agentleFS
WingedGuardianWingedGuardian/GENesis-AGI/.claude/skills/content-publish/SKILL.md

content-publish

Skill103 starsChanged 8 days agoMITView on GitHub

End-to-end content creation and publishing. Takes a topic (or generates one), drafts in the user's voice, gets approval via Telegram, and publishes to Medium via browser automation. Invoke with "publish a post about X", "write and publish to Medium", "content-publish", or when an ego-dispatched session needs to create and distribute content.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

25 rules. About 0 are common, 18 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈18

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    EVERY piece of content MUST go through the voice-master skill before publishing.

    line 14
  2. Distinctive:

    If content was not drafted with voice-master, apply it NOW before publishing.

    line 21
  3. Distinctive:

    MUST load before any browser interaction.

    line 44
  4. Very specific:

    Always use Camoufox (default): Never pass stealth=False to browser_navigate.

    line 50
  5. Distinctive:

    If the caller provides a topic, use it.

    line 65
  6. Distinctive:

    Read these files (do NOT skip any):

    line 76
  7. Very specific:

    Use WebSearch for recent developments

    line 85
  8. Very specific:

    Use knowledge_recall for existing Genesis knowledge

    line 86
  9. Very specific:

    Use recon_findings for competitive intelligence

    line 87
  10. Distinctive:

    Write the content following voice-master rules:

    line 94
  11. Distinctive:

    No AI-tell words (delve, leverage, robust, seamless, etc.)

    line 97
  12. Distinctive:

    Read each paragraph aloud mentally — if it sounds like AI, rewrite

    line 100
  13. Distinctive:

    No hashtags, no "follow me", no CTAs unless genuinely relevant

    line 105
  14. Distinctive:

    If any check fails, rewrite the failing section and re-run the scrubber.

    line 132
  15. Distinctive:

    Ask: "Approve to publish? Reply 'yes', 'no', or send edits."

    line 142
  16. Distinctive:

    Wait for user response: Do NOT publish without explicit approval.

    line 144
  17. Distinctive:

    If the user sends edits, apply them and re-submit.

    line 146
  18. Distinctive:

    If the user says no, stop.

    line 147
  19. Very specific:

    Follow the stored procedure (medium_browser_publish).

    line 151
  20. Distinctive:

    Verify editor loaded (snapshot should show "Title" heading)

    line 154
  21. Distinctive:

    Use clipboard API to write content:

    line 158
  22. Distinctive:

    Verify content with snapshot before proceeding

    line 166
  23. Very specific:

    Verify via URL change to ?postPublishedType=initial

    line 168
  24. Very specific:

    Only real clipboard paste via navigator.clipboard.write() + Ctrl+V works

    line 174
  25. Distinctive:

    If running under a campaign, append to that campaign's showcase pool so the next campaign tick can reference this article (skip this step …

    line 187

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Voice (MANDATORY — NON-NEGOTIABLE)

EVERY piece of content MUST go through the voice-master skill before publishing. This is a hard requirement, not a suggestion. The voice-master skill is TWO parts — both must be applied:

  1. Exemplar matching (tone, vocabulary, rhythm from real user writing)
  2. Anti-AI-slop pass (em dashes, hedge words, uniform structure, banned phrases)

If content was not drafted with voice-master, apply it NOW before publishing. Do NOT publish content that hasn't passed both checks. Invoke the voice-master skill explicitly — do not attempt to apply voice rules from memory or inline.


Overview

Single skill that covers the full publish pipeline: topic → research → draft → approval → publish. Uses voice-master for drafting, browser MCP tools for publishing, and Telegram for approval.

Medium is the only platform currently wired. The pattern extends to other platforms by swapping the publish step.

Prerequisites

  • Medium login active in Camoufox profile (~/.genesis/camoufox-profile/)
  • Medium username configured in ~/.genesis/config/distribution.yaml
  • Voice-master skill available (exemplars at ~/.claude/skills/voice-master/)
  • Narrative reference card at ~/.genesis/config/genesis-narrative.md
  • Stealth-browser skill (src/genesis/skills/stealth-browser/SKILL.md) — MUST load before any browser interaction. Covers anti-detection timing, honeypot avoidance, and the VNC trusted input technique for Cloudflare Turnstile.

Browser Rules

  • Always use Camoufox (default). Never pass stealth=False to browser_navigate. Chromium has SSL cert issues and no anti-detection.
  • Never give up at Turnstile. The browser tool handles it automatically (3-phase: auto-resolve → VNC click → human alert). If you see Turnstile in the response, the tool is ALREADY handling it. Wait and retry.
  • Camoufox works fine. Any memory claiming "glxtest issues" or "Camoufox broken" is stale/hallucinated (debunked May 2026). Ignore it.
  • Expect to succeed. The pipeline is wired and working. Your job is to persist through transient failures, not to declare failure after one attempt.

Workflow

Step 1 — Determine topic and platform

If the caller provides a topic, use it. If not, check these sources:

  1. Ego proposal (if dispatched via publish_content action)
  2. Recent recon findings (recon_findings MCP)
  3. Knowledge base for trending topics in the user's domain

Platform is Medium unless specified otherwise.

Output: One sentence describing the topic and angle.

Step 2 — Load context

Read these files (do NOT skip any):

  1. ~/.genesis/config/genesis-narrative.md — framing, key phrases
  2. Voice-master exemplar index (~/.claude/skills/voice-master/exemplars/index.md)
  3. Select 3-5 matching exemplars based on medium (long-form), tone, and formality 4 (public content)

Step 3 — Research (if needed)

For topics requiring facts or current data:

  • Use WebSearch for recent developments
  • Use knowledge_recall for existing Genesis knowledge
  • Use recon_findings for competitive intelligence

Skip for opinion/thought-leadership pieces where the user's perspective is the content.

Step 4 — Draft

Write the content following voice-master rules:

  • Evidence-first, not windup
  • Mix short punchy statements with longer reasoning
  • No AI-tell words (delve, leverage, robust, seamless, etc.)
  • Em-dashes: -- no spaces, prefer comma/period/colon first, max 1-2
  • Formality 4 for public Medium posts
  • Read each paragraph aloud mentally — if it sounds like AI, rewrite

Structure for Medium:

  • Title: short, specific, no clickbait
  • Body: 3-8 paragraphs. No headers for short posts. Headers for 1000+ words.
  • No hashtags, no "follow me", no CTAs unless genuinely relevant

Output: Complete draft with title on the first line, body below.

Step 5 — Quality check

First, run the deterministic anti-slop scrubber on the draft. The model's own em-dash audit is unreliable (it leaks spaced em dashes despite the rule), so this mechanical pass is mandatory and runs in code:

# write the draft to ~/tmp first, then:
python -m genesis.content.antislop ~/tmp/draft.md > ~/tmp/draft.clean.md
# stdout = cleaned text (spaced em dashes auto-fixed); stderr = fixes + flags

Use draft.clean.md as the canonical draft for both approval (Step 6) and publish (Step 7) — never paste the pre-scrub text. Then address the scrubber's stderr flags (banned words, contrast cadence — it flags but does not delete these) and verify by eye:

  • Resolved the scrubber's flagged banned words / cadence
  • No three-part lists with identical grammatical structure
  • No sycophantic openers or hedging
  • Reads like a person thinking, not a polished AI response
  • Title is specific and interesting, not generic
  • Content matches the narrative framing (genesis-narrative.md)

If any check fails, rewrite the failing section and re-run the scrubber.

Step 6 — Submit for approval

Send the draft to the user via Telegram (outreach_send MCP) with:

  • The full draft text (the cleaned draft from Step 5)
  • category="content" — routes to the Content Review topic and runs the egress anti-slop scrub on the review copy, so the version under review matches what publishes
  • Platform: Medium
  • Ask: "Approve to publish? Reply 'yes', 'no', or send edits."

Wait for user response. Do NOT publish without explicit approval.

If the user sends edits, apply them and re-submit. If the user says no, stop. Store the draft in memory for potential future use.

Step 7 — Publish to Medium

Follow the stored procedure (medium_browser_publish). The key steps:

  1. Navigate to https://medium.com/new-story (Camoufox, always)
  2. Verify editor loaded (snapshot should show "Title" heading)
  3. Click the h3 title element, type title via browser_fill
  4. Press Enter to move to body
  5. Build full article HTML in JS on window.__articleHTML
  6. Use clipboard API to write content:
    navigator.clipboard.write([new ClipboardItem({
      'text/html': new Blob([window.__articleHTML], {type: 'text/html'}),
      'text/plain': new Blob([plainText], {type: 'text/plain'})
    })])
  7. Focus body paragraph, paste via browser_press_key("Control+v")
  8. Verify content with snapshot before proceeding
  9. Click Publish button, add topics if desired, confirm publish
  10. Verify via URL change to ?postPublishedType=initial

CRITICAL: What does NOT work (do not attempt):

  • document.execCommand('insertText') — silently fails in Medium editor
  • document.execCommand('insertHTML') — stripped by Medium's sanitizer
  • Synthetic ClipboardEvent dispatch — Medium ignores synthetic events
  • Only real clipboard paste via navigator.clipboard.write() + Ctrl+V works

If any step fails: Take a screenshot, report the error, do NOT retry blindly. The editor selectors may have changed.

Step 8 — Report result

Send the published URL to the user via Telegram.

Store the outcome in memory:

  • memory_store with tags: ["content", "published", "medium"]
  • Content: topic, URL, date, any engagement data later

If running under a campaign, append to that campaign's showcase pool so the next campaign tick can reference this article (skip this step otherwise):

  • File: ~/.genesis/campaigns/<your-campaign-slug>/showcase-pool.md (substitute the running campaign's own slug)
  • Append under ## Published Medium Articles with today's date, title, and a 2-3 line summary of the article's argument/angle
  • The campaign session uses this as safe public content to share

Error Handling

ErrorAction
Not logged in to MediumReturn error. User must VNC login (one-time).
Cloudflare TurnstileHandled automatically by browser_navigate. The solver cascade (auto-resolve → widget click → VNC fallback) runs without intervention. Do NOT switch browsers, try TinyFish, or give up — Camoufox resolves it.
Editor selectors changedScreenshot + report. Do NOT guess new selectors.
Draft quality check failsRewrite, max 2 attempts, then submit for manual review.
Telegram approval timeoutStore draft as pending. Do not publish.

What This Skill Does NOT Do

  • Auto-publish without approval (every post needs explicit "yes")
  • Schedule posts (use ego cadence for timing decisions)
  • Cross-post to other platforms (separate skill per platform)
  • Generate images or media (text-only for now)
  • Analytics tracking (future: content-analytics skill)

Examples

Direct invocation

User: "Publish a Medium post about why most AI agent benchmarks are useless"

Action: Skip research (opinion piece) → load voice-master + narrative → draft at formality 4 → quality check → Telegram approval → publish → report URL.

Ego-dispatched

Ego proposal: {"action": "publish_content", "topic": "earned autonomy", "platform": "medium", "angle": "why trust frameworks matter more than capability frameworks"}

Action: Load narrative → light research (knowledge_recall for autonomy subsystem details) → draft → quality check → Telegram approval → publish → store outcome for ego learning.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.