The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the 'hcom' utility using the uv package manager and references the author's GitHub repository (github.com/aannoo/hcom) for installation and troubleshooting. These resources are associated with the skill's author.
[COMMAND_EXECUTION]: The skill uses various hcom CLI commands (e.g., hcom status, hcom list, hcom reset all) to manage the multi-agent environment, troubleshoot connectivity, and control agent sessions.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to allow agents to message each other, watch transcripts, and monitor file edits in real-time. This capability creates a pathway where malicious content or instructions processed by one agent could influence the behavior of other agents participating in the shared 'hcom' session.
Ingestion points: Inter-agent messages, watched transcripts, command history, and file edit notifications monitored via hcom hooks.
Boundary markers: No explicit boundary markers or instructional delimiters are specified to distinguish between agent-generated content and potential embedded instructions.
Capability inventory: The skill allows agents to spawn, fork, and kill processes in other terminals, and execute shell commands via the hcom CLI.
Sanitization: There is no evidence of sanitization or filtering for data exchanged between agents.
[DATA_EXFILTRATION]: The skill documentation explicitly identifies the locations of sensitive configuration and environment files (e.g., ~/.hcom/env, ~/.hcom/config.toml), which often contain API keys or session credentials. While it does not automatically exfiltrate these files, it provides the agent with the specific paths needed to access them.

hcom-agent-messaging