Skills
skills/malinskibeniamin/skills/setup-sandcastle/Gen Agent Trust Hub

setup-sandcastle

Pass

Audited by Gen Agent Trust Hub on May 15, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill templates ingest external content from GitHub issues directly into agent prompts without sanitization or protective delimiters. This creates a surface for indirect prompt injection attacks.\n
  • Ingestion points: Data is fetched from GitHub issues using the gh CLI in orchestration scripts.\n
  • Boundary markers: Missing in implement.md and review.md templates.\n
  • Capability inventory: Orchestration scripts use execSync. Delegated agents (e.g., Claude Code) have broad tool capabilities including file and shell access.\n
  • Sanitization: Content is interpolated raw into markdown templates.\n- [COMMAND_EXECUTION]: The orchestration logic relies on execSync to interact with local CLI tools like the GitHub CLI (gh).\n- [EXTERNAL_DOWNLOADS]: The skill's orchestration scripts fetch and install external skills from the author's repository using bunx skills@latest.
Audit Metadata
Risk Level
SAFE
Analyzed
May 15, 2026, 09:07 AM
Security Audit — agent-trust-hub — setup-sandcastle