The skill is distributed publicly and works with any MCP-compatible agent that supports skills (Claude Code, OpenClaw, Hermes Agent, and others). All Postzee work is driven through the MCP HTTP server — never through direct REST calls.
What’s new in v4.0.0
v4.0.0 simplifies the creative loop and tightens the agent’s discipline around the things that affect you directly: cost transparency, plan capability, your assets, and the way the agent talks to you.🎯 Render-direct workflow
The agent goes from approved script straight to a real, rendered carousel — no intermediate preview to maintain. You see the actual slides in the conversation. Iteration runs on the real slides via single-slide replace or append.
💰 Cost transparency, enforced
The agent never quotes a credit cost from memory. Every number it tells you came from a live pricing call in that same response. Carousel rendering, slide replacement, uploads, and posting are all credit-free — only AI image and video generation consume credits.
✅ Plan capability, proactive
The agent reads your plan capability at the start of the conversation. If you’re on a plan that can’t publish to social channels, it tells you upfront and offers the upgrade — never a surprise paywall after you invested time scripting.
📎 Your images, one step
Drop an image in the chat — attachment, link, paste — and the agent uploads it to Postzee for you in one go. No more “host this somewhere and paste the URL”, ever.
What it does
The skill makes your agent fluent in five disciplines at once:Creative Director
Crafts concepts that fit your goal, audience, and platform. Asks the right questions during the brief, proposes alternative angles, suggests hook variants.
Copywriter
Writes hooks, captions, and CTAs using proven frameworks (AIDA, PAS, BAB, FAB, 4Ps). 80+ hook templates across 10 categories.
Video Producer
Decomposes ideas into scenes, maintains character consistency, composes multi-clip narratives. ffmpeg recipes for shell-capable agents.
Social Media Manager
Knows each platform’s algorithm, format, and engagement patterns. Per-platform aspect ratios, captions limits, hashtag strategy.
Key features
🌍 Multi-language native
Your agent always responds in your language — Portuguese, English, Spanish, French, German, Japanese, or any other. Even though the skill is written in English, the agent translates everything to your language and adapts the tone culturally (PT-BR is warmer than EN-US, FR can be more formal in B2B, etc.).💳 Plan and credit awareness
Before generating anything, the agent reads your full account context — plan tier, credit balance, storage, channel status — and anticipates issues instead of hitting them mid-flow:- Free plan + wants to post? Generates the asset, then proposes the right paid plan with persuasive copy and offers file delivery as a fallback.
- Low credits? Shows the shortfall and recommends the matching credit pack — no surprise paywalls.
- Channels need reconnection? Tells you which one and how (reconnect vs re-enable — they’re different actions).
- Storage near full? Warns you before generation, so credits aren’t wasted on files that fail to save.
💰 Cost transparency
The agent always shows generation cost in credits before generating — never in dollars, never as a surprise. It uses the livePOSTZEE_ESTIMATE_GENERATION_COST tool to compute exact numbers from the current pricing — no guessing, no hardcoded values.
🎨 Image & Carousel mastery
Carousels and text-heavy single images go through a dedicated editorial pipeline: the agent composes the slides with pixel-perfect text and your typography, then Postzee renders them. AI image models render text poorly — this pipeline removes that constraint entirely. The skill ships a complete editorial methodology that produces work indistinguishable from what a top human editorial team would publish — strong angle selection, rigorous quality validation, and a visual system that earns the scroll. The user sees the proposal, approves, and the agent ships the final carousel ready to publish. Iteration after the carousel exists is surgical — change one slide without touching the others, append a new slide to extend a growing carousel.⚙️ Per-platform publish awareness
The skill knows each social network’s settings — Instagram story vs feed, TikTok privacy + duet + music controls, YouTube visibility + tags, LinkedIn carousel mode, Pinterest board target, and more. When you say “post that as a story” or “private upload to TikTok”, the agent routes the right setting automatically; you don’t have to spell out the platform mechanics.✅ Platform-aware publish guards
Postzee enforces each social network’s native carousel limits before dispatching to the platform API — no more burning 4 minutes of retries and 30+ media containers on a carousel the platform was always going to reject.
When a carousel exceeds the destination limit (e.g., 13 slides → Instagram), Postzee fails fast with a message translated to the user’s language — “Instagram accepts at most 10 slides per carousel. Your post has 13.” — instead of letting the social network’s API return a
fbtrace_id. The skill also caps slide count at brief time when the destination platform is known, so the user is corrected before the first render call.
🎬 Multi-scene video
For stories with 2+ connected scenes, the skill orchestrates character and scene consistency across multiple generations. Strategies are gated by what’s currently available in the MCP — the agent only proposes what works end-to-end.🎙️ Talking head decisions
When you ask for someone speaking specific text, the skill picks intelligently between HeyGen (perfect lip-sync, voice control, multi-language, long-form) and Sora 2 / Veo 3.1 (cinematic scenes with native audio) — and tells you upfront which one bills on Postzee credits and which uses HeyGen credits.🎞️ ffmpeg + Whisper composition
For shell-capable agents (OpenClaw, Hermes, Claude Code with shell access), the skill includes a complete cookbook: concat with crossfades, audio mixing with sidechain ducking, loudness normalization (-14 LUFS), aspect ratio conversion, Ken Burns, picture-in-picture. Plus Whisper subtitle workflows with 5 styles (standard, highlighted-word, single-word, karaoke, type-on).🔍 Trend awareness
For creative work, the agent runs a quick web search for current viral patterns before proposing concepts — without citing sources unless you ask. The result feels like working with someone who’s plugged into what’s actually trending right now.📅 Smart scheduling
The agent knows the best posting times per platform, in your timezone (read from your Postzee org settings) — and converts to UTC ISO when scheduling. No mental math.🧭 Smart model routing
You don’t have to know which model to use. Say “create a sunset photo for Instagram” and the agent picks the right family (text-rendering vs photoreal vs illustration), the right tier (fast / balanced / premium), and the right aspect ratio for the platform — automatically. Default tier is fast to keep automation costs predictable. Carousels go through the editorial pipeline and don’t need model selection at all. If you DO know what you want — say “using Nano Banana” or “with the highest quality” — the agent honors your override. The multilingual signal vocabulary (PT / EN / ES / FR / DE) is built in.🔄 Self version-check
Every new session, the agent checks whether your installed skill version matches the currently published one. If you’re outdated, it tells you once (in your language) with the upgrade command for your client — and keeps working with the version you have. No nagging.Architecture
The skill ships as an orchestrator file that’s loaded on every session, plus a set of on-demand reference files the orchestrator pulls in when it needs them (creative brief, model routing, video composition, etc.). This keeps the session context lean while giving the agent deep knowledge exactly when each part of the workflow needs it. The skill is published as an open archive distributed through GitHub Releases — you can read it end-to-end if you want to understand exactly what your agent is doing.Installation
Claude Code
gh skill install Zee-Labs/postzee-skillOpenClaw
clawhub install postzeeHermes Agent
hermes skills install https://github.com/Zee-Labs/postzee-skill --path skills/postzeeExample interactions
The skill shines when you let the agent run a real creative loop. A few examples:Quick generation
“Generate a professional product photo for Instagram.”The agent picks the right photoreal model, validates the cost in credits, enhances your prompt, generates, and asks if you want to publish.
Full carousel with brief
“Create a 10-slide LinkedIn carousel about productivity mistakes.”The agent runs an editorial workflow: it gathers the brief, drafts a strong angle, validates it, and proposes the final headline + caption for you to approve. Once approved, it ships the rendered carousel ready to publish. If you want to tweak slide 4 after the carousel exists, the agent updates just that slide — the other 9 keep their identity.
Editorial single-image post
“Create a hero image for Instagram announcing the new pricing.”Same editorial pipeline but for standalone single-image posts: the agent drafts the hook, proposes a layout treatment, composes the design, and ships the rendered image. Useful for text-heavy posts where the value is in the typography + composition rather than in a swipeable sequence.
Multi-scene story
“Make a 30-second video showing a transformation journey for Reels.”The agent storyboards 3-4 scenes with a locked character and scene bible, picks the strategy that’s currently driveable through MCP, validates costs upfront, generates each scene, and (with shell access) composes them with crossfades.
Cost-aware mode
“I want to generate 100 images for a campaign.”Before generating anything, the agent computes total cost in credits, compares it to your balance, and either proceeds (if you have enough) or proposes the right credit pack with persuasive copy — no surprise paywalls mid-batch.
What the skill never does
- ❌ Quote prices in dollars for AI generations (always credits —
$1 USD = 1,000 creditsis the display ratio) - ❌ Recommend a feature that’s not currently driveable through MCP
- ❌ Generate without showing you the cost first
- ❌ Skip the version check on a new session
- ❌ Auto-publish to a channel that’s disconnected or needs reauth
- ❌ Translate captions literally across languages — it adapts culturally instead
Version
The currently published skill version is reported by the MCP server on everyPOSTZEE_GET_CONTEXT call. Your installed version may differ if you haven’t updated lately — the skill itself will tell you.
Updating
Claude Code Setup
Install and configure the Postzee Skill in Claude Code.
OpenClaw Setup
Install and configure the Postzee Skill in OpenClaw.
Hermes Agent Setup
Install and configure the Postzee Skill in Hermes Agent.
Tools Reference
The 25 MCP tools the skill orchestrates.