The Agent Skills Directory

[SAFE]: The skill implements a robust 'plan-search-extract-synthesize' loop designed to maximize factual accuracy and evidence-based reporting.
[SAFE]: Instructions include explicit quality filters to avoid low-quality content like landing pages and aggregators, and it utilizes a 'recovery ladder' for blocked content (e.g., using Wayback Machine or archive.today).
[SAFE]: It mandates the use of evidence labels such as [VERIFIED], [SOURCED], and [REASONED] to clearly distinguish between corroborated facts, single-source claims, and AI-generated inferences.
[SAFE]: The references/tools.md file defines abstract tool capabilities that map to common environment-specific tools (like web_search or curl), providing a clear boundary for tool usage without introducing high-risk commands or exfiltration patterns.
[SAFE]: The skill requires citing sources with tiers (primary, secondary, tertiary) and access dates, ensuring transparency and provenance in the generated reports.

research