researching-a-topic

Pass

Audited by Gen Agent Trust Hub on Mar 9, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill utilizes the webfetch tool to retrieve and process content from external URLs and the gh_grep tool to search through GitHub repositories.- [COMMAND_EXECUTION]: The bash tool is made available to the agent for data processing tasks, which presents a potential risk if the tool is used on maliciously crafted data retrieved from the web.- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted data from external sources without sufficient safety boundaries.
  • Ingestion points: External content enters the agent context through webfetch and gh_grep (SKILL.md).
  • Boundary markers: The instructions lack explicit requirements for the agent to use delimiters or 'ignore embedded instructions' markers when processing retrieved text.
  • Capability inventory: The skill allows for command execution via bash and file modification via the write tool.
  • Sanitization: No sanitization, escaping, or validation steps are defined for the data retrieved from external sources before it is analyzed or used.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 9, 2026, 06:36 PM