oss-bucket-exploit
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches the
cosclimanagement tool from Tencent Cloud's official domain (cosbrowser.cloud.tencent.com). - [COMMAND_EXECUTION]: Provides instructions for using cloud CLI tools (
aws,ossutil,coscli) andcurlto interact with bucket APIs for auditing purposes. - [DYNAMIC_EXECUTION]: Dynamically generates XML and JSON policy files to test or modify bucket access controls.
- [INDIRECT_PROMPT_INJECTION]: The skill surface allows for indirect prompt injection through the processing of untrusted external bucket data.
- Ingestion points: Object listing and file content retrieval from cloud buckets during Phase 3 and 4 in
SKILL.mdandreferences/bucket-attack-methods.md. - Boundary markers: None; bucket data is processed directly within the agent's context.
- Capability inventory: High capabilities including cloud CLI command execution and file management across all scripts.
- Sanitization: No sanitization of bucket contents before processing is specified.
Audit Metadata