Menü
Discover and run arbitrary FAL operations that do not fit the narrower image-blast skills.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Generate one atomic 3D object from an approved object.json or direct source image.
Generate one clean plate image for a world by removing confirmed foreground objects or user-specified content.
Create, inspect, and manage an image-blaster project envelope under worlds/<slug>. Use before other image-blast skills or whenever the user asks about project state.
Generate world ambience, object impact sounds, or custom sound effects through the FAL ElevenLabs SFX endpoint.
Analyze source images, write image.json, present object candidates, and create object.json files after user confirmation.
Generate or resume the static 3D environment for a world using World Labs Marble.
Basierend auf der SOC-Berufsklassifikation
| name | image-blast-wildcard |
| description | Discover and run arbitrary FAL operations that do not fit the narrower image-blast skills. |
Use this skill as an escape hatch for FAL models outside the standard world, image-edit, 3D, and SFX flows.
This skill runs in two modes:
run-fal.mjs, or trigger any background execution 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.In discovery mode, do not run paid FAL requests until the user confirms one
exact endpoint by name (e.g. confirm fal-ai/flux/dev).
Discover candidates through the FAL Platform Model Search API (not the Explore page):
https://api.fal.ai/v1/models?q=<query>&status=active&limit=5
https://api.fal.ai/v1/models?category=<category>&status=active&limit=5
https://api.fal.ai/v1/models?endpoint_id=<endpoint>&expand=openapi-3.0
Use https://fal.ai/docs/llms.txt and individual model API docs only as
fallback context when the model search response is insufficient.
Present the best candidate endpoint(s) with category, description, and any relevant schema notes, then ask the user to confirm one exact endpoint.
After confirmation, fetch the confirmed endpoint with expand=openapi-3.0,
then build schema-shaped JSON from the user's literal inputs. Use schema
defaults for optional fields. Ask only if a required field cannot be
inferred, a referenced local file is missing, or FAL_KEY is unavailable.
Resolve 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. Do not use a
dedicated wildcard directory by default. If no output location can be
inferred, ask before execution. Use ls -a before reading generated state.
For local files, pass them with --file <schema_key>=<path> so the helper
converts them to model input URLs. Use dot paths for nested keys
(e.g. image_urls.0).
Run:
node .claude/scripts/fal/run-fal.mjs \
--endpoint "<fal endpoint>" \
--input-json '<schema-shaped JSON input>' \
--output-dir "<output directory>" \
--output-slug "<short output slug>" \
--user-prompt "<literal user request>"
The default queue mode persists request metadata before polling and
downloads any returned file URLs. Use --mode run only when the FAL API
page requires a direct fal.run call instead of the queue API.
Repair missing local files from request metadata when needed:
node .claude/scripts/project/ensure-local-assets.mjs --from "<request-json-path>"
Report endpoint, input summary, output directory, downloaded files, request metadata, and any raw result fields that were not downloadable.