mention-radar
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from untrusted external sources, exposing the agent to potential instruction injection via social media content.
- Ingestion points: Fetches mention text from X/Twitter (via X.AI API), Reddit, GitHub Discussions, and various blogs/newsletters via WebSearch.
- Boundary markers: While it uses jq to parse API responses, the actual text of the mentions is not wrapped in protective delimiters or warnings to ignore embedded instructions.
- Capability inventory: The skill has access to network tools (curl), GitHub CLI (gh), file system logs (memory/logs/), and notification services (./notify).
- Sanitization: No logic is present to sanitize or escape the content of retrieved mentions before they are categorized by the agent.
- [COMMAND_EXECUTION]: Employs shell-based tools to interact with APIs and manage data.
- Evidence: Utilizes bash, jq, curl (via ./secretcurl), and the gh CLI tool to perform monitoring and reporting tasks.
Audit Metadata