hcom-agent-messaging

Warn

Audited by Gen Agent Trust Hub on Feb 22, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • EXTERNAL_DOWNLOADS (MEDIUM): The skill requires the installation of the 'hcom' package via pip install hcom. The repository (github.com/aannoo/hcom) is not from a trusted organization or repository list.
  • COMMAND_EXECUTION (MEDIUM): The skill executes multiple shell commands including hcom hooks add, hcom reset all, and hcom start. These commands modify the local development environment and manage system processes (spawn, resume, kill).
  • DATA_EXFILTRATION (LOW): The tool supports 'Cross-device' communication via an 'MQTT relay'. This implies that agent transcripts, event histories, and terminal screens are transmitted over the network to a relay server.
  • INDIRECT_PROMPT_INJECTION (LOW):
  • Ingestion points: The skill ingests data from other agents via @mentions, transcripts, and event history (SKILL.md).
  • Boundary markers: There are no explicit instructions to treat inter-agent messages as untrusted or to use specific delimiters.
  • Capability inventory: The skill can spawn/kill processes, inject text into terminals, and modify file system hooks.
  • Sanitization: No sanitization or validation of the messages/transcripts from other agents is mentioned.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 22, 2026, 02:51 PM