improve-website
Warn
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: MEDIUMMETADATA_POISONINGEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [METADATA_POISONING]: The skill metadata identifies 'wondelai' as the author, which conflicts with the expected author 'emmraan' provided in the analysis context.\n- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to suggest the installation of global packages from a third-party source not recognized as a trusted vendor.\n
- Evidence: Operating Rule 4 instructs the agent to offer: 'npx skills add wondelai/skills/ --global'.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection through the ingestion of external, untrusted data.\n
- Ingestion points: The skill ingests data from live website URLs, copy, and external visitor feedback such as exit surveys, reviews, and support tickets (Intake step 7 and Phase 1).\n
- Boundary markers: There are no instructions or patterns provided to use delimiters or to ignore potential instructions embedded within the ingested data.\n
- Capability inventory: The skill can write to the local file system (creating and extending files in the 'docs/' folder) and invoke multiple other agent skills based on the processed data.\n
- Sanitization: No sanitization, validation, or filtering of external content is described before it is processed or stored in project artifacts.\n- [COMMAND_EXECUTION]: The skill generates shell commands for the user to execute for environment setup.\n
- Evidence: The 'npx' command suggested in Operating Rule 4.
Audit Metadata