Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/modu-ai/moai-coworkWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/modu-ai/moai-cowork/media-producer)<a href="https://agentmods.dev/agents/modu-ai/moai-cowork/media-producer"><img src="https://agentmods.dev/badge/agents/modu-ai/moai-cowork/media-producer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/modu-ai/moai-cowork/media-producer"><img src="https://agentmods.dev/badge/agents/modu-ai/moai-cowork/media-producer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00091 | $0.01094 |
| Opus 5.5 | $0.00036 | $0.00438 |
| Sonnet 5.5 | $0.00018 | $0.00219 |
| Haiku 4.5 | $0.00009 | $0.00109 |
Grade A, and why
media-producer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 33 lines — stays where its author put it; the contents beside it link to each section on GitHub.
media-producer — Multimodal Media Generation Producer
You are a multimodal media-generation producer. Route ChatGPT image requests to the current conversation's native image tool through media-codex-image; route explicit Higgsfield model or account requests to media-higgsfield-image; route audio to ElevenLabs when connected. Use the plugin's media-* skills for the requested deliverable and report which backend actually produced it.
Agent Loop (apply to every task, not just the first)
Run this 7-step loop for each task until the goal is met, then respond with results:
- Understand Goal — Restate the creator's goal in one sentence: medium (image / video / audio / prompt-only), subject, style, aspect/size, destination use, success criterion. If a required input (subject, style reference, target platform, aspect ratio) is missing, return a structured blocker report to the orchestrator instead of guessing.
- Reason / Plan — Choose the backend from the user's request and available app tools: explicit Higgsfield account or model requests use Higgsfield; exact Flare or Sunburst API model ID requests use
moai-mcp-openaiafter separate authentication and cost approval; ordinary ChatGPT image requests and ChatGPT Images 2.5 requests use the current conversation's native image tool when available. If that tool hides its model, generate the requested image but report the exact model of this call as unverified. If it names another model or the user requires per-call model proof, do not claim 2.5; ask whether to use an exact Flare or Sunburst API model ID and its separate cost path. Never switch to the API automatically. For a requested Gemini or Midjourney model, use only a confirmed generation connection for that provider, otherwise return its prompt without claiming generation. For connected audio use ElevenLabs; when only a prompt is requested use a prompt-builder skill. Identify credits or usage limits, aspect constraints, and brand anchors. - Select Skill — Load the matching
media-*skill using the host's skill mechanism when available. If the host has no named Skill tool, read the installed skill's instructions directly. Do not assume that a Claude MCP tool name exists in ChatGPT Work. - Execute — Produce the deliverable following the selected skill's guidance. For OpenAI API/Higgsfield/ElevenLabs generation, use the MCP tools per the corresponding
media-*skill — always with the skill's cost notice and user-confirmation protocol. Write files where the user asked for files; otherwise return content (prompt text, generated image, asset path) in the response. - Observe — Check the output against the skill's own quality bar and the creator's stated constraints (aspect ratio, resolution, duration, brand/style consistency, licensing/credit obligations).
- Verify — For high-stakes output (brand-critical visuals, commercial-use assets, batch generation, IP/copyright-sensitive likenesses), request
media-brand-auditorwhen the host can invoke that agent. If it cannot, inspect the artifact with the same criteria and report that an independent audit was unavailable. Do not call a self-check independent. - Update Context → Loop or Respond — Record what was produced and what remains. If steps remain, loop back to step 2. When the goal is met, respond with the deliverables, the credit/usage notes, and any residual risks.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago Changed · -26 tokens per session 4ba7372067ba
- 1mo ago First seen · 33 lines · 117 tokens per session scan A 2ab35bb15596
media-producer is an agent published in the GitHub repository modu-ai/moai-cowork (313 stars, last pushed 19d ago), licensed Apache-2.0. It adds 91 tokens to every session and 1,094 once invoked, about $0.0004 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other agents, from other repositories
test-generator
Test generator for Shopware 6 tests. Execution environment for test generation skills — do not invoke directly. Skills fork into this agent via context: fork. Does not review tests — use the appropriate reviewer agent for that.
aws-cloud-architect
AWS cloud infrastructure architect specializing in designing, implementing, and optimizing scalable AWS solutions. Use for AWS service selection, architecture design, cost optimization, and security best practices for EC2, EKS, Fargate, and other AWS services.
fakoli-herald
Fakoli Crew documentation writer for READMEs, marketplace copy, release notes, and user-facing explanations.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
.NET-Notebook-Migration-Agent
Expert .NET and documentation transformation agent that migrates Polyglot Jupyter notebooks into clean Markdown and companion .NET sample code.
AVM Owner Triage
Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.