The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download a binary Python wheel (sn_proactive_agent-0.1.3-py3-none-any.whl) and associated checksums from the vendor's GitHub repository (OpenSenseNova/SenseNova-Skills-ProactiveAgent).
[DYNAMIC_EXECUTION]: The installation process involves the sn-proactive-agent setup command, which performs automated modification of the Hermes agent's source code (specifically in the ui-tui/ directory) and triggers a build process to inject a web-only bridge.
[COMMAND_EXECUTION]: The skill requires the execution of multiple system-level commands for installation and environment setup, including pipx install, brew install, python3 -c for environment validation, and the sn-proactive-agent CLI for setup and diagnostics (doctor).
[INDIRECT_PROMPT_INJECTION]: The skill processes multi-turn conversation logs and QA history to generate project updates and proactive suggestions. While it includes a web-based approval step and instructions not to interpret suggestions as shell commands, the ingestion of arbitrary external dialogue data constitutes an attack surface.
Ingestion points: Conversation signals (turn.started, turn.completed) and QA history harvested via the Hermes connector.
Boundary markers: Explicit instructions to the agent to avoid interpreting suggested text as commands and the requirement for user approval via a separate web interface.
Capability inventory: The skill can perform binary installations (pipx), modify and build source code (setup), and initiate network services (serve).
Sanitization: The documentation states that the bridge submits actions via a native prompt.submit interface and should not interpret text as Shell or slash commands.

sn-proactive-agent