THE AGENT SKILLS DIRECTORY
Reusable skills for AI agents. Install any of them into a Cohub world's agent with one command.
npx -y @neta-art/neta-search-cli@latest skills find "<what you need>"
Wrote a skill? Publish it here →
Basic Cohub abilities. Most agents already have these mounted; listed here for reference.
Paste this into your agent's chat — it fetches the skill and follows it.
Use the skill: background-removal. Run `npx -y @neta-art/neta-search-cli@latest skills use neta-skills/background-removal --via site` and follow what it prints. Read its complete output, redirecting it to a temporary file first if necessary, and resolve relative paths from the skill directory it prints.npx -y @neta-art/neta-search-cli@latest skills use neta-skills/background-removalRemove backgrounds from images to create transparent cutouts of characters and objects.
# Background Removal Use Cohub `birefnet-general` (BiRefNet) for background removal. This skill and its wrapper use only the Python standard library. Do not install or invoke local image libraries or alternate background-removal providers. ## Supported Inputs BiRefNet accepts image input as an HTTP(S) URL. The wrapper supports: - any model-supported image format supplied as an HTTP(S) URL; - local JPEG and WebP files, uploaded unchanged to Cohub temporary asset storage. Local PNG and other formats are not converted by this skill. For those files, obtain an HTTP(S) URL through an existing approved upload path and use that URL as `--input`. The temporary upload endpoint accepts JPEG/WebP and has an 8 MiB limit. The wrapper uses `COHUB_EXECUTION_TOKEN`, `COHUB_SPACE_ID`, and `COHUB_SESSION_ID`, which are normally present in Cohub sessions. Do not upload assets the user marked as secret without confirmation. ## Output Limits The wrapper preserves the raw BiRefNet result. It does not inspect or transform image pixels. BiRefNet may: - return WebP bytes regardless of the output filename; - crop the result to the foreground bounding box; - resize or upscale the foreground. This dependency-free skill therefore does not promise real PNG encoding, original canvas dimensions, alpha-channel validation, no-crop restoration, or pixel-level QA. Do not claim those guarantees after a successful command. ## Dependency Check Only verify the Cohub CLI, model, and upload environment: ```bash command -v cohub || true cohub models show birefnet-general --json env | grep -E 'COHUB_(EXECUTION_TOKEN|SPACE_ID|SESSION_ID)' | sed 's/=.*/=present/' ``` Do not install image packages, image command-line suites, or another background-removal implementation. ## Recommended Script For an HTTP(S) URL: ```bash python3 .agents/skills/background-removal/scripts/birefnet_cutout.py \ --input https://example.com/path/to/input.png \ --output path/to/cutout.webp \ --log path/to/background-removal.json ``` For a local JPEG or WebP: ```bash python3 .agents/skills/background-removal/scripts/birefnet_cutout.py \ --input path/to/input.webp \ --output path/to/cutout.webp \ --log path/to/background-removal.json ``` Use `.webp` as the default output extension because the provider commonly returns WebP. Always preserve the source file. The wrapper handles one input per invocation. For multiple files, invoke it once per supported input and keep a separate output path for each. ## Direct Cohub Call When an HTTPS URL is already available: ```bash cohub generate "remove background" \ -m birefnet-general \ --image https://example.com/path/to/input.png \ -o output.webp ``` Do not pass a local path directly to `cohub generate`; the provider rejects it. Do not pass a data URL because the CLI interprets it as a file path. ## Validation Without reading image pixels, validation is limited to: - command exit status is zero; - output exists; - output is non-empty; - log records provider, model, input mode, output path, and byte size. For visual or format-sensitive validation, use an explicitly requested workflow that provides suitable tooling. This skill itself must remain free of local image-library dependencies. ## Failure Handling - If local input is not JPEG/WebP, request or obtain an HTTPS URL through an existing approved upload mechanism. - If a local file exceeds 8 MiB, use an existing approved hosting path and pass its HTTPS URL. - If Cohub credentials or CLI are unavailable, report the exact blocker. - If BiRefNet fails, preserve the source and report provider stdout/stderr. - Do not install a local substitute or silently switch providers.