buzz-self-hosting
Warn
Audited by Gen Agent Trust Hub on Aug 5, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous shell commands for managing a Docker Compose stack, including service management and executing commands inside running containers.
- Evidence: Provides complex commands for
docker compose,pg_dump, andopensslintended for setup and administration. - [DATA_EXFILTRATION]: The guide identifies specific sensitive file paths that contain plaintext private keys and provides commands for data extraction and backup.
- Evidence: Identifies
%APPDATA%\xyz.block.buzz.app\agents\managed-agents.jsonas a location for plaintext private keys on Windows systems. - Evidence: Provides instructions to back up the
.envfile (containing sensitive secrets) and dump the application database. - [EXTERNAL_DOWNLOADS]: The skill instructs the user to download and run external Docker images and perform network-based operations.
- Evidence: Usage of
docker run --rm ... alpineand mentions ofgit clonefor data migration. - [PROMPT_INJECTION]: The skill suggests analyzing agent logs which contain external runtime data, creating a surface for indirect prompt injection.
- Ingestion points: Agent harness logs (e.g.,
%APPDATA%\...\logs\*.log). - Boundary markers: Absent.
- Capability inventory: Database access and shell commands via Docker.
- Sanitization: No sanitization or filtering of log content is suggested.
Audit Metadata