The Mandatory Entrypoint for MiniMax H3 Video Prompts

Where this fits

You ask your agent to "extend this 4-second clip to 12 seconds" with a reference image, and it replies with a generic keyframe prompt—now the video looks nothing like the subject. That is the workflow gap: MiniMax H3 has multiple generation modes, and choosing the wrong one before formatting causes expensive retries. This skill acts as a mandatory front door: it reads the official h3-prompt-writing specification, defaults image-based work to full-reference consistency, and routes to the formatting specialist without making the user understand internal routing.

Why agents benefit

• Reads the official spec in the required order: h3-prompt-writing/SKILL.md first, then mode-specific references for T2VA, I2VA, FL2VA, or L2VA.
• Enforces a non-skippable multishot question workflow for explicit multishot requests or accepted 10-second-plus proposals, so the agent does not hallucinate structure.
• Defaults image-based work to full-reference consistency; pure keyframe mode is allowed only for explicitly declared boundary-only images.
• Routes to the final formatting specialist, keeping agent orchestration clean and consistent.
• Works as the single entrypoint for creation, animation, extension, editing, restyling, reference, and prompt review, preventing duplicate or conflicting instructions.

Practical scenarios

Extending a video from a single image

The agent reads the reference spec, asks the required multishot questions, formats with full-reference consistency, then sends to the specialist—no accidental identity drift.

Restyling an existing clip without keyframe mode

A user says "make it anime but keep the motion." The skill recognizes reference generation and prevents pure keyframe mode unless boundary-only images are explicitly declared.

Prompt review before submission

A user asks "is my H3 prompt good?" The skill reads the official spec, checks the mode, reports missing constraints, and routes to the reviewer instead of guessing.

Add it to your agent workflow

Install with:

npx skills add unknowlei/minimax-h3-opencode-skills --skill minimax-h3-creative-director

Then add one instruction to your agent: "For any MiniMax H3 request, invoke minimax-h3-creative-director before other H3 skills." The skill handles read order and routing; you do not need to manually load prompt-writing references.

Compare before adopting

Compare against building your own H3 wrapper. Look at quality signals: how often the skill is updated as MiniMax H3 changes, maintainer responsiveness, community stars (currently 105), and whether the non-skippable question workflow matches your agent's tolerance for user interaction. If your workflow is fully automated, this skill may need adjustments. Also compare with alternatives in OpenAgentSkill—for example, the dedicated h3-prompt-writing skill if you only need formatting.

Why it is worth tracking

This skill is worth tracking because it codifies the official H3 spec into an agent-accessible process, not just a prompt template. Community momentum is early, and the mandatory-entrypoint design reduces prompt-format errors. Re-evaluate it when MiniMax H3 prompt specifications change or after using it on real video projects—does it actually cut retries? See the OpenAgentSkill listing at https://www.openagentskill.com/skills/unknowlei-minimax-h3-creative-director for quality signals and future updates.