skills/runwield/skills/guide/Gen Agent Trust Hub

guide

Pass

Audited by Gen Agent Trust Hub on Nov 2, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external websites through web search and page fetching, alongside reading repository source code, configuration, and documentation. This content is used to answer user queries and can lead to the creation or modification of markdown files. If these external sources or repository files contain hidden malicious instructions, the agent might execute them, potentially misusing its shell or file-writing capabilities.
  • Ingestion points: External web content (search/fetch) and local repository artifacts (SKILL.md, project source, docs).
  • Boundary markers: While the instructions emphasize verification and authority hierarchies, they do not include explicit technical delimiters or instructions to ignore embedded commands within the processed data.
  • Capability inventory: Execution of shell discovery commands (git, ls, grep), web navigation tools, and markdown file modification (via the 'documentation' skill).
  • Sanitization: No explicit sanitization or filtering of external input is described.
  • [COMMAND_EXECUTION]: The skill explicitly allows the use of shell commands for "safe discovery." Although it restricts these to read-only discovery commands and prohibits state-altering actions, the underlying capability to execute shell commands remains a surface that could be targeted by sophisticated prompt injection.
  • [DATA_EXFILTRATION]: The skill has combined access to local repository files (including configuration and source code) and external network tools (web search and page fetching). This creates a technical capability for data exfiltration if the agent is manipulated into reading sensitive local data and transmitting it to an external endpoint via the search/fetch tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 2, 2026, 04:20 PM