unicorn-query

name	unicorn-query
description	Query unicorn behavioral test results from the database. Filter functions by verdict (EQUIVALENT/DIVERGENT), divergence class (logic/build_env/regalloc), and unit. Use to find functions with real behavioral bugs to fix.
argument-hint	[--class logic|build_env|regalloc] [--verdict DIVERGENT|EQUIVALENT] [--unit pattern] [--status workable|complete|all] [--limit N] [--summary-only]
allowed-tools	Bash(python3 scripts/unicorn/query.py *)

Unicorn Query Skill

Query the decomp database for unicorn behavioral test metadata. Surfaces functions by their runtime behavioral equivalence status.

Arguments

$ARGUMENTS

Steps

1. Build the command from the arguments and run:

   python3 scripts/unicorn/query.py $ARGUMENTS
   

   If no arguments were provided, default to --verdict DIVERGENT to show all divergent functions.

2. Present results to the user. Key things to call out:

   • logic class: real behavioral bugs, fixable from source — primary work targets
   • build_env class: unfixable from source (linker/environment differences like __FILE__ strings, merged symbols)
   • regalloc class: rare register allocation edge cases, usually not worth pursuing
   • Functions marked COMPLETE but with logic divergence may have hidden behavioral bugs

Common Queries

• /unicorn-query --class logic — all logic-divergent functions (primary work targets)
• /unicorn-query --class logic --status workable — logic bugs not yet marked done
• /unicorn-query --class logic --unit system/char/* — logic bugs in a specific subsystem
• /unicorn-query --verdict EQUIVALENT --status workable — behaviorally correct but asm not matching yet
• /unicorn-query --summary-only — just show counts by verdict/class
• /unicorn-query --class build_env --summary-only — count of environment-only issues

Note

The unicorn DB columns are in decomp.db. This skill is available for when unicorn behavioral testing is added to the rb3-xenon pipeline.