agentleFS
vericontextvericontext/vibeframe/.agents/skills/vibe-pipeline/SKILL.md

vibe-pipeline

Skill185 starsChanged todayMITView on GitHub

Author and run VibeFrame YAML pipelines with vibe run, checkpoints, and budget controls.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 1 is distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when the user wants a reproducible multi-step media workflow, checkpointed execution, or budget-limited provider calls.

    line 12
  2. Very specific:

    Use scene-build and scene-render when a storyboard project already exists:

    line 52
  3. Very specific:

    Keep this in sync with packages/cli/src/pipeline/types.ts and packages/cli/src/pipeline/executor.ts.

    line 80
  4. Very specific:

    Keep step ids short and stable (image, motion, title, render).

    line 135
  5. Very specific:

    Use $<step-id>.output rather than repeating file paths.

    line 137
  6. Very specific:

    Add a budget: block for any pipeline that can call image/video providers.

    line 138
  7. Very specific:

    Use scene-build for storyboard projects and generate-* / edit-* for standalone media chains.

    line 139
  8. Very specific:

    When converting a shell sequence, map each vibe ...

    line 142

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Vibe Pipeline

A VibeFrame pipeline is a YAML manifest executed by vibe run. Steps can reference earlier outputs with $<step-id>.output, and the executor writes a checkpoint under the output directory so failed runs can resume.

Use this skill when the user wants a reproducible multi-step media workflow, checkpointed execution, or budget-limited provider calls.

Minimal Skeleton

name: promo-video
budget:
  costUsd: 8
  maxToolErrors: 1
steps:
  - id: image
    action: generate-image
    provider: openai
    model: "2"
    size: 1536x1024
    quality: hd
    prompt: "sleek product shot on a quiet studio background"
    output: hero.png

  - id: video
    action: generate-video
    provider: seedance
    image: $image.output
    prompt: "slow cinematic camera push, polished product demo motion"
    duration: 5
    ratio: "16:9"
    output: hero.mp4

  - id: titled
    action: edit-motion-overlay
    input: $video.output
    description: "minimal lower-third title, clean sans-serif, subtle grain"
    model: gemini
    understand: auto
    output: titled.mp4

Scene Project Skeleton

Use scene-build and scene-render when a storyboard project already exists:

name: storyboard-render
budget:
  costUsd: 2
  maxToolErrors: 1
steps:
  - id: build
    action: scene-build
    project: my-video
    mode: batch
    composer: openai
    tts: kokoro
    skipBackdrop: true
    skipRender: true

  - id: render
    action: scene-render
    project: my-video
    output: renders/final.mp4
    quality: standard
    fps: 30
    format: mp4

Supported Actions

Keep this in sync with packages/cli/src/pipeline/types.ts and packages/cli/src/pipeline/executor.ts.

  • Generate: generate-image, generate-video, generate-tts, generate-sfx, generate-music, generate-storyboard, generate-motion
  • Edit: edit-silence-cut, edit-jump-cut, edit-caption, edit-noise-reduce, edit-fade, edit-translate-srt, edit-text-overlay, edit-motion-overlay, edit-grade, edit-speed-ramp, edit-reframe, edit-interpolate, edit-upscale, edit-image
  • Audio: audio-transcribe, audio-isolate, audio-dub, audio-duck
  • Detect: detect-scenes, detect-silence, detect-beats
  • Analyze/review: analyze-video, analyze-media, review-video
  • Scene: scene-build, scene-render

Each step names its main input input and its file output (relative to the output directory); other parameters follow the matching vibe command or MCP tool. If unsure, read the executor source. There is no public vibe run --list-actions command.

vibe run refuses an unknown action before any step runs (exit 2). A run with a failed step exits 1 with the per-step report in the error's data; fix the step and rerun with --resume. A failed step carries errorCode (AUTH_ERROR, API_ERROR, NOT_FOUND, ...) and, for a provider failure, errorKind (auth, quota, rate-limit, ...): fix the key for auth, wait for rate-limit, and retry provider/network.

Running

vibe run pipeline.yaml --dry-run
vibe run pipeline.yaml -o pipeline-output
vibe run pipeline.yaml -o pipeline-output --resume
vibe run pipeline.yaml -o pipeline-output --budget-usd 5
vibe run pipeline.yaml -o pipeline-output --json

Useful flags:

  • --dry-run validates the graph and shows upper-bound cost estimates.
  • On a real run each step reports costUsd: its metered provider usage priced from the model catalog, or its upper-bound estimate (flagged costEstimated) when nothing was metered. budget.costUsd counts these, including what failed steps were billed.
  • --resume skips completed checkpointed steps.
  • --budget-usd, --budget-tokens, --max-errors override or extend the manifest budget.
  • --effort low|medium|high|xhigh forwards an LLM effort hint where supported.
  • --json returns structured output for agents.

Checkpoint file: <output-dir>/.pipeline-state.yaml.

Authoring Rules

  1. Dry-run before executing paid provider steps.
  2. Keep step ids short and stable (image, motion, title, render).
  3. Name outputs with the real extension (.png, .mp4, .mp3, .json).
  4. Use $<step-id>.output rather than repeating file paths.
  5. Add a budget: block for any pipeline that can call image/video providers.
  6. Use scene-build for storyboard projects and generate-* / edit-* for standalone media chains.

When converting a shell sequence, map each vibe ... command to one action and turn downstream input paths into step references.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.