turingcoder-skills-integration

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: CRITICALEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • EXTERNAL_DOWNLOADS (MEDIUM): The skill is installed from and references an untrusted GitHub repository (https://github.com/suskycode/turingcoder-skills). [TRUST-SCOPE-RULE] applied: While downloads from vercel-labs and anthropics are downgraded to LOW, the primary skill source remains untrusted.\n- REMOTE_CODE_EXECUTION (MEDIUM): The bootstrap.sh script and the deployed workflows execute bash commands and Node-based tools fetched from remote sources. This pattern allows the author to execute arbitrary logic on the host system during the integration process.\n- COMMAND_EXECUTION (MEDIUM): scripts/deploy-workflow.sh uses sed to dynamically inject shell commands (__BOOTSTRAP_CMD__) into markdown templates. This generated content is then executed by the agent, creating a risk of command injection if the source variables are manipulated.\n- INDIRECT_PROMPT_INJECTION (LOW): The skill creates an attack surface by managing the AGENTS.md file which controls agent behavior.\n
  • Ingestion points: bootstrap.sh and deploy-workflow.sh generate project configuration files.\n
  • Boundary markers: Absent; the generated AGENTS.md lacks delimiters or instructions to ignore embedded malicious instructions within skill descriptions.\n
  • Capability inventory: Access to bash, npx, and file system operations (ln -s, mkdir).\n
  • Sanitization: No validation or escaping is performed on the repository URLs before they are processed by shell scripts.\n- AUTOMATED SCAN ALERT: An automated scanner flagged references/requirements.md for containing a blacklisted URL, which correlates with the untrusted nature of the primary repository.
Recommendations
  • Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Feb 17, 2026, 06:48 PM