Post Ideas

post-ideas

End-to-end workflow for Justin's build-in-public content on LinkedIn, X, and Threads.

Required reading every invocation

Before doing anything else, read these two files in full:

1. /root/wsg-voice/voice.md — canonical voice spec. Anti-AI rules, banned vocabulary, length targets, hook patterns, output contract. Do not paraphrase or compress these rules at runtime.
2. /root/.claude/projects/-root/memory/feedback_no_twitch_retrospectives.md — hard rule against Twitch-anchored posts.

Step 1 — Research recent work

Pull the last 14 days of activity from these sources in parallel:

# Infra audit (single best source — CHANGELOG-grade entries)
cd /root/wsg-infra-audit && git log --since="14 days ago" --pretty=format:"%ad %s" --date=short
head -150 /root/wsg-infra-audit/CHANGELOG.md

# Per-project commits
for repo in /root/dealophant /root/my-glp-shot /root/.openclaw/workspace/daily/wsg-cp; do
  echo "=== $repo ==="
  git -C "$repo" log --since="14 days ago" --pretty=format:"%ad %s" --date=short 2>/dev/null
done

# Per-client recent work
for d in /root/wsg-client-projects/*/; do
  echo "=== $d ==="
  git -C "$d" log --since="14 days ago" --pretty=format:"%ad %s" --date=short 2>/dev/null | head -10
done

# Project state for context on current focus
head -120 /root/.claude/projects/-root/memory/PROJECT_STATE.md

If the omnisocials MCP is connected, also call mcp__omnisocials__list_posts (status=scheduled, then status=published) to avoid pitching topics already queued.

Step 2 — Cluster and map to archetypes

From the raw activity, identify 5-8 distinct post-worthy topics. Each must be anchored to a real artifact, number, or specific event. Map each to the rotation:

Client Wins: only use a real client name if Justin confirms permission. Otherwise generic ("a client", "a contractor in California").

Step 3 — Present slate, get confirmation

Show the candidate slate as a table or short list. For each candidate include:

• One-line summary
• Suggested archetype + day
• One sentence on why it works

Recommend a strongest pick or full week/two-week lineup. Wait for Justin to pick or edit before drafting.

Step 4 — Draft (when Justin confirms)

Per voice.md output contract, generate three variants per topic:

• Variant A — most direct telling, LinkedIn long-form (1,000-1,800 chars)
• Variant B — different archetype angle on the same topic, LinkedIn long-form
• Variant C — X-native, under 280 chars

For Threads: clone the X version. Threads supports 500 chars and X copy fits cleanly.

Show all drafts with char counts. Wait for Justin to pick variants (or "push all").

Step 5 — Push to OmniSocials

Channels: 679855_linkedin, 679855_x, 679855_threads.

Use mcp__omnisocials__create_post. Schedule patterns Justin uses:

• Default if no preference stated: create as draft (no scheduled_at)
• Common explicit pattern: Mon-Fri at 4am Pacific (11:00 UTC in PDT, 12:00 UTC in PST)
• Common variant: random between 3:00-4:15am Pacific (10:00-11:15 UTC PDT, 11:00-12:15 UTC PST) — pick one time per channel per day

After pushing, return a table of post IDs + scheduled times + OmniSocials URLs.

Step 6 — LinkedIn tag follow-up

For LinkedIn drafts that mention specific products / companies / people that should be @ mentions: the API publishes them as plain text. Remind Justin to open each LinkedIn draft in OmniSocials' composer once and retype the @ to convert plain text into real entity-mention chips. X and Threads @ handles auto-link from plain text and don't need this.

Hard rules

• No em dashes anywhere — use commas, periods, or rewrite the sentence
• No banned vocabulary: delve, leverage, unlock, harness, robust, seamless, streamline, transform (as verb), elevate, humbled, blessed, game-changer, no-brainer, 10x, "let that sink in"
• No "I noticed your post" surveillance language
• No fake or made-up quotes / dialogue
• No inflated percentages without absolute numbers
• No Twitch-anchored posts (see feedback memory)
• No filler closers; closes should be a specific question, a forward commitment, or a concrete punch line
• Every post anchored to a real artifact, number, or specific event from the actual work

Voice anchoring check

Before finalizing any draft, re-read it once asking: "could a human who actually shipped this thing have written this." If it sounds even slightly AI-flavored — parallel three-bullet structures, generic insight after generic insight, motivational closer — regenerate from a different opening.