01Safety
0.0 / 30
What changed in the harness
Selection accuracy 80%, destructive-action safety rate 0% (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
25.0 / 30
03Economics
18.8 / 20
04Discoverability
15.9 / 20
Highest-impact fix
Estimated gain +30 pointsAdd explicit identity and permission preflight tools
Expose machine-readable principal/tenant confirmation and a non-mutating permission check so agents can verify both before destructive actions.
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.
5 pairs where similar names or overlapping descriptions may send an agent toward the wrong tool.
| Tool A | Tool B | Confidence | Why they collide |
|---|---|---|---|
reasoning |
adaptive-reasoning |
medium | adaptive-reasoning states 'Same triggers as reasoning' and lists the identical trigger phrases; the only differentiators are tier requirement and a 'use when generic is insufficient' judgment call, so a plain analytical task can plausibly land on either. |
code |
adaptive-code |
medium | adaptive-code states 'Same triggers as code' and reuses the same trigger list; the discriminator is only severity/tier, so a normal code task could plausibly pick either member. |
memory |
adaptive-memory |
medium | adaptive-memory states 'Same triggers as memory' with identical perception-layer triggers and the same 'observe FIRST' framing; the base-vs-adaptive choice rests on sharpness/stakes judgment, leaving real ambiguity. |
anti-deception |
adaptive-anti-deception |
medium | adaptive-anti-deception states 'Same triggers as anti-deception' and mirrors the signal list; an integrity-tension task triggers both, and nothing in the definitions decisively picks one for a mid-stakes case. |
adaptive-reasoning |
adaptive-code |
medium | Trigger lists overlap on diagnostic language: reasoning lists 'diagnose'/'root cause' and code lists 'debug this error'/'what's wrong with this code', and both schemas even share the identical example 'diagnose why a microservice returns 503s under load', so a bug-diagnosis task is plausibly routed to either. |
Compare the field