01Safety
30.0 / 30
What changed in the harness
Selection accuracy 96%, destructive-action safety rate 100% (baseline only -- no rewrite pass applied).
Category breakdown
Where the score comes from.
Earned points across the four signals Gradable measures. Safety and Legibility are scored out of 30; Economics and Discoverability are scored out of 20.
02Legibility
28.6 / 30
03Economics
20.0 / 20
04Discoverability
18.9 / 20
Highest-impact fix
Estimated gain +1 pointDisambiguate overlapping tool choices
Give confusable tools explicit decision boundaries and examples of when to choose one instead of the other.
Description evidence
Defects and rewrites.
0 defects found across the exposed tool descriptions. Suggested rewrites make purpose, inputs, boundaries, and returns easier for an agent to understand.
| Tool | Defect types | Suggested rewrite |
|---|---|---|
| No description defects were flagged in this assessment. | ||
Selection evidence
Confusable tool pairs.
3 pairs where similar names or overlapping descriptions may send an agent toward the wrong tool.
| Tool A | Tool B | Confidence | Why they collide |
|---|---|---|---|
lookup_icd10 |
search_icd10 |
medium | A task phrased 'look up the ICD-10 code for type 2 diabetes' mirrors the word 'lookup' but actually needs reverse lookup (condition→code); an agent could pick lookup_icd10, which requires a code rather than a condition. |
lookup_icd10 |
validate_icd10 |
medium | Both tools accept a code and both can return a description, so a task like 'check code E11.9 and tell me what it means' plausibly maps to either; validate emphasizes existence while lookup emphasizes the official description. |
validate_icd10 |
search_icd10 |
low | A task like 'find/confirm the correct valid code for this condition' is ambiguous, but the schemas diverge sharply (validate needs an existing code, search needs a condition query), making wrong selection less likely when no code is supplied. |
Compare the field