What problem does it solve? Research findings in a growing knowledge base often exist in isolation, missing connections to related findings that enable, contradict, extend, or solve the same problem. Manually reviewing thousands of possible pairs to build these relationships is impractical, and naive linking produces false positives that pollute the knowledge graph. ## Core Features & Use Cases - Four Binary-Tested Relationship Types: Classifies links as enables, contradicts, extends, or same-problem using strict two-question tests, with anti-pattern guidance to reject false positives. - Parallel Subagent Evaluation: Batches candidate pairs (default 25 per batch) and dispatches concurrent subagents to evaluate relationships at scale across 24K+ pair spaces. - Human-Gated Writes with Validation: Produces a proposal report first, writes approved links to YAML frontmatter via a safe parser, then runs a mandatory post-write validation pass on stratified samples. - Use Case: After backfilling 20 new findings into a category, run the skill with --category to evaluate only affected pairs, review the proposal report, approve links, and let it write bidirectional related_findings entries with validation. ## Quick Start Run the finding-crosslink skill with --dry-run to generate a proposal report of candidate relationships across the knowledge base before approving any writes.
