agent-advisor
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied content from repository scans and chat descriptions to determine workload requirements. The instructions explicitly direct the agent to treat this input as data rather than instructions to mitigate potential prompt injection during the assessment and classification phases.
- [EXTERNAL_DOWNLOADS]: The skill is configured to fetch live service limits, feature statuses, and pricing documentation from official AWS domains and docs.temporal.io using web-fetch tools. These references target well-known technical services and are documented neutrally within the skill's freshness policy.
- [COMMAND_EXECUTION]: The skill generates deployment scripts (e.g., deploy.sh) and, with explicit user opt-in for 'Assisted Build' mode, executes AWS CLI and toolkit commands. These operations are protected by mandatory identity checks, per-step confirmation gates, and a resource ledger for subsequent teardown.
- [DYNAMIC_EXECUTION]: Test scripts within the repository (e.g., test_agentcore_platform.py) use exec() to validate logic snippets extracted from the skill's own documentation files. Additionally, generated deployment scripts use shell-embedded Python snippets for identity derivation. These operations are localized and part of the skill's core auditing and generation functionality.
Audit Metadata