Agent-Skills.md

Agent Skills: Grok Aurora CLI

Use when users need multi-provider image/video generation workflows in `<workspace-root>/grok-aurora-cli`, including running `grok make`, provider-specific `grok video` and image commands, tracking async tasks, stitching clips, and evaluating outputs with `grok eval`.

UncategorizedID: dbmcco/claude-agent-toolkit/grok-aurora-cli

Repository

dbmcco/claude-agent-toolkit
dbmccoLicense: MIT
122

Install this agent skill to your local

pnpm dlx add-skill https://github.com/dbmcco/claude-agent-toolkit/tree/HEAD/skills/grok-aurora-cli

Skill Files

Browse the full folder contents for grok-aurora-cli.

Download Skill

Loading file tree…

skills/grok-aurora-cli/SKILL.md

Skill Metadata

Name
grok-aurora-cli
Description
Use when users need multi-provider image/video generation workflows in `<workspace-root>/grok-aurora-cli`, including running `grok make`, provider-specific `grok video` and image commands, tracking async tasks, stitching clips, and evaluating outputs with `grok eval`.

Grok Aurora CLI

Overview

Run the existing grok-aurora-cli project reliably and reproducibly. Prefer direct CLI commands over ad-hoc scripts, and produce concise run summaries with output paths, provider/model used, and any failures.

Quick Start

  1. Move into the project: cd <workspace-root>/grok-aurora-cli
  2. Activate environment: source activate.sh
  3. Confirm CLI: grok --help
  4. Check provider keys: ../claude-agent-toolkit/skills/grok-aurora-cli/scripts/check-env.sh

Execution Workflow

  1. Resolve intent and provider
  • Use grok make for guided, interactive runs.
  • Use direct commands for deterministic execution (grok video, grok fal-image, grok novita-image, grok venice-image, grok gemini-image, grok kling-image, grok banana).
  1. Validate prerequisites before long runs
  • Ensure the selected provider key exists.
  • Ensure prompt and image paths exist.
  • Keep output naming explicit for repeatability (-o output/<name>.mp4).
  1. Run generation and capture IDs/paths
  • For async providers, record task/request IDs immediately from CLI output.
  • For long video runs, prefer one command at a time to avoid confusing status tracking.
  1. Poll and retrieve assets
  • Use grok video-status for pending tasks.
  • Re-run with explicit status flags (--fal, --novita, --kling, --seedance2, --veo) as needed.
  • Use grok veo-4k for Veo upgrade flows.
  1. Evaluate and summarize quality
  • Use grok eval <image-paths...> with optional --reference to score identity/style consistency.
  • Report score totals, pass/fail threshold, and top failure reasons.
  1. Manage outputs and logs
  • Use grok stitch for chain assembly.
  • Use grok log --errors for debugging.
  • Keep generated artifacts in output/ with scenario-based subfolders.

Command Selection

Use the smallest command that satisfies the request:

  • Text-to-image quick test: grok generate "..." or grok gemini-image ...
  • Image-to-image: grok fal-image ..., grok novita-image ..., grok venice-image ..., grok kie-image ..., grok banana ...
  • Image-to-video: grok video <prompt-file> -i <image> --provider <provider>
  • Guided selection: grok make
  • Evaluation: grok eval ...
  • Logs: grok log, grok clear-log

Prefer prompt files under prompts/ for reproducibility rather than inline prompt text for long runs.

Reporting Format

After each run, provide:

  1. Command executed (redact secrets)
  2. Provider/model and key options
  3. Output file paths
  4. Task/request IDs (if async)
  5. Evaluation result (if run)
  6. Failures and next corrective action

Resources

  • scripts/check-env.sh: Print provider key readiness and fail if no providers are configured.
  • scripts/run-grok.sh: Run grok with project activation handled automatically.
  • references/command-recipes.md: Common command templates for repeatable generation and evaluation.

Use GUIDE.md in the CLI project as the authoritative command/options reference when a flag question is ambiguous.