The Agent Skills Directory

[COMMAND_EXECUTION]: The RELEASE_NOTES.md file (v1.1.0) explicitly documents the use of a "minified + obfuscated production build". Code obfuscation is used to hide the underlying logic and intent of executable files, making it difficult to verify the security and safety of the skill's operations.
[PROMPT_INJECTION]: There is a metadata discrepancy between the platform-identified author ('onimusya') and the author name ('Francis Hor') listed in the SKILL.md frontmatter and the LICENSE file. This mismatch is deceptive and categorized as metadata poisoning.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection due to the way it handles untrusted text inputs.
Ingestion points: The skill ingests arbitrary external data via the --prompt and --text flags in various media generation commands (image, video, and text-to-speech) as described in SKILL.md.
Boundary markers: Absent. There are no delimiters or instructions to the agent to treat these specific inputs as potentially malicious or untrusted content.
Capability inventory: The skill has the capability to perform network operations (connecting to multiple AI provider APIs) and file system writes (saving outputs to specified paths).
Sanitization: Absent. There is no evidence of validation, filtering, or escaping performed on the prompt or text strings before they are processed by the underlying models.
[EXTERNAL_DOWNLOADS]: The installation scripts (scripts/install.sh, scripts/install.ps1, scripts/install.cmd) execute npm install, which fetches unverified dependencies from the public npm registry during the setup phase.
[COMMAND_EXECUTION]: The installation process executes npm run build, which can trigger arbitrary shell commands defined in the project's build configuration script.

media-generation