Runway Agent skills vs a saved Sume Format you call by API

A Runway Agent skill is saved instructions you trigger with / in a chat. A Sume Format is a saved recipe your backend calls by handle and slug. Which fits when.

5 min readSume
All posts

A Runway Agent skill is a saved set of instructions, up to 5,000 characters, that you trigger by typing / in the Agent composer, so it is a tool for a person working in a chat. A Sume Format is a saved production recipe that your own backend calls over HTTP by handle and slug, so it is a tool for code. If a human steers each step, a skill fits; if a service starts the job and nobody is watching, call a Format.

Side by side

Both save a repeatable workflow so you stop re-typing it. They differ in who starts the work and what comes back.

Runway Agent skill vs Sume Format (read 2026-10-04)
QuestionRunway Agent skillSume Format
Where it is savedSkills menu in the Agent composerFormat library, or the Contents API
Size of the instructionsUp to 5,000 charactersA SKILL.md plus reference files; no cap beyond 100 MiB per file
SharingPrivate, or Share with workspaceTeam handle; grants to other workspaces
How it startsType / and pick the skill by namePOST /v1/formats/{handle}/{slug}/runs
Who is in the loopSkill checks in with you along the wayUnattended: approvals are pre-granted
Your inputsDetails typed next to the skillinput object: up to 64 top-level keys, 2 MiB

What a Runway skill is

Runway's page lists built-in skills (Ad Campaign, Commercial, Mood Board, Motion Graphic, Resize Image, Social Thumbnail, UGC Video, Workflow) and a New Skill option. Each skill breaks its task into steps and checks in with you. Runway calls skills starting points, not scripts: you can redirect Agent at any step.

What a Sume Format adds

A Format run boots a fresh sandbox with the Format's package on disk, follows the recipe, and returns one receipt. The receipt carries the Format version that actually ran, so a later edit never rewrites history. Your own words go in instruction and your data in input, which the agent reads as data and not as instructions.

The catalog is callable without authoring anything. Any key with formats:write can post to the sume handle.

curl -sS -X POST "https://api.sume.com/v1/formats/sume/sume-product-commercial/runs" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: sku-1042-v1" \
  -d '{
    "instruction": "Vertical 9:16 hero clip, no captions.",
    "input": { "product_url": "https://shop.example.com/p/1042" },
    "generation_spend_cap_usd": 25
  }'

Which one to use

  • Use a skill when you are in the chat, want to review each step, and the output goes to you.
  • Use a Format when an order, a webhook or a cron job in your own system should start the video and read back a URL.
  • Keep the recipe in one place. If you maintain both, the Format is the copy that code can version, because every edit bumps its version.

Sources

Related posts

More in Comparisons

All Comparisons posts

Written by Sume