FLUX agent skills for video direction, and Sume's skill install
BFL released six Agent Skills for FLUX 3 video direction on 2026-08-07. Sume ships its own skill through the CLI. Where each installs and what it cannot do.

BFL released six Agent Skills for directing FLUX 3 video from a coding agent on 2026-08-07, and they apply to BFL's model. Sume ships a separate packaged skill: sume skills install writes it into .agents/skills or .claude/skills, and it is a Sume skill, not a FLUX 3 one.
BFL's line is from its release notes; Sume's is from the Agent skills page, both read 2026-09-29. The release note gives a count and a purpose only, so this post says nothing about what BFL's skills contain.
How do I install Sume's skill?
The CLI has commands to list, install, update, export and remove it. Export first if you want to read the source files before a custom install.
sume skills list
sume skills install
sume skills export sumeWhat does a skill not do?
A skill is instructions for an agent, not a permission. Sume's docs tell agents to prefer read commands while planning and not to submit paid provider work without explicit operator confirmation. Image, Video and Music generation are not CLI submit commands yet, so an agent generates video through the Developer API or hosted MCP, and a skill cannot change that.
| Item | BFL | Sume |
|---|---|---|
| What ships | six Agent Skills for FLUX 3 video direction | one packaged Sume skill |
| Install path | not stated in the release note | .agents/skills or .claude/skills |
| Paid work | not stated | needs explicit operator confirmation |
Which one should my coding agent use?
If the video comes from FLUX 3, use BFL's skills and BFL's API. If it comes from Sume's catalog, use Sume's skill for the CLI conventions, and hosted MCP or the API for generation. They do not conflict, since they teach different services.
How do I keep a skill from spending money?
Use the gates the docs give. In the CLI, --confirm-paid marks generation that can reserve or spend credits, and --confirm-submit covers non-paid writes such as cancelling a job. On hosted MCP, paid tools need an idempotency_key and support dry_run=true.
A skill can tell an agent to use these, but the gates are what enforce it, so put the rule in both places.
Can I read a skill before I trust it?
For Sume's, yes: sume skills export sume writes the source files out so you can review them before a custom install. Do the same for any third-party skill, BFL's included, before you point an agent that can spend at it.
What should I check after installing one?
List what landed on disk, confirm the directory is the one your agent reads, and run one read-only task before any paid one. For Sume, sume doctor --agent --json and sume catalog list --json are the read commands the docs give for planning. Then try a dry run of a paid call and confirm you see the cost preview before anything is submitted.
Sources
Related posts
More in Developers
- FLUX MCP server generate_video: what Sume's hosted MCP has
BFL's FLUX MCP server gained generate_video and enhance_video on 2026-08-11. Sume's hosted MCP has its own generate_video and video_upscale_create tools.
- 413 payload_too_large on a Sume Format run: the 4 MiB body cap
A Format run body over 4 MiB answers 413 payload_too_large. Shrink input, send media by URL, and tell it from attachment_too_large and a null payload.
- Cap what one Format run can spend with generation_spend_cap_usd
Every Sume Format run has a generation spend cap. Send generation_spend_cap_usd to set it: up to 500, null runs at 500, 0 is rejected. What happens at the cap.
- Video judder with mixed frame rates: Timeline's warning
Judder on a joined video comes from resampled frame rates. Sume's Timeline warns output_fps_resamples_sources when a source rate differs. How to read it.
Written by Sume