01Safety
0.0 / 30
What changed in the harness
Selection accuracy 80→72, token cost up 2%, 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
25.6 / 30
03Economics
20.0 / 20
04Discoverability
8.5 / 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.
2 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 |
|---|---|---|
collect_base |
no_return_description |
Run the base diagnostic group on a host — docker_ps, df, df_inode, free, uptime, dmesg_oom, docker_daemon — as a cheap first look, and return the commands' redacted, truncated output. |
container_deep_dive |
params_unexplained no_return_description |
Run docker_logs, docker_inspect (redacted), container_state, and docker_stats for a single container on a host, and return their redacted, truncated output. The container parameter is the name of the target container to diagnose on that host. |
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 |
|---|---|---|---|
collect |
collect_base |
low | Descriptions clearly separate the generic command runner from the fixed base-group shortcut, so an agent reading them is unlikely to pick wrong. Only a vague task like "collect base host info" could blur them, but collect_base's explicit 'shortcut for the base group' wording disambiguates. |
Compare the field