Verified Agent Skill record

Comprehensive Research Agent

This skill addresses common failures in multi-step research tasks—unhandled tool errors, missing validation, opaque reasoning, and premature conclusions—by providing structured protocols for source validation, error recovery, and thinking transparency to improve research quality and reliability.

Tasks

Conduct multi-source web research using search, read_url, or fetch operations requiring 3+ tool interactions, applying pre-reading source ranking and post-reading validation checkpoints
Execute file operations (save, write, read) and verify content using read_file instead of list_directory to avoid caching false negatives

Inputs

Web research search queries
Source URLs identified for reading
Tool responses containing search results, fetched content, or error messages

Outputs

Structured thinking blocks documenting what was learned, source connections to the goal, contradictions/gaps, and strategic decisions
Pre-completion validation checklist confirming requirements coverage, source retrieval status, claim validation, and documented gaps
Final research report including a 'Limitations & Gaps' section documenting failed attempts and unverified claims

Limitations and checks

Complex research tasks with multiple tools (6+) typically achieve scores in the 65-75 range due to inherent variability in tool outputs, multiple valid reasoning paths, and stochastic nature of long-horizon agent execution
Score improvement focus is relative (5-10% is significant) rather than absolute, reflecting the difficulty of complex multi-step tasks
Verify that every thinking block after a tool call includes 3-5+ sentences covering: what was learned, connections to the goal, gaps/contradictions, and next steps
Confirm that a source tracking table exists and no cited URLs remain unretrieved (prohibit citation of sources never successfully fetched)
Ensure that all file write verifications use read_file rather than list_directory to avoid caching false negatives
Check that the final report includes a 'Limitations & Gaps' section documenting what was attempted but failed or remains unverified
Validate that key claims are cross-validated across at least 2 sources where possible, with explicit notes on consensus or contradictions