The Agent Skills Directory

Prompt Injection (LOW): The SKILL.md file contains a directive: "IMPORTANT: Don't read scripts unless absolutely needed". This instruction discourages the agent from inspecting the skill's underlying executable code, which is a pattern that can be used to obscure malicious logic or bypass agent oversight.
Data Exposure & Exfiltration (LOW): The scripts perform network requests to api.elections.kalshi.com. This domain is not on the trusted whitelist. While appropriate for the skill's purpose, it represents an external data flow to a non-whitelisted destination.
Unverifiable Dependencies & Remote Code Execution (LOW): The scripts use uv inline metadata to install unversioned dependencies (httpx, click) from the public PyPI registry at runtime. This introduces a minor risk of dependency confusion or supply chain compromise.
Indirect Prompt Injection (LOW): The skill possesses a vulnerability surface for indirect prompt injection. Untrusted data (market titles, rules, descriptions) is ingested from the Kalshi API and displayed to the agent via click.echo without sanitization or boundary markers to delimit external content from instructions.

kalshi-markets