nikhail bench — skill-activation
router gpt-5.6-terra · effort high · fixer claude-fable-5 · gbrain 3cc34c92eec2 · 150 intents × k=5 · 0 errors · tokens: 3.5M in, 0% cached · 72.3k (45.4k reasoning) out · ≈$9.85 est. · generated 2026-07-28 10:59
Accuracy by intent source
arm B — original descriptionsarm C — linter findings applied
Paired outcome (per intent)
C better on 22 · B better on 12 · 116 tied · sign-test p = 0.12145
Top regressions (C worse)
| intent | expected | B /5 | C /5 | Δ |
|---|
| I need to improve this skill's Anti-Patterns section, can you help? | skill-creator | 3 | 0 | -3 |
| Check if any pending migrations or stale embeddings need attention today. | skillpack-check | 3 | 0 | -3 |
| I need to file this in the brain — what path? | brain-taxonomist | 4 | 2 | -2 |
| what's the weather in SF this weekend | (null) | 3 | 1 | -2 |
| Turn my notes at path/to/page into a clean shareable document I can send over Telegram | brain-pdf | 4 | 3 | -1 |
Top improvements (C better)
| intent | expected | B /5 | C /5 | Δ |
|---|
| Run a skill health check and tell me what's broken | testing | 0 | 5 | +5 |
| I wrote a script for parsing logs, does it need tests and a resolver entry? | skillify | 0 | 5 | +5 |
| Give me a health check on how my notes system is configured | gbrain-advisor | 0 | 5 | +5 |
| Check my knowledge base setup and tell me what's wrong with it | gbrain-advisor | 0 | 5 | +5 |
| let's DRY this up across our skills | eiirp | 0 | 5 | +5 |