hermes-in-buzz

Pass

Audited by Gen Agent Trust Hub on Aug 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill performs shell commands to configure the gateway and manage system services across different operating systems.
  • The instructions utilize hermes gateway install and service management tools like systemctl or launchd for persistence.
  • The scripts/update_buzz_credentials.py utility executes the buzz CLI via subprocess.run to verify relay connections.
  • [EXTERNAL_DOWNLOADS]: The skill retrieves project assets and source code from established platforms.
  • Downloads the stable release of the Buzz CLI and source code from https://github.com/block/buzz.git.
  • Fetches documentation and integration guides from https://hermes-agent.nousresearch.com.
  • [DYNAMIC_EXECUTION]: The setup process involves compiling the Buzz CLI and a credential helper from source code.
  • The workflow uses cargo build --release to build tools from the cloned official repository.
  • It involves compiling the embedded Rust source scripts/hermes_buzz_credential_helper.rs into a local binary for cryptographic operations.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages sensitive cryptographic keys using security best practices to prevent leakage.
  • scripts/update_buzz_credentials.py uses getpass to ensure private keys are never passed via command-line arguments or stored in shell history.
  • Credentials are saved to a .env file with restricted read/write permissions (chmod 0o600) to ensure they are accessible only by the current user.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external configuration data such as relay URLs and user identities which are interpolated into local configuration.
  • Ingestion points: User-provided relay URLs, public keys, and channel UUIDs are collected during the discovery and configuration phases.
  • Boundary markers: The instructions do not use explicit delimiters to isolate external data from internal logic.
  • Capability inventory: The skill has capabilities for filesystem writes, subprocess execution, and service installation.
  • Sanitization: Inputs are validated against cryptographic regex patterns (e.g., KEY_RE, PUB_RE) in the Python helper script before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 22, 2026, 11:55 AM