video-batch-runner

Video Batch Runner

This is the single PostPlus video-generation skill. It owns creative request assembly and execution across the six Seedance 2.5 generation modes; do not create a separate architect, preflight report, reference contract, or model-specific submitter.

Workflow

1. Read the brief and inspect the supplied media. Use media-analysis first only when understanding an existing video materially changes the new render.
2. Run postplus media schema --json, choose the smallest Seedance 2.5 endpoint whose input roles match the job, then load its exact fields with postplus media schema --endpoint <endpoint> --json:
   • no driving media: text-to-video
   • one opening image: image/first-frame-to-video
   • locked opening and closing images: first-last-frame-to-video
   • several images, videos, or audios that guide the result: reference video
   • change an existing video's content: video edit
   • continue an existing video: video extend
3. Decide each asset's role from the task. Bind identity, product, first/last frame, timing, or motion only when the asset must control that property. Treat style and benchmark media as inspiration, and omit media that should not influence the render. Never infer a role merely because a file is present.
4. Write one self-contained, model-facing prompt per clip. Include the visible subject and action, product behavior, scene, camera, timing, speech/sound, continuity, and the role of every submitted reference. Use explicit [image N], [video N], and [audio N] bindings when the selected endpoint accepts numbered references.
5. Validate fields, enum values, defaults, cardinality, and duration against the current schema. Do not carry private execution field tables in this skill. If one clip exceeds the supported duration or creative load, split it into independent prompts before spending.
6. Pass each local path, HTTPS URL, existing PostPlus media reference, or data URI directly to the matching role flag. The CLI validates and prepares local media before a single hosted submit. Use only fields published by the selected endpoint; do not pre-upload media or construct private payloads.
7. Prefer --wait when the expected runtime fits the command budget. If create still returns pending, poll the same run with postplus media poll --handle <output.data.id>. Prefer the CLI-returned action or resume command for the same operation. Honor its wait/recovery boundary and never submit a replacement request merely to check status.
8. When a local final file is needed, download the completed output with postplus media-file download --url <fresh-output-url> --output-file <path> --skill video-batch-runner.
9. On successful generation, deliver the raw result, actual model/endpoint, submitted prompt and reference roles, and original run information. Inspect the finished clip only when the user asks for that check. Generation delivery does not wait for transcription, frame extraction or editing.

Execution Rules

• PostPlus schema is the source of truth for endpoint availability, input fields, enums, and defaults.
• Use the released postplus CLI. Do not call underlying services directly, invent private fields, or create a service-specific submitter skill.
• Keep private request/result files under the active work folder's .postplus/ state; do not manufacture contract or planning artifacts for handoff.
• Requested tasks use existing PostPlus credits after the current price and balance are checked. Estimate-only requests do not execute. Stop on insufficient credits; do not recharge, buy a subscription, or repeat an uncertain paid operation.
• On an auth, transport, quota, malformed-request, capability, or execution failure, report the typed error and stop. Do not switch endpoints, rewrite the request, or claim a local plan is an executed render.
• In a batch, only the typed error postplus_cli_hosted_media_content_policy_blocked is item-local: record that item as blocked without rewriting or retrying it, continue the independent items, and report the incomplete set. Every other error stops the batch.

Public Command Boundary

• Readiness: postplus doctor --skill video-batch-runner
• Discover endpoints: postplus media schema --json
• Read selected endpoint fields: postplus media schema --endpoint <endpoint> --json
• Submit local or remote role media directly: postplus media create <endpoint> --skill video-batch-runner ... --wait
• Resume: postplus media poll --handle <output.data.id>
• Download: postplus media-file download --url <fresh-output-url> --output-file <path> --skill video-batch-runner

postplus media create video-seedance-2-5-text \
  --prompt "Describe the result you need" \
  --wait \
  --output ./result.json

Follow the CLI's structured result and reported next action; do not infer recovery from free-text messages. Wait for explicit user approval when requested; an action does not authorize spending, publishing, or overwriting. Resume the same operation through its returned checkpoint or action; never resubmit uncertain work, repeat exhausted recovery, or switch providers to bypass failure.