •Engage with replies - questions without follow-up feel hollow
• •Mix in ~15-20% question posts for engagement balance
• •Use learnings from answers in future content (close the loop)

Algorithm Awareness

What X rewards (latest known)

What hurts reach

• •External links (algorithm downgrades)
• •Heavy hashtags
• •Posting and leaving (no engagement)

Content Voice

Frame as: human building products with autonomous tools (not "AI doing everything").

Use: creating, building, generating, exploring, shipping, launching Avoid: testing, experimenting, trying (passive/uncertain) Say: product, tool, solution (never "content")

✅ "Exploring vibe coding with autonomous agents to ship faster" ✅ "Building automated workflows - here's what's working" ❌ "I'm an AI agent, no human writes these tweets" ❌ "Testing if this works..."

The story is human innovation, not AI replacement.

Author Info

To reference/promote repo owner:

gh api users/{owner}

For links not in GitHub API, see ME.md. Never hardcode or guess links.

Promotional Content

~20% of posts include soft promotion:

• •This repo - link to the autonomous agent experiment
• •Author's profile - GitHub, LinkedIn, blog
• •Author's products - if mentioned in repo/profile

Examples:

• •"Building this in public → [repo link]"
• •"More on my approach → [blog/profile link]"

Keep natural, not salesy. Tie to value.

Promotional Templates (Week 1 Learning)

Only ~10% of Week 1 tweets included links. Target is 20%. Use these templates:

• •BIP update: "[Insight or milestone]. Building this in public → [repo link]"
• •Learning share: "[Key takeaway from session]. Following the whole journey: [repo link]"
• •Profile soft plug: "[Valuable observation]. More on how I build with AI → [profile link]"

Evidence: Week 1 retro showed 3/30 posted tweets had links vs. 6/30 target. Week 2 retro: dropped further to 4.3% (2/47 tweets). This is a critical gap.

Content Creation Checklist (Week 2 Learning)

Before committing any tweet, verify ALL items:

1. •Length: Check X_MAX_TWEET_LENGTH GitHub var for current limit (run gh variable list). Premium = 25,000 chars, Free = 280. Write as long as the content needs — but keep it concise and valuable (not padded).
2. •Link check: Does this session's batch include at least 1 tweet with a repo/profile link? Target 20% of all output.
3. •BIP balance: Is BIP content at least 25% of recent output? If not, make next tweet BIP.
4. •Category: What category is this? (Authority / Personality / Shareability). Avoid imbalance.
5. •Hook: Does the first line stop the scroll? Apply hook formula.

Evidence: Week 2 retro — 4 tweets skipped for over-length, promotional links dropped to 4.3%, BIP dropped from 40% to 23%.

Engagement-First Strategy (Week 2 Learning)

For accounts under 100 followers, engagement is more important than content.

5 followers after 176 tweets = content-only doesn't drive growth for small accounts. Discoverability is the bottleneck, not volume.

Session Allocation (< 100 followers)

• •50% of session time: Find and create replies to larger accounts in your niche
• •50% of session time: Create original content

Why Engagement First

• •Reply-to-reply = 75x algorithm multiplier
• •One good reply to a 50K-follower account = more visibility than 10 original tweets
• •Replies show up in other users' feeds, bringing profile visits
• •Established accounts may follow back or retweet

How to Execute

1. •During reading sessions, note recent posts from top voices that you can reply to
2. •Use web search to find tweet IDs: WebSearch: "site:x.com @handle {topic}"
3. •Create reply files using the commenting skill format
4. •Aim for 2-3 replies per session alongside 1-2 original tweets

Evidence: Week 2 retro — 0 replies in 2 weeks despite commenting skill existing. Follower growth: <1/day.

Commenting / Engagement

See @.claude/skills/commenting/SKILL.md for reply strategy, file format, and engagement approach.

Gist Fallback

When no platform integration exists:

gh gist create --public -f "filename.md" content.md
gh gist edit <gist-id> -f "filename.md" updated.md

Track gist URLs in state file under "## External Outputs".