youtube-cumulative-research
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes multiple subcommands of the
yt-insightsCLI tool to manage research state, discover content, and export data. This tool is identified as the primary source of truth for the workflow. The instructions explicitly forbid automatic retries and autonomous decision-making.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through ingested YouTube metadata.\n - Ingestion points: External content from YouTube (titles, channel names, metadata) enters the context via the
yt-insights research candidatesandacquirecommands inSKILL.md.\n - Boundary markers: The instructions do not specify explicit delimiters (like XML tags) for untrusted data, instead relying on CLI-provided JSON structures.\n
- Capability inventory: The skill has the capability to execute shell commands (
yt-insights) and write data to local directories (exportcommand).\n - Sanitization: The skill mitigates risks by requiring the agent to treat invalid JSON as unverified state, requiring explicit human approval for candidate selection, and forbidding the use of model-generated prose as source material.
Audit Metadata