The Agent Skills Directory

[SAFE]: The skill follows security best practices for AI agents, including the use of hard gates to prevent unverified recommendations and a dedicated skeptic role for evaluating data.\n- [PROMPT_INJECTION]: The skill ingests untrusted data from external sources such as property listing links (Redfin, Zillow, Realtor) and listing remarks. Ingestion points: Property links and listing marketing text. Boundary markers: Requirement for explicit grounding and [UNAVAILABLE] markers for missing data. Capability inventory: Local file read via Bun.file in underwrite.ts, command execution of the TypeScript script via bun. Sanitization: The skill enforces a 'Required source hierarchy' and 'Hard gates' to prevent synthetic or hypothetical data from resulting in a 'BUY' recommendation.\n- [COMMAND_EXECUTION]: Executes a local script (underwrite.ts) using the bun runtime. This is used for financial calculations and does not perform network operations.\n- [EXTERNAL_DOWNLOADS]: References various official US government and legal resources (IRS, California FTB, County websites) for tax and legality grounding. These are well-known, trusted sources.

rental-underwriting