alibabacloud-iac-code

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches a runtime manifest and binary archive from the vendor's official storage service.
  • The skill downloads its manifest from https://ros-public-tools.oss-cn-beijing.aliyuncs.com/github-releases/aliyun/iac-code/skill-runtime/releases/v0.15.0/runtime-manifest.json.
  • It verifies the manifest's integrity against a hardcoded SHA256 hash (50b819fb372ad3ea8171fe5556af5c187680f06c95e3e89226b1c190955482b9) and further validates the digest of the runtime artifacts before extraction.
  • [COMMAND_EXECUTION]: Spawns the verified vendor runtime as a local server to handle infrastructure tasks.
  • The script scripts/iac_code.py uses subprocess.Popen to launch the downloaded binary. Execution is restricted to the validated vendor artifact and is necessary for the skill's primary function of managing cloud resources via IaC (Infrastructure as Code).
  • [DATA_EXPOSURE]: Implements a proactive secret redaction mechanism to prevent credential leaks in agent logs or user-facing messages.
  • The _sanitize_text function in scripts/iac_code.py utilizes a regular expression (_SECRET_PATTERN) to identify and redact Bearer tokens, access keys, and passwords from internal text streams.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user requests and external template data which may contain embedded instructions.
  • Ingestion points: The skill reads user requests from prompt files within the local workspace using the --prompt-file argument in scripts/iac_code.py.
  • Boundary markers: The SKILL.md defines an explicit inputRequired protocol. The agent is instructed to stop and approved by the user for every permission, candidate_selection, and deployment_confirmation event, preventing autonomous execution of potentially injected instructions.
  • Capability inventory: The skill has capabilities to write files to the local workspace, perform authenticated network requests to Alibaba Cloud APIs (via the runtime), and execute binary code (the verified runtime).
  • Sanitization: All progress updates and final outputs are sanitized to remove sensitive credentials before display.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 02:02 AM