agent-workbench

Pass

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill dynamically fetches its core operating rules and instructions from a remote MCP server, creating a surface for indirect prompt injection.
  • Ingestion points: SKILL.md instructs the agent to search for and read canonical instructions from MCP resources at agent-protocol/70-agent-workbench.md and ai-workspace/60-skills/agent-workbench/SKILL.md.
  • Boundary markers: Absent. The instructions lack delimiters or warnings to ignore potentially malicious embedded instructions in the remote content.
  • Capability inventory: The agent utilizes MCP tools like resource_read_budgeted, context_prepare, and resource_get_content to fetch and act upon remote data.
  • Sanitization: Absent. The skill does not describe any validation or filtering mechanisms for the dynamically loaded instructions.
  • [COMMAND_EXECUTION]: The workbench.mjs module executes the local codex CLI tool using node:child_process.spawnSync to manage and validate MCP server configurations.
  • [EXTERNAL_DOWNLOADS]: The doctor functionality in workbench.mjs performs network operations to a remote MCP server (defaulting to the author's domain agent-mcp.devkimson.kro.kr) to verify tool availability and authentication status.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 6, 2026, 02:42 AM