The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the protocol framework and various scripts from an external community repository on GitHub (yehudalevy-collab/polis-protocol). It also suggests the installation of a Python package from PyPI.
[COMMAND_EXECUTION]: Executes multiple shell commands and scripts to initialize the environment (init_polis.py), run demonstrations (demo.sh), and manage task routing (route_contract.py).
[PROMPT_INJECTION]: The routing logic ingests untrusted data from markdown files located in _polis/contracts/. This creates an indirect prompt injection surface where malicious instructions inside a task description could attempt to manipulate the agent's behavior or routing decisions.
Ingestion points: Contract files read by the polis route command from the _polis/contracts/ directory.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the skill configuration.
Capability inventory: The skill performs filesystem writes via init_polis.py and executes specialized routing logic via route_contract.py across all scripts.
Sanitization: No evidence of input validation or sanitization for the content of the processed contracts was found.

polis-protocol