The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to operate the forecast CLI, enabling the agent to list markets, quote prices, execute trades, and create new prediction markets on platforms like Polymarket and Kalshi.
[PROMPT_INJECTION]: The skill ingests untrusted data from external prediction market providers, which could contain malicious instructions within market descriptions or resolution criteria.
Ingestion points: Data enters the agent context through the output of forecast market list, forecast market search, and forecast market get subcommands.
Boundary markers: The workflow documentation explicitly requires the agent to present results to the user and obtain manual approval before using the --confirm flag for trades or market creation.
Capability inventory: The skill has the capability to execute financial transactions, sell owned positions, and create new blockchain-based markets via the Precog platform.
Sanitization: The skill lacks automated sanitization or filtering of external market strings, relying solely on human review of the generated previews.
[CREDENTIALS_UNSAFE]: The skill handles highly sensitive information, including blockchain private keys and API credentials.
Sensitive file paths: The references/config-and-auth.md file references private_key_file paths for Polymarket, Precog, and Kalshi.
Mitigation: The instructions mandate best practices for secret hygiene, such as using environment variables (e.g., POLYMARKET_PRIVATE_KEY) or ignored key files instead of passing secrets as command-line arguments.

forecast-cli