Social media manager

social-media-manager — run the accounts, make the ads, post to Reddit

Turn "run my socials" into a standing operation: Claude plans content, writes copy, generates ad creatives, posts (including to subreddits), schedules, and engages — with every public or money-spending action behind a configurable approval gate (AUTONOMY.md). Emphasis: creating ads and posting to communities (Reddit/subreddits).

No social MCP exists in this environment. X/Reddit/LinkedIn/TikTok have no MCP here, so posting runs through (a) the platform's own API with app credentials the user provides, (b) n8n nodes for scheduling/posting, or (c) browser automation (claude-in-chrome / playwright) as a fallback for UI-only flows. Creatives come from the higgsfield skills. Pick per platform in platforms/* and wire what the user authorizes.

---

What it is

A multi-platform manager. Use the parts the goal needs; they compose.

• Platforms (platforms/*) — per-network posting, scheduling, ToS: reddit.md (deep), x-twitter.md, linkedin.md, instagram-facebook.md, tiktok.md.
• Ads (ads/*) — ad-creation.md (hooks, copy, creative gen, A/B), ad-platforms.md (Meta/Google/Reddit/TikTok campaign setup, targeting, tracking, launch).
• Playbooks (playbooks/*) — content-engine.md (plan→create→schedule→publish→engage→ measure), engagement.md (replies/DMs/community).
• account-creation.md — real-world playbook for creating accounts (browser-assisted; what's automatable vs needs-human; per-platform reality; IMAP verification codes; file-upload limits).
• copy-rules.md — how every post/ad must be written (name real platforms, no bare lists, lead with what it does, balanced messaging, tone). Read it before drafting any copy.
• AUTONOMY.md — how far Claude may act per action class. GUARDRAILS.md — hard rules.

Mandatory grill-questions (fold into the Definition of Ready)

• Goal — awareness, leads, sales, community, traffic? Sets platforms + metrics.
• Platforms — which accounts, and what's the priority order?
• Brand voice — tone, do/don'ts, banned topics, emoji use (default none — ask), examples.
• Accounts & access — for each platform: API app credentials/OAuth, or browser-session access, or "drafts only"? What can Claude actually reach? (→ environment readiness)
• Autonomy level — per action class (draft / schedule-to-queue / publish / spend). Default conservative; raise deliberately. (→ AUTONOMY.md)
• Cadence + volume — posts/week per platform; ad budget + ceiling.
• Reddit specifics — which subreddits, are you an established account there, what's the self-promo ratio you'll hold? (→ platforms/reddit.md)
• Compliance context — sponsored/#ad disclosure needs, regulated industry?, regions.

Autonomy tiers (summary — full in AUTONOMY.md)

1. Draft-only — write copy + generate creatives, nothing leaves.
2. Schedule-to-queue — schedule into a review queue / native scheduler; human approves.
3. Publish (gated) — auto-publish a whitelisted, low-risk class (e.g. a pre-approved content calendar) with logging.
4. Spend (always gated) — launching/raising ad budget ALWAYS needs explicit confirmation. Per platform + action class. Public/irreversible/spend actions never bypass the gate.

Project sub-agents to generate (into .claude/agents/)

• content-strategist (delegate-by-default) — turns goals into a content plan/calendar.
• copywriter (delegate-by-default) — platform-native copy (hooks, AIDA/PAS), per voice.
• ad-creative-generator (delegate-by-default) — drives the higgsfield skills for image/ video creatives + variants; scores with the virality predictor.
• reddit-poster — rules-aware posting to subreddits: checks subreddit rules/flair, holds the self-promo ratio, avoids karma/age gates and shadowban patterns.
• scheduler — queues/schedules across platforms; spaces posts; respects best-time windows.
• engagement-responder — drafts replies to comments/DMs; escalates sensitive ones.
• brand-safety-reviewer (adversarial — before any publish) — fails the action if a post risks ToS (automation/spam limits, fake engagement), missing #ad disclosure, Reddit self-promo/astroturf rules, brand-unsafe content, or skips the approval gate.

Tools / CLIs / MCP / skills to chain

Check in environment-readiness; offer install, never auto-install or invent credentials.

• higgsfield skills — higgsfield-generate (image/video), higgsfield-soul-id (consistent faces), higgsfield-product-photoshoot, higgsfield-marketplace-cards; higgsfield MCP generate_image / generate_video, virality_predictor (score video before posting), marketing studio. Primary creative engine.
• creative-media sub-skill — compose it for the full produce→publish media pipeline.
• build-business sub-skill — source of the offer/ICP/angles for ad copy.
• n8n MCP — scheduling + cross-posting automation; approval-gated publish workflows.
• claude-in-chrome / playwright — browser automation for platforms without API access (e.g. Reddit posting, LinkedIn) — a fallback, ToS-aware, human-gated.
• magic — generate any companion UI (a small dashboard/approval queue) if wanted.
• design-harvest + web-aeo — reference design + discoverability for linked content.
• Reddit: PRAW (Python Reddit API Wrapper) with a Reddit app (client id/secret + OAuth).
• Platform APIs (X, Meta Graph, LinkedIn, TikTok) — user provides app creds; keys in .env.

File / asset nudges (on top of the base project set)

• .env.example — every platform's app id/secret/token named (real .env gitignored).
• brand/voice.md — tone, do/don'ts, banned topics, examples; the single source of voice.
• calendar/ — the content calendar (planned → drafted → scheduled → published).
• ads/ — ad sets (copy + creative refs + targeting + budget) per campaign.
• assets/ — generated creatives (or references to where they live).
• queue/ — the approval queue: drafts awaiting human go before publish/spend.
• metrics.md — what's tracked per platform + a weekly review note.

Stack defaults & done-bar

Default: higgsfield for creatives + virality scoring; n8n for scheduling/cross-post; PRAW/browser for Reddit; platform APIs where the user grants them; a queue/ approval gate in front of every publish/spend. Reddit posting is rules-first (read each subreddit's rules).

Done-bar (verified, not reasoned): a configured account can produce a scheduled post and a complete ad set through the approval gate; Reddit posts respect each subreddit's rules + the self-promo ratio; the virality predictor scored any video before it queued; and nothing published or spent without passing the autonomy gate.

Guardrails (standing footer — enforced; full in GUARDRAILS.md)

• Nothing public or money-spending without the approval gate. Publishing and ad budget are gated by default.
• Platform ToS: respect automation/rate limits; no fake engagement, no bot networks, no vote manipulation, no mass-DM spam, no astroturfing.
• Reddit: follow each subreddit's rules + the ~90/10 self-promo guideline; participate as a real member, not a spammer. Never manipulate votes or sock-puppet.
• Disclosure: FTC #ad/sponsorship disclosure where required; honesty in claims.
• No emojis by default unless the brand voice explicitly calls for them (ask first).
• Commits under the user's own name only (Skryx-L-A); never add Claude as co-author.
• Never commit account secrets/tokens — .env only (gitignored).