agent-mesh-creator

Warn

Audited by Gen Agent Trust Hub on Mar 8, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill directs the agent to execute shell commands to manage agent lifecycles and start a persistent background daemon process on the host machine.\n- [EXTERNAL_DOWNLOADS]: The workflow involves downloading and installing the @annals/agent-mesh package from the npm registry, which is a resource owned by the vendor 'annals-ai'.\n- [CREDENTIALS_UNSAFE]: The skill handles sensitive authentication material, including user tokens for CLI login and bearer tokens for configuring external network exposure.\n- [DATA_EXFILTRATION]: The agent-mesh agent expose command facilitates the creation of network tunnels from the local environment to external providers like agents-hot and generic-a2a, potentially exposing local services to the internet.\n- [PROMPT_INJECTION]: The skill includes an attack surface for indirect prompt injection by processing untrusted Markdown files from local project directories.\n
  • Ingestion points: Local configuration and documentation files such as SKILL.md and CLAUDE.md.\n
  • Boundary markers: No delimiters or safety instructions are used to separate external content from the agent's instructions.\n
  • Capability inventory: The agent has the ability to execute CLI commands and configure network-facing bridges.\n
  • Sanitization: There is no evidence of validation or sanitization of the content within the ingested Markdown files.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 8, 2026, 04:43 PM