01Safety
0.0 / 30
What changed in the harness
Selection accuracy 100→98, token cost up 1%, unconfirmed writes 0%→0%.
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
29.5 / 30
03Economics
20.0 / 20
04Discoverability
8.6 / 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.
1 defect 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 |
|---|---|---|
growthbook_list_skills |
no_return_description |
List available GrowthBook agent skills and return each skill's name plus a short description of what it does. Use growthbook_read_skill to load the full workflow for a skill before acting; skills encode how to accomplish GrowthBook tasks well, and use growthbook_call_api to execute the REST calls they describe. |
Selection evidence
Confusable tool pairs.
1 pair where similar names or overlapping descriptions may send an agent toward the wrong tool.
| Tool A | Tool B | Confidence | Why they collide |
|---|---|---|---|
growthbook_list_skills |
growthbook_read_skill |
medium | A user asking to 'read' or 'show' a skill (e.g., 'show me the flag-create skill contents') could trigger the wrong member: 'read the skills' is plausibly interpreted as listing skill names rather than loading a specific skill's workflow, and agents may skip the required list-first step when a user names a skill directly. The descriptions distinguish names+descriptions from full content, but the verb overlap ('list' vs 'read'/'show'/'get') makes the choice genuinely ambiguous for vague requests. |
Compare the field