The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as an interface layer that relays user input to the NemoVideo backend and processes streaming responses (SSE). It lacks explicit boundary markers or sanitization logic to prevent the agent from acting on potentially malicious instructions embedded in the external API responses or user-provided descriptions.
Ingestion points: User-provided video descriptions and editing requests (SKILL.md Section 2); Streaming data received from the backend SSE endpoint (SKILL.md Section 3.1).
Boundary markers: Absent. The instructions do not define delimiters or specific "ignore" commands for the ingested data.
Capability inventory: Network access via curl (SKILL.md Section 3), filesystem access for session state (Section 1), and file upload capabilities (Section 3.2).
Sanitization: Not present, although the skill includes instructions to filter GUI-specific directions from the backend.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands using curl, uuidgen, mkdir, and echo. These tools are used for authenticating with the NemoVideo service, uploading user files, and managing the video production lifecycle.
[PERSISTENCE]: The skill stores a unique client_id in the user's home directory at ~/.config/nemovideo/client_id. This is used to maintain session continuity across requests and avoid rate-limiting on anonymous authentication tokens. The file contains only a UUID and no sensitive credentials.
[DYNAMIC_EXECUTION]: The skill provides procedural shell logic for the agent to execute, such as scripts for generating anonymous tokens and handling session failures. This involves generating commands based on runtime variables like the CLIENT_ID and NEMO_TOKEN.

wechat-video-editor