ai-video-script
Pass
Audited by Gen Agent Trust Hub on Jul 16, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to how it handles external data.
- Ingestion points: Untrusted data enters the agent context via the
topic,purpose, andsource_materialfields defined in theSKILL.mdinput contract. - Boundary markers: The output templates (e.g.,
templates/video-script.md) do not utilize delimiters or specific warnings to instruct downstream agents to ignore embedded instructions within the user-provided content. - Capability inventory: The skill generates structured files (YAML manifests) that trigger actions in downstream services like
image-serviceandvideo-creator, which perform file creation and media processing. - Sanitization: No evidence of filtering or escaping is present for the user-supplied text before it is interpolated into the generated image prompts and narrations.
- [COMMAND_EXECUTION]: The
README.mdfile provides shell commands for validating and packaging the skill, such aspython3 .opencode/skills/tests/test_candidate_skills.py. These commands execute local scripts residing within the skill's repository structure. Users should ensure the development environment is trusted before running these diagnostic tools.
Audit Metadata