azure-to-aws

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • Regulated Command Execution: The skill is designed to generate and execute local Python or Node.js scripts to facilitate data extraction from Azure and various AI usage APIs (OpenAI, Anthropic, OpenRouter). It also relies on the user's authenticated az CLI for live discovery.
  • Evidence: Found in discover-live.md, discover-openai-api.md, discover-anthropic-api.md, and discover-openrouter-api.md which contain detailed logic for generating and running these capture scripts.
  • Context: These executions are non-mutating (read-only) and are used strictly to build a migration inventory. The skill employs a 'dispatched worker' model with restricted tool access to minimize risk.
  • Sensitive Data Handling: The discovery process for AI workloads involves collecting administrative API keys from the user.
  • Evidence: Procedures in discover-anthropic-api.md and discover-openai-api.md instruct the agent to write these keys to local files (e.g., .anthropic-admin-env) with restricted permissions (chmod 600) and delete them once the capture is complete.
  • Context: The skill implements a 'Hard Rule' that keys must never enter the chat transcript. This localized, temporary handling follows security best practices for credential management.
  • Consent-Gated Telemetry: The skill includes a telemetry protocol for reporting migration progress.
  • Evidence: references/vendored/telemetry/PROTOCOL.md and associated scripts define an opt-in flow that sends anonymized data to *.activate.aws.dev.
  • Context: Reporting is strictly optional, anonymized (verified against leaks of resource names or IDs), and uses standard AWS-owned infrastructure for collection.
  • Prompt Injection Defense: The skill explicitly acknowledges the risk of malicious instructions within user-supplied files.
  • Evidence: SKILL.md contains an 'Input Security' section directing the agent to treat all user files as untrusted data and to ignore any attempts to override migration instructions.
  • Context: This proactive instruction helps the agent maintain its intended behavior when processing attacker-influenced content like resource names or tags.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 12:39 AM