aca-sandboxes

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill includes a CLI example that upserts secrets by passing key/value pairs directly on the command line (aca sandboxgroup secret upsert --values "K=V"), which would require the LLM to emit secret values verbatim if it filled in real secrets, creating an exfiltration risk.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill contains runtime fetch-and-execute one-liners (curl | sh and irm | iex) and a wheel download used to install software at runtime — specifically https://aka.ms/aca-cli-install, https://aka.ms/aca-cli-install-ps, and https://aka.ms/aca-sdk-python — which download and execute remote code during skill use.