delegate-x-research

Warn

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute local shell scripts (run-x-research.sh, prepare.sh, etc.), which in turn invoke node to run a local JavaScript file (delegate-cli.mjs) for task management and X research delegation.
  • [PRIVILEGE_ESCALATION]: The SKILL.md documentation explicitly states that the Grok CLI requires network access and local session state, recommending the use of escalated permissions (require_escalated in Codex) to function correctly.
  • [DYNAMIC_EXECUTION]: Multiple shell scripts in the scripts/ directory (e.g., prepare.sh, run.sh, dispatch.sh) dynamically compute the path to and execute a local Node.js script (delegate-cli.mjs) using the $dir variable, which is a form of dynamic loading from computed paths.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted Markdown data from the user (Objectives, Scope, Context, etc.) and passes it via stdin to sub-processes for research. This creates a surface where embedded malicious instructions could influence the sub-agent's behavior.
  • Ingestion points: Research requests (Objective, Scope, Context, Acceptance criteria) passed via stdin to run-x-research.sh and prepare.sh.
  • Boundary markers: Absent; the shell shims pass the raw input string directly to the underlying Node.js tool.
  • Capability inventory: Network access and X.com interaction via Grok CLI, as well as file system write operations to the run directory and response files.
  • Sanitization: No sanitization or input validation is performed at the shell shim level before passing data to the execution layer.
  • [EXTERNAL_DOWNLOADS]: The model-token-prices.json file contains numerous reference URLs to official AI provider pricing pages (e.g., Anthropic, OpenAI, Cursor). These are documented for reference purposes and do not trigger automatic downloads or code execution.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 6, 2026, 12:38 PM