maintaining-ai-models
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides a GitLab CI automation template that downloads a Node.js script (ai-models.mjs) from the author's official GitHub repository (lenneTech/claude-code). This process is secured by pinning the script to a specific commit SHA and performing a SHA256 checksum verification before the script is executed by Node.js.
- [COMMAND_EXECUTION]: The skill facilitates the execution of local project commands defined in a .ai-models.json configuration file. Specifically, it executes a benchmark.command to run project-specific benchmarks and an inventory.runtime.command to inspect the project's runtime environment for model IDs. These are executed within the project's context.
- [INDIRECT_PROMPT_INJECTION]: The skill represents a vulnerability surface because it processes untrusted project-local data (.ai-models.json configuration and benchmark results in JSONL format) and uses this data to influence its analysis and reporting. Findings follow the mandatory evidence chain:
- Ingestion points: .ai-models.json (adapter configuration) and .jsonl benchmark result files.
- Boundary markers: Absent; the skill reads structured data without explicit instructions to ignore embedded natural language commands.
- Capability inventory: Shell command execution (via ai-models.mjs) and network requests to external LLM providers.
- Sanitization: Implements a mask() helper that anonymizes digits and letters in output logs and reports to protect potentially sensitive values.
Audit Metadata