renku-skills

Write and revise model-specific prompts for Renku generation workflows. Use when Codex needs prompts for Seedance, Kling, Veo, Wan, LTX, image generation/editing, reference workflows, start/end-frame clips, multi-shot clips, native audio video, TTS narration, or one-time video inputs after the workflow and producer contract are known.

Audit and design condition-heavy Renku blueprint graphs. Use when a blueprint has optional branches, workflow routing, sparse fan-in, reference-vs-plain asset paths, StartEnd/image anchor logic, model-specific branches, or when Codex must prevent producers from running without available inputs or running in branches where their outputs are unused.

Author or repair Renku LLM director prompt producers. Use when creating or updating producer.yaml, prompts.toml, output-schema.json, director output schemas, prompt-producer variable bindings, or high-level prompt producers that generate downstream image/video/audio prompts for a blueprint.

Design reusable Renku blueprints for asset-only pipelines, compositing workflows, and full video generation workflows. Use when Codex needs to create or substantially restructure a Renku workflow, choose graph shape, define loops/outputs/imports/connections, decide whether to use prompt producers or direct inputs, or turn user requirements into a reusable blueprint.

Pick the right Renku workflow, producers, providers, and model configurations for generation tasks. Use when choosing between text-to-video, image-to-video, start/end-frame, reference-to-video, multi-shot, talking-head, lipsync, image edit/compose, TTS, music, or composition workflows, especially for capability-rich models like Seedance, Kling, Veo, Wan, or LTX.

Validate Renku blueprints and builds before calling work complete. Use when Codex needs to run blueprint validation, dry-run simulation, produced condition path inspection, costs-only checks, stage-by-stage --up or --up-to-layer validation, root build/test verification, or diagnose validation/runtime errors without invoking paid generation.

Create or iterate a one-time Renku video/build with an agent and the viewer. Use when the user wants to make or edit a specific video from an existing blueprint, populate inputs.yaml, generate session-time prompts instead of reusable prompt producers, run stage-by-stage, inspect viewer artifacts, pin good outputs, regenerate weak artifacts, or steer models/prompts interactively.

Maintain Renku model-specific prompt guide references. Use when Codex needs to add a new model prompt guide, refresh Seedance/Kling/Veo/Wan/LTX/image model guidance from vendor docs, update source provenance, apply the prompt guide template, or keep renku-write-prompts references current as models and producer contracts change.