comprehensive-research-agent_skill

Overview

This skill provides protocols and checkpoints to ensure rigorous, verifiable research workflows when an agent makes multiple tool calls. It focuses on source validation, explicit error recovery, traceable reasoning, and cross-source verification to prevent hallucination and premature conclusions. The result is clearer audit trails and higher-confidence research outputs.

How this skill works

The skill inserts validation checkpoints after each tool call and enforces substantive thinking blocks that record what was learned, how it maps to goals, contradictions, and next steps. It requires source tracking (which URLs were fetched vs. failed), mandated error acknowledgments with recovery strategies, and a pre-completion validation checklist before declaring tasks done. It also prescribes cross-source comparison for key claims and file-content verification using direct reads.

When to use it

• Web research involving search, read_url, fetch, or multiple file operations
• Tasks that require gathering or synthesizing information from 3+ distinct sources
• Research with explicit completeness, verification, or auditability requirements
• Multi-step agent workflows where tool failures or caching could affect results
• Building, debugging, or hardening agents that must avoid hallucination or unsupported citations

Best practices

• Rank and pre-evaluate search results by relevance, credibility, recency, and authority before reading
• After every tool call, explicitly check for errors and document recovery strategy if needed
• Use structured thinking blocks: what was learned, connection to goal, contradictions/gaps, and next steps
• Maintain a source-tracking table indicating fetched, failed, and cited sources; never cite unretrieved content
• Cross-validate key claims across at least two independent sources and record consensus or contradictions
• Before finishing, run a pre-completion checklist verifying coverage, retrievals, validations, and documented gaps

Example use cases

• Aggregating technical specs from vendor docs where authoritative verification is required
• Long-form literature reviews that must note contradictions and unverifiable claims
• Multi-step data collection pipelines that write and verify files across tools
• Incident investigations needing traceable source retrievals and explicit error handling
• Producing final reports with a ‘Limitations & Gaps’ section documenting failed fetches and unresolved items