social-media-manager

name	social-media-manager
description	Social calendar, branded Gemini images (brand-images/ + post-image.png), bundles, HypeEngine POST /posts with date+time (scheduled auto-publish, UUID idempotency); cron AM/PM; Drive for LinkedIn articles only.
metadata	{"clawdbot":{"emoji":"📱"},"openclaw":{"emoji":"📱"}}

social-media-manager

The parent social agent: owns consistency (voice, CTA ladder, disclosure) and file-based handoffs to sub-skills—mirroring how qf-course-researcher orchestrates web → API → Notion with explicit steps.

Posting model (this workspace)

HypeEngine is already connected to LinkedIn and Twitter/X. For approved feed slots (linkedin_feed, twitter_x), send each row once via hype-engine: Media API + POST /posts with date + time from APPROVAL.md—HypeEngine schedules and auto-publishes (no separate “publish now” call). See hype-engine → Scheduling — no separate “publish now”.

Pipeline order (how teams use it):

1. First — this skill + HypeEngine: calendar → writers → auto-image-generation (per slot, brand-images/ / BRAND_IMAGES_DIR) → bundles → APPROVAL.md → hype-engine: upload post-image.png / article-hero.png, POST /posts scheduled (record HypeEngine post UUID; never duplicate the same Post ID).
2. Last — LinkedIn articles: linkedin-article-writer produces article.md + article-hero.png + teaser.md. Interns publish the article body (composer / Drive). teaser.md feed row: same HypeEngine flow (scheduled POST /posts with hero image when approved).

Operator summary (canonical playbook)

1. Batch write (~2 weeks): Using calendar.md, x-post-writer and social-content-writer (and linkedin-article-writer for article rows) generate copy in advance; then run auto-image-generation per slot so every twitter_x / linkedin_feed / article row has post-image.png or article-hero.png before bundling.
2. Approve: Fill APPROVAL.md with Approved Y/N and go-live datetime matching each row’s Local time (e.g. morning 09:00, evening 18:00).
3. Cron (twice daily): Two jobs (AM + PM) are not duplicates of each other: each must publish only rows whose calendar.md Local time matches that window (e.g. 09:00 vs 18:00) and APPROVAL.md has HypeEngine post UUID empty. See hype-engine → Idempotency.
4. Trends: Before each publish run, optionally refresh or read social-trend-monitor and merge 0–2 vetted trending hashtags into X and LinkedIn feed copy (see below).
5. LinkedIn articles (last in the chain): linkedin-article-writer on a separate cadence (~every 2 days). Interns post the article body (LinkedIn UI + optional Drive 04-articles/ per INTEGRATIONS.md). HypeEngine publishes teaser.md to the LinkedIn feed when approved—same connected account as other feed posts; it does not replace intern posting of the long-form article.

OpenClaw: Cron documentation · jobs persist in ~/.openclaw/cron/jobs.json. Finish skill setup first, then add cron so execution does not depend on manual chat triggers.

Prerequisites

• Read USER.md, SOUL.md, AGENTS.md (group vs main session rules).
• Google Drive (optional — articles + intake refs only): If ~/.config/social/drive_folder_id or SOCIAL_POSTING_DRIVE_FOLDER_ID is set (see workspace/INTEGRATIONS.md → Google Drive — LinkedIn articles only), then:
  • Read: At intake, you may pull reference files the team placed there (guidelines, notes) and summarize into 00-intake.md with Drive paths.
  • Do not use this folder as the default export for Twitter/LinkedIn feed posts—those are pushed via hype-engine from post-bundle.md after approval.
  • Articles: linkedin-article-writer may export long-form copy to 04-articles/ for interns; link in publish-handoff.md / README-handoff.md. Routine social slots are not mirrored to Drive as “ready-to-post” Docs.
• Default output root:

  workspace/drafts/social/<YYYY-MM-DD>-<campaign-slug>/
  

• Publish policy: default draft-only; auto-post only if human has documented allowlist in workspace (e.g. TOOLS.md or USER.md).

Cadence: two-week batch + twice daily (AM / PM)

• Planning: Build calendar.md for the next ~14 days from marketer + social-content-planning.
• Writing + images: social-content-writer / x-post-writer / linkedin-article-writer fill post-body.md, teaser.md, article.md; then auto-image-generation (Gemini, already configured) produces posts/<post-id>/post-image.png or workspace/drafts/linkedin/.../article-hero.png—required for this workspace unless a row is explicitly text-only in 00-intake.md.
• Calendar column: add Local time (or Slot) per row, e.g. 09:00 and 18:00 for morning and evening posts. hype-engine date + time must match APPROVAL.md.
• At publish time: read the approved post-bundle.md for that slot only—do not improvise a different topic.

Trending hashtags before X / LinkedIn feed posts

• When the human wants live trends in the copy: run or refresh social-trend-monitor (or read the latest file under workspace/drafts/social/trends/).
• Merge rule: add 0–2 hashtags or phrases that pass brand fit and risk from the trend brief—no spam-stacking; no unrelated viral tags.
• Record in post-bundle.md a line Trend source: + file path and which tags were added so publishes are auditable.

Credentials & API (qf-style)

• Draft-only: No secrets; all bundles under workspace/drafts/social/... as above.
• Twitter/X + LinkedIn feed (live post): HypeEngine is already connected for these channels. When APPROVAL.md allows it, use hype-engine: Accounts API → Posts API with the correct account UUIDs (~/.config/hype-engine/*) to push the approved bundle—no assumption that OAuth setup happens inside this skill. Do not use raw LinkedIn or X API from this workflow unless HypeEngine is unavailable and TOOLS.md / USER.md explicitly allows a fallback.
• Other live publish / analytics: OpenClaw channels in ~/.openclaw/openclaw.json when configured (Telegram, Discord, etc.) where the team uses them.
• Per-platform API keys (Meta, Google Ads, TikTok Marketing, Reddit, direct LinkedIn if fallback): use workspace/INTEGRATIONS.md. Sub-skills (adverts-creator, …) spell out which service each needs.

High-level Workflow

1. Intake (00-intake.md)

   • Capture: goals, KPIs, platforms, blacklist topics, legal/compliance (region, disclaimers), deadlines, asset links.
   • Images: default every scheduled twitter_x / linkedin_feed / article row gets auto-image-generation. List text-only post IDs here if any row must ship without an image.
   • Visual consistency: optional campaign visual-dna.md (illustration style, banned motifs). Product logo & colors: always from workspace/brand-images/ or BRAND_IMAGES_DIR (e.g. /home/admin/.openclaw/workspace/brand-images)—see auto-image-generation → Brand kit and workspace/brand-images/README.md.
   • Optional: path to marketer-agent output (workspace/drafts/marketing/.../README-handoff.md or 06-channel-plan.md) to align pillars and CTAs.

2. Trend pass (optional)

   • Invoke or simulate social-trend-monitor → save trends-in.md reference in folder.

3. Calendar (calendar.md)

   • Produce or merge output from social-content-planning:
     • Columns: Date | Local time (e.g. 09:00 / 18:00 for twice daily) | Platform | Format | Topic slug | CTA type | Asset needs | Owner | Post id | Draft path | Image path (TBD → filled) — expect posts/<post-id>/post-image.png or LinkedIn folder article-hero.png.
   • Calendar rows are schedule + intent (“shadows”); they are not a substitute for full post copy.

4. Per-slot production loop (order is mandatory) For each calendar row:

   • 4a — Publish-ready copy first: Use the one writer that matches the slot (writers include hashtags / thread layout in-file by default—no separate caption step unless the human opts in):

     • LinkedIn article (long-form): linkedin-article-writer only → Draft path → workspace/drafts/linkedin/... (article.md + teaser.md + article-hero.png per that skill).
     • X only: x-post-writer → posts/<post-id>/post-body.md.
     • Short LinkedIn feed, Reddit, Facebook, mixed calendar: social-content-writer → posts/<post-id>/post-body.md (or platform-specific notes inside file).
     • Fallback: agency-marketing persona if no dedicated skill fits—still require post-body.md with publish-ready sections.

   • 4b — Image (required unless intake marks row text-only): Run auto-image-generation immediately after 4a for that slot—load that skill fully: STYLE LOCK must lead prompt-master.txt, benchmark negatives included, post-render visual QA (no sci‑fi/holographic/anime drift for IT topics). Write posts/<post-id>/post-image.png (+ image-alt.txt) or article-hero.png per auto-image-generation → Chaining. If Gemini refuses, document in bundle under ## Image as STATUS: missing — reason and escalate in APPROVAL.md.

   • 4c — Optional caption polish: Only if the human asks: social-caption-writer → captions/<date>-<platform>-<slug>.md (otherwise skip).

   • 4d — Video (optional): auto-video-generation per assets-needed.md when applicable (separate from the static image default).

   • 4e — Bundle: posts/<post-id>/post-bundle.md for every calendar row with a Post id (including LinkedIn teaser rows): final body + hashtags + suggested time + disclosure line + ## Image block:

     ## Image
     - File: posts/<post-id>/post-image.png  (or path to article-hero.png)
     - Alt: <from image-alt.txt or article>
     - HypeEngine: upload via Media API → save MEDIA_UUID in this file after upload
     

     Sources: post-body.md, teaser.md, optional caption from 4c.

5. Engagement queue

   • If community work requested: social-community-engagement outputs under replies/.

6. Analytics loop

   • When metrics provided: social-analytics → analytics/<period>-summary.md; feed next calendar adjustments.

7. Approval package (APPROVAL.md)

   • Table: Post ID | Platform | Preview link or file path | Image path | Approved Y/N | Publisher | Go-live datetime | HypeEngine post UUID (empty until published—never re-post if filled) | Media UUID (optional).

8. HypeEngine handoff (X + LinkedIn feed only, after approval)

   • For each row where Platform is twitter_x or linkedin_feed and Approved Y/N is yes: if HypeEngine post UUID is already set → skip (idempotent). Else hype-engine: upload media if needed, then POST /posts once with date + time from APPROVAL.md—HypeEngine queues and auto-publishes at that time (no separate “publish now” call). Record post UUID + media UUID in APPROVAL.md. If no image and row was not text-only, do not send without human OK. Trends: merge before POST. See hype-engine → Scheduling — no separate “publish now” and Idempotency.
   • LinkedIn long-form articles: not published as the article via HypeEngine—interns post article.md (Drive / LinkedIn composer). HypeEngine only handles the teaser.md feed promo when that calendar row is approved. See hype-engine → LinkedIn: feed post vs long-form article.
   • Other platforms: unchanged (drafts only, or channels/APIs per INTEGRATIONS.md / TOOLS.md).

9. Close-out message to user

   • List exact paths created; state nothing was sent to HypeEngine unless step 8 ran (scheduled posts still “go live” later via HypeEngine, not via a second agent step).

10. Scheduling

• Weekly campaigns: one folder per week slug.
• Always-on: date-prefix daily subfolders day-YYYY-MM-DD/.

Sub-skills (explicit contracts)

Skill	Delivers
marketer-agent (optional upstream)	ICP, positioning, channel plan, briefs/ for social
social-content-planning	calendar.md skeleton
linkedin-article-writer	LinkedIn articles + teaser.md (not short feed posts)
x-post-writer	X-only post-body.md (publish-ready, hashtags in-file)
social-content-writer	Multi-platform short posts post-body.md (LinkedIn feed, X, Reddit, etc.)
agency-marketing	Fallback personas into posts/<id>/post-body.md
social-caption-writer	Optional second pass on copy + hashtags
social-community-engagement	Reply drafts
social-analytics	Readout markdown
social-trend-monitor	Trend brief file
auto-image-generation	post-image.png / article-hero.png + alt text for every slot (Gemini)
hype-engine	Media upload + live Twitter/X and LinkedIn feed posts after approval (Accounts + Media + Posts API)

Job chain & “when done → next” (pipeline-state.md)

For multi-step automation, keep a single status file in the campaign folder (same directory as calendar.md):

workspace/drafts/social/<campaign>/pipeline-state.md

Template (copy and keep updated):

Step	Phase	Status	Gate (required artifact)	Next skill to run
1	strategy	pending | complete	workspace/drafts/marketing/.../README-handoff.md	social-media-manager (intake + calendar)
2	calendar	pending | complete	calendar.md with Post id + Draft path columns	linkedin-article-writer / x-post-writer / social-content-writer (per row)
3	post_bodies	pending | complete	Every slot: post-body.md or LinkedIn teaser.md + article.md and matching post-image.png or article-hero.png (+ image-alt.txt where applicable)	auto-image-generation (if images missing) → social-media-manager (bundles)—or optional social-caption-writer
4	bundles	pending | complete	posts/<id>/post-bundle.md + APPROVAL.md table filled for review	Human approval
5	publish	pending | complete	Approved rows in APPROVAL.md with empty HypeEngine post UUID	hype-engine (POST /posts + date/time; record UUID)

Rules

• Whoever finishes a phase updates pipeline-state.md: set Status to complete, set Last updated (ISO date) in a line under the table, and optionally add NEXT_PROMPT for the operator (one sentence: what to invoke next).
• Immediate focus when the calendar already exists: set step 2 → complete; run step 3 until every slot has copy and post-image.png / article-hero.png (unless text-only flagged in intake); then step 4 (bundles)—skip caption writer unless requested.
• Host automation: use OpenClaw cron (Gateway scheduler) so jobs persist across restarts. Official docs: Scheduled tasks (cron). Jobs are stored at ~/.openclaw/cron/jobs.json. For what counts as “auto handoff” (cron vs webhooks vs custom sessions), see multi-agent-orchestrator → Automatic handoffs — what OpenClaw can (and cannot) do.
• Cron jobs (marketing workflow) — full spec below. Replace CAMPAIGN_FOLDER and TIMEZONE in CLI --message / --tz before running.

1) Twice daily — publish only (HypeEngine)

Does not run marketer-agent, writers, or auto-image-generation. Only ships already-approved rows with post-bundle.md and images ready.

• Today in TIMEZONE; row = twitter_x or linkedin_feed, Local time matches job (e.g. 09:00 vs 18:00), Approved Y, HypeEngine post UUID empty.
• hype-engine: Media upload → one POST /posts per row with date/time from APPROVAL.md (scheduled auto-publish; no publish-now). Record UUIDs in APPROVAL.md.
• Trends: If USER.md allows, read latest workspace/drafts/social/trends/ file; merge 0–2 brand-safe hashtags into HTML before POST; note Trend source:.
• Do not invent new posts or draft new post-body.md.

2) Weekly (optional) — continue pipeline through bundles — no HypeEngine

Read workspace/drafts/social/CAMPAIGN_FOLDER/pipeline-state.md, run next incomplete phase among calendar / post_bodies / bundles, update pipeline-state.md. Skills: social-media-manager, social-content-planning, x-post-writer, social-content-writer, linkedin-article-writer, auto-image-generation (brand-images/), bundles + APPROVAL prep. Do not POST /posts here. Optional social-trend-monitor for calendar hints. strategy phase usually needs marketer-agent + Drive—run when pending or skip if human owns it.

3) Every 48h (optional) — LinkedIn article draft

linkedin-article-writer once (marketer + CAMPAIGN_FOLDER calendar themes), article-hero.png, workspace/drafts/linkedin/, publish-handoff.md if Drive.

CLI examples (openclaw cron add)

# 1a) Morning publish — set TIMEZONE and replace CAMPAIGN_FOLDER in --message and --tz
openclaw cron add \
  --name "Social AM publish" \
  --cron "0 9 * * *" \
  --tz "TIMEZONE" \
  --session isolated \
  --message "Open workspace/drafts/social/CAMPAIGN_FOLDER/calendar.md and APPROVAL.md. Today in TIMEZONE only. Rows: platform twitter_x or linkedin_feed, Local time 09:00, Approved Y, HypeEngine post UUID empty. hype-engine only: upload media from bundle, POST /posts once per row with date+time from APPROVAL, record UUIDs in APPROVAL. Skip if UUID set. Do not write new posts. If USER.md allows: read latest workspace/drafts/social/trends/ file, merge 0-2 brand-safe hashtags into body HTML before POST."

openclaw cron add \
  --name "Social PM publish" \
  --cron "0 18 * * *" \
  --tz "TIMEZONE" \
  --session isolated \
  --message "Open workspace/drafts/social/CAMPAIGN_FOLDER/calendar.md and APPROVAL.md. Today in TIMEZONE only. Rows: twitter_x or linkedin_feed, Local time 18:00, Approved Y, HypeEngine post UUID empty. Same hype-engine rules: media, one POST /posts per row from APPROVAL, record UUIDs. Skip if UUID set. Do not process 09:00 rows."

# 2) Weekly pipeline — example Monday 07:00 (adjust cron day/hour)
openclaw cron add \
  --name "Social pipeline continue" \
  --cron "0 7 * * 1" \
  --tz "TIMEZONE" \
  --session isolated \
  --message "Read workspace/drafts/social/CAMPAIGN_FOLDER/pipeline-state.md. Execute next incomplete phase only: social-media-manager, social-content-planning, writers, auto-image-generation with brand-images, bundles per skills. Do NOT run hype-engine or POST /posts. Update pipeline-state.md. Optional social-trend-monitor for calendar planning only."

# 3) LinkedIn article every 48h
openclaw cron add \
  --name "LinkedIn article draft every 2d" \
  --every "48h" \
  --session isolated \
  --message "Run linkedin-article-writer once: article from marketer outputs + workspace/drafts/social/CAMPAIGN_FOLDER/calendar.md themes; article-hero.png via auto-image-generation; save under workspace/drafts/linkedin/; update publish-handoff.md if Google Drive in use."

openclaw cron list

Replace literal TIMEZONE and CAMPAIGN_FOLDER in each command (e.g. America/New_York, 2026-04-acme).

OpenClaw GUI prompt (paste to agent)

Read workspace/skills/social-media-manager/SKILL.md section "Cron jobs (marketing workflow)". Run openclaw cron add on this host with my values:

CAMPAIGN_FOLDER = [paste folder name under workspace/drafts/social/]
TIMEZONE = [e.g. America/New_York]

Create four jobs: (1) Social AM publish 0 9 * * * (2) Social PM publish 0 18 * * * (3) Social pipeline continue weekly Monday 07:00 — or ask me for day/hour (4) LinkedIn article draft every 2d --every 48h.

Use the exact --message strings from the skill, with CAMPAIGN_FOLDER and TIMEZONE substituted. Session isolated. Then openclaw cron list and paste output. Do not delete unrelated jobs.

• Main session (lighter): openclaw cron add --session main --system-event "..." --wake now — good for reminders; isolated is better for end-to-end post + API work.
• Full DAG: multi-agent-orchestrator/chain-templates/social-feed-pipeline.md

Outputs (required per campaign folder)

• 00-intake.md, calendar.md, APPROVAL.md, README-handoff.md
• pipeline-state.md — required when running the marketer→publish chain (so the next session knows what to do).
• Optional: publish-handoff.md / article links only — long-form LinkedIn articles handed to interns via Drive (04-articles/); feed posts do not require a Drive row.

Agent Checklist

• Intake captured goals, platforms, compliance, deadlines.
• Every post has a unique post id and file path.
• post-body.md (or linked teaser.md) exists with publish-ready copy before bundling (or explicit defer).
• post-image.png / article-hero.png exists for each non-text-only row, or APPROVAL.md notes an explicit exception.
• If social-caption-writer was skipped: post-bundle.md still reflects final copy from the writer files.
• If caption polish was used: captions/*.md merged; else note “caption pass skipped.”
• Disclosure / AI-assisted / sponsored lines present when applicable.
• APPROVAL.md present; human sign-off assumed empty until filled.
• User told no live publish unless tools + approval satisfied; X/LinkedIn use hype-engine when posting live.
• Sub-skill outputs referenced by path, not vague “see above.”
• If using trends: Trend source: line in bundles; hashtags brand-safe and sparse.
• Cron (if used): user has openclaw cron list and Gateway running; timezone matches Local time on calendar; cron message filters by Local time + empty HypeEngine post UUID (see hype-engine idempotency).
• No duplicate HypeEngine publishes: HypeEngine post UUID filled in APPROVAL.md after each successful post.
• If social/drive_folder_id (or env) is set: used for article handoff (04-articles/) per INTEGRATIONS.md, not for exporting every feed post-bundle to Drive.