stevekaplanai/agenthost/execute

Execute — From Draft to Published

You have a post. This skill runs the final checks before it goes live — voice, ICP, hook, and format — and outputs a clean, publish-ready version.

After posting, it logs the content to what-works.md so the system starts learning.

Phase 1: Load Context

Read ~/.claude/CLAUDE.md and its sub-files:

• brand-voice.md — voice check reference
• icp.md — audience alignment check
• offer.md — CTA and positioning check
• what-works.md — pattern context for comparison

If CLAUDE.md doesn't exist: "Your content brain isn't set up yet. Run /brain-builder — this check only works when I know your voice and audience."

Phase 2: Accept the Draft

The member pastes their post draft. Accept it as-is — don't rewrite without permission. Your job in Phase 3 is to evaluate, flag specific issues, and suggest targeted edits. Not to rewrite the whole thing.

If they paste without context, ask one question: "Which platform is this for?"

Phase 3: The Pre-Post Checklist

Run each check. Score as ✓ (pass) or ✗ (fix needed). If ✗, give the specific edit — not a vague note.

HOOK ☐ Does the first line create enough tension or curiosity to force a second line? ☐ Is it specific? (number, name, result, or concrete scene — not vague claims) ☐ Does it avoid starting with "I" followed by a boring setup? (Fine to start with "I" if the next word is a result or action) ☐ Is it short enough to work as the preview line before "...see more"?

VOICE ☐ Does this sound like them — not polished AI, not corporate, not someone else? ☐ Are there any phrases that feel off-brand? (Check against brand-voice.md "What to Avoid") ☐ Is the energy consistent throughout, or does it shift partway through?

ICP MATCH ☐ Would the intended avatar read this and feel like it was written for them? ☐ Does the post use their language — not industry jargon, not too advanced, not too basic? ☐ Is the problem or desire named in a way that lands?

SPECIFICITY ☐ Are there real numbers, real timelines, real results — not vague claims? ☐ Is the proof grounded in their actual experience (from offer.md)?

STRUCTURE ☐ Does the post have a clear arc? (Hook → setup → insight → proof or value → CTA) ☐ Is the ending strong, or does it trail off? ☐ Is there a CTA? Is it specific enough to act on?

FORMATTING (platform-specific) For LinkedIn: ☐ One sentence per line, blank lines between paragraphs? ☐ No markdown bold or headers in body? ☐ Hook is visible before "...see more"?

For TikTok script: ☐ Written for the ear — does it flow spoken? ☐ Hook line is under 10 words? ☐ Runs 45–90 seconds?

Phase 4: Output the Scorecard

Present the checklist results clearly.

PRE-POST SCORECARD

[List each check with ✓ or ✗]

Issues to fix: [If any ✗ items exist, list them with the specific edit needed. Be direct: "Line 2 is vague — replace 'better results' with your actual client outcome." Not: "Consider adding specificity."]

Verdict: [READY TO POST / FIX FIRST]

If READY TO POST:

Show the clean final version of the post (no preamble, just the post) and ask:

"Ready. Want me to post this now, schedule it, or just keep the draft?"

If FIX FIRST:

Show what needs changing and offer: "Want me to make these edits, or will you do it? Just say go and I'll fix them."

After fixes are confirmed, re-run the scorecard and show the clean final version.

Phase 5: Publish (Optional)

If the member wants to publish via Blotato:

Step 1 — Get account ID

Call blotato_list_accounts for their platform. Use the id returned.

Step 2 — Post options

Publish now:

{
  "accountId": "<id>",
  "platform": "<linkedin|tiktok|etc>",
  "text": "<clean post text>",
  "mediaUrls": []
}

Schedule:

{
  "accountId": "<id>",
  "platform": "<platform>",
  "text": "<clean post text>",
  "mediaUrls": [],
  "scheduledTime": "<ISO 8601 UTC datetime>"
}

Queue to next free slot:

{
  "accountId": "<id>",
  "platform": "<platform>",
  "text": "<clean post text>",
  "mediaUrls": [],
  "useNextFreeSlot": true
}

Step 3 — Confirm

• If live publicUrl returned: "Posted. [URL]"
• If postSubmissionId in-progress: call blotato_get_post_status after 10 seconds
• If scheduled: confirm the time

Phase 6: Log to what-works.md

After any post is published or confirmed ready, append an entry to ~/.claude/what-works.md:

---
 
**[Platform] — [Date]**
Hook type: [type used]
Topic: [1-line topic description]
Format: [format type]
Hook (first line): [first line of the post]
 
_[Leave performance stats blank — member fills these in after posting]_
Engagement: 
Notes: 

Tell them: "Logged to what-works.md. Come back after 48 hours and add the engagement stats — that's how the system learns what to repeat."