The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill requires the user to execute multiple Python and shell scripts downloaded from an external GitHub repository, including scripts/init_polis.py, scripts/demo.sh, and scripts/route_contract.py. These scripts perform file system operations, such as writing configuration files to the project root and the .agents/skills/ directory.
[EXTERNAL_DOWNLOADS]: The protocol framework and its dependencies are fetched from a community-managed GitHub repository (yehudalevy-collab/polis-protocol) and an unverified package on PyPI (polis-protocol).
[INDIRECT_PROMPT_INJECTION]: The skill implements a routing and governance model that integrates agent-generated outputs into its decision-making core.
Ingestion points: The agent processes and incorporates data from markdown files in _polis/contracts/ and _polis/citizens/ into the _polis/CONSTITUTION.md and the router's history.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are documented for the incorporation of settled work lessons into the routing logic.
Capability inventory: The skill executes Python scripts with file-write permissions across the project and agent skill directories.
Sanitization: There is no evidence of sanitization or validation for agent-proposed "amendments" or "lessons" before they are filed and used to influence future agent routing and protocol rules.
[METADATA_POISONING]: The internal metadata in SKILL.md identifies the author as yehudalevy-collab, which conflicts with the platform-provided author context (sickn33). This discrepancy could lead to confusion regarding the provenance and security review status of the external code.

polis-protocol