The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages a Higgsfield API key stored in a local configuration file at ~/.config/higgsfield/api.env. While the skill includes scripts to set this up with restricted permissions (mode 600) and handles the secret by passing it to curl via stdin to avoid process list exposure, the presence of a local plaintext credential file is a security consideration.
[DATA_EXFILTRATION]: The scripts/hf-upload.sh script uploads media files to public, unauthenticated URLs provided by the Higgsfield API. The documentation correctly warns that these URLs are public and should not be used for confidential data, but this remains a potential data exposure vector if misused.
[COMMAND_EXECUTION]: The skill makes extensive use of shell scripts (hf-api.sh, hf-gen.sh, encode.sh) that execute system commands including ffmpeg, ffprobe, curl, jq, and cwebp. These commands process data fetched from the external Higgsfield API and are used for file encoding and API communication.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests and processes untrusted data from the external Higgsfield API.
Ingestion points: API responses processed in scripts/hf-gen.sh and scripts/hf-upload.sh to extract URLs and job statuses.
Boundary markers: None explicitly defined for API response parsing, though jq is used for structured data extraction.
Capability inventory: The skill has the ability to write to the file system (media/, public/film/) and execute shell commands (ffmpeg, curl, jq).
Sanitization: Relies on jq for parsing JSON responses from the API, but lacks explicit sanitization of the values (like URLs) before they are passed to other commands.
[EXTERNAL_DOWNLOADS]: The skill downloads media files (images and videos) from external URLs provided by the Higgsfield API using curl. It also installs standard Node.js packages (lenis, playwright) and Astro project dependencies during the build step.

scroll-film-landing