skill-selection
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses
bunx skills add <source>to fetch and install agent skills from external sources, including GitHub repositories and theskills.shregistry. It provides specific examples for fetching content from well-known organizations and the author's own public repositories. - [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands such as
bunx skills find,bunx skills add, andbunx skills updateto manage local skill installations, verify file hashes for theskills-lock.jsonfile, and perform updates. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from external repositories and local configuration files (
capabilities.tsv,candidates.tsv). This constitutes a vulnerability surface if the ingested data contains malicious instructions. The risk is addressed by an internal security protocol: - Ingestion points:
catalog/capabilities.tsv,catalog/candidates.tsv, and content from external skill repositories. - Boundary markers: Not explicitly defined in the data files; the skill relies on the agent's logic to separate evaluation from execution.
- Capability inventory: File system writes to the
skills/directory, network operations, and CLI execution viabunx. - Sanitization: Explicitly requires manual source code review of candidates to detect hidden command execution or network exfiltration before committing them to the project.
Audit Metadata