// Discover and run any FAL API model or operation the user requests. Use this as a generic FAL escape hatch when the user wants to generate something that does not fit a narrower Image Blast skill.
Create, inspect, and manage an IMAGE-BLAST project envelope under worlds/<slug>. Use before other image-blast skills (image-blast-uncover, image-blast-world, image-blast-3d, etc.) or whenever the user asks about active project state.
Main image analysis skill, object generation, plate decisionmaking, and initial scene description. Use this skill when user wants to create objects and necessary for main image analysis.
Generate the 3D static environment of a world from a source or plate image.
Generate one specified atomic 3D object. Use when the user names exactly one object instance to make, or provides one image plus the object name/description.
Generate one clean plate/source cleanup image for an IMAGE-BLASTER world. Use after confirmed objects exist, when removing foreground objects or other specified content from a source image.
Generate one image edit from explicit input images and a prompt. Use for source cleanup, clean plates, object removal, or other FAL-backed image edits.
| name | image-blast-wildcard |
| description | Discover and run any FAL API model or operation the user requests. Use this as a generic FAL escape hatch when the user wants to generate something that does not fit a narrower Image Blast skill. |
Resolve one arbitrary FAL API operation from $ARGUMENTS, confirm it with the user, then run it only after confirmation.
run-fal.mjs, or launch the background image-blast-wildcard agent until the user confirms the exact endpoint.CONFIRMED_FAL_ENDPOINT: <endpoint>. Do not ask for model confirmation again; validate inputs and run exactly one request.https://api.fal.ai/v1/models?q=<query>&status=active&limit=5https://api.fal.ai/v1/models?category=<category>&status=active&limit=5https://api.fal.ai/v1/models?endpoint_id=<endpoint>&expand=openapi-3.0https://fal.ai/docs/llms.txt and the model API docs only as fallback context when the model search response is insufficient.confirm fal-ai/flux/dev.expand=openapi-3.0, build schema-shaped JSON from the user's literal inputs, resolve the output location from the user's request or surrounding project context, and launch Agent(image-blast-wildcard) with a prompt that starts with CONFIRMED_FAL_ENDPOINT: <endpoint>.FAL_KEY is unavailable.ls -a before reading generated state. Do not use a dedicated wildcard directory by default; choose the output directory contextually from the user's request, the active Image Blast project/world, an input file's surrounding generated-output directory, or another clear local workflow context. If no output location can be inferred, ask before execution.--file <schema_key>=<path> so the helper converts them to model input URLs. For nested keys use dot paths, such as image_urls.0.The confirmed background agent should run:
node .claude/scripts/fal/run-fal.mjs \
--endpoint "<fal endpoint, such as fal-ai/flux/dev>" \
--input-json '<schema-shaped JSON input>' \
--output-dir "<output directory>" \
--output-slug "<short output slug>" \
--user-prompt "<literal user request>"
Use --mode run only when the FAL API page requires a direct fal.run call instead of the queue API. The default queue mode persists request metadata before polling and downloads any returned file URLs.
If request metadata records provider URLs but local files are missing, fill them from the matching hidden request JSON:
node .claude/scripts/project/ensure-local-assets.mjs --from "<request-json-path>"
Final response before confirmation: ask for confirmation of the exact endpoint. Final response after execution: report the endpoint, input summary, output directory, downloaded output files, request metadata, and any raw result fields that were not downloadable.