The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of external CLI tools to perform its primary functions.
Evidence: The instructions in SKILL.md and references/dreamina-cli.md explicitly direct the agent to execute commands such as dreamina text2video, dreamina query_result, and plotloom video submit.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external markdown files to derive arguments for shell commands, creating an attack surface where malicious data could influence agent behavior.
Ingestion points: The skill reads from video-prompts.md and video-prompts-en.md as specified in SKILL.md.
Boundary markers: Instructions suggest extracting "model-facing prose," but there are no formal delimiters or systemic protections against instructions embedded within those files.
Capability inventory: The skill has the capability to execute shell commands (dreamina, plotloom) and perform file system operations (writing .mp4 files and markdown notes).
Sanitization: There is no evidence of programmatic sanitization or escaping of the content extracted from the prompt files before it is passed to the CLI.
[DATA_EXPOSURE]: Documentation within the skill contains hardcoded absolute paths that reveal specific host environment details.
Evidence: references/dreamina-cli.md contains the hardcoded path /Users/wangguiping/.hermes/profiles/nova/home/.local/bin/dreamina, which exposes a specific username and filesystem structure.

plotloom-video-adapter