Diagnose FLUX 3 Briefs Before They Break

Where this fits

You've just received a brief for a FLUX 3 video generation task. The client wants a clip that starts from an existing video, but the brief only mentions a single keyframe and says "match the style." If you send that straight to the API, you'll either get a routing error or a generation that ignores the intended starting point. The flux-3-prompt-doctor skill steps in before any prompt is written, forcing you to resolve the blocking decisions that determine mode, payload, and production method—so your agent doesn't waste a call on a doomed request.

Why agents benefit

• Asks only blocking questions – It filters out nice-to-have details and surfaces only the requirements that change routing or payload, keeping the diagnostic loop short.
• Preserves hard constraints verbatim – Product names, exact copy, source roles, and fixed constraints are carried through untouched, preventing accidental drift.
• Routes by what must survive – It maps the request to t2v, i2v, or v2v based on whether you need full generation, keyframes, or a start video, eliminating guesswork.
• Labels assumptions instead of guessing – When a detail isn't specified, it uses documented defaults or marks the assumption explicitly, so you always know what was inferred.
• Prevents schema mismatch – If the brief implies a field that doesn't exist in the current FLUX 3 schema, it catches that before the API call, saving debugging time.

Practical scenarios

Missing keyframes for an image-to-video request

The brief says "animate this photo" but provides no ending frame. The skill detects that mode: "i2v" requires at least one keyframe and prompts for a closing frame or a storyboard, or it falls back to a documented default and labels the assumption.

Conflicting modes in a single brief

The brief says "generate a clip from this text" but also includes a start video file. The skill flags the conflict between t2v and v2v, asks which requirement is authoritative, and then routes accordingly—before you waste tokens on a mixed request.

Schema changes between FLUX versions

Your agent was built for FLUX 2, but the brief references a new field like start_video that only exists in FLUX 3. The skill checks the current schema, resolves the mismatch, and updates the routing logic without you manually tracking changelogs.

Add it to your agent workflow

Install the skill via npx:

npx skills add black-forest-labs/skills --skill flux-3-prompt-doctor

Then, in your agent's pre-generation step, call it with the raw brief:

Use flux-3-prompt-doctor on this brief: "Generate a 5-second clip of a cat walking, starting from the attached video, with the exact text overlay 'Meow'."

The skill will return a structured list of blocking questions and assumptions, which your agent can resolve before building the final prompt. For a deeper look at how this fits into a broader agent skill library, see the OpenAgentSkill entry.

Compare before adopting

Check whether the skill's routing logic matches your current FLUX 3 API version—schema changes can render it stale. Compare its question-asking behavior against your own heuristics: does it ask too few or too many questions for your typical briefs? Also evaluate alternatives like custom validation scripts or prompt templates; this skill is useful only if you need structured diagnosis rather than a fixed prompt formatter.

Why it is worth tracking

With 106 stars and active maintenance from Black Forest Labs, this skill signals that the official team cares about pre-generation quality. It's worth tracking because it evolves with FLUX 3's schema—when new modes or fields appear, the skill updates to handle them. Re-evaluate it whenever you upgrade your FLUX integration or when your briefs start hitting routing errors that this skill could have caught earlier.