The Agent Skills Directory

[EXTERNAL_DOWNLOADS / REMOTE_CODE_EXECUTION] (HIGH): The README instructions guide users to clone a project from an untrusted GitLab repository (https://gitlab.com/outsharp/shipp/alph-bot.git) and execute it using yarn migrate and ./index.ts. This bypasses typical package manager security checks and executes code from an unverified third-party author.
[CREDENTIALS_UNSAFE] (MEDIUM): The skill requires the use of RSA-PSS private keys (kalshi-private.pem). While required for the Kalshi API, the documentation's instructions for local storage and path referencing increase the risk of accidental credential exposure or exfiltration if the agent is compromised.
[COMMAND_EXECUTION] (LOW): The 'Getting Started' section provides several terminal commands for the user/agent to run, including cloning repos and running scripts, which can be dangerous if the target source is malicious.
[INDIRECT PROMPT INJECTION] (LOW): The skill describes an automated trading bot ('Alph Bot') that ingests external, attacker-influenced data (market descriptions and orderbook data) to make financial decisions.
Ingestion points: Market discovery fetches event contracts from Kalshi; real-time sports data from Shipp.
Boundary markers: None mentioned in the README.
Capability inventory: Order execution (placing/canceling orders), portfolio management.
Sanitization: Not documented; the bot relies on AI (Claude) to estimate probabilities based on external data.

kalshi