Loading ToolBench
45 tools with mixed quality. Naming is strong (verb_noun pattern, clear intent). Descriptions present but often terse (avg ~60 chars, well below 194-char baseline). Critical gap: input schemas are inferred from code, not explicitly visible in tool registration. Most tools lack documented output schemas. Parameter descriptions exist but are minimal. Error handling and recovery guidance largely absent. No tool annotations (readOnlyHint/destructiveHint) despite risk classifications in metadata.
Abort currently running operation
Adjust orbit periapsis and apoapsis
Change longitude of ascending node
Change orbit shape by adjusting eccentricity
Change orbital inclination to target angle
Rotate orbit orientation by changing periapsis longitude
Change orbital period by adjusting semi-major axis
Circularize current orbit at apoapsis or periapsis
Input schemas inferred from code, not explicitly visible in tool registration. Cannot verify parameter types, constraints, or descriptions in source. Per HARD SCORING RULE, caps schema scores at 35-40 max.
No output schemas documented. Tools return data but LLMs cannot plan downstream calls or extract required fields. Violates pattern:tool and mxe:response-field-naming.
Descriptions are terse (avg ~50 chars, baseline 194 chars). Many lack context for when/why to use tool. E.g., 'Warp to target time or event' does not explain prerequisites or side effects.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | <=2025-11-25 | v2 |
Delete all maneuver nodes
Clear navigation target
Run raw kOS command
Configure landing parameters
Continue a paused operation
Fine-tune approach trajectory to target
Avoid crash by raising periapsis
Disconnect from kOS server
Set both apoapsis and periapsis to create elliptical orbit
Execute next maneuver node with automatic warp and steering
Find suitable landing site on target body
Get predicted landing location
Get current target information
List available targets (bodies and vessels)
Transfer to target body using Hohmann transfer orbit
Match target plane or set inclination angle
Transfer to another planet with automatic window waiting
Start/abort/status landing (action: start, abort, status)
Launch into orbit with automatic staging, gravity turn, and fairing deployment
List available CPUs on connected kOS server
List available saved games
List available ships in current save
Load a saved game
Load a ship from the VAB/SPH
Align orbit plane with target
Match velocity with target vessel
Create a quicksave
Create resonant orbit with target body
Return from moon to Kerbin with specified periapsis
Run a kOS script from archive
Set surface landing target by latitude and longitude
Set navigation target by name
Get ship info: orbit, fuel, position, and current operation status
Switch to a different CPU
Switch to a different vessel
Auto-detect transfer type (moon/planet/return)
Warp to target time or event (soi, apoapsis, periapsis, etc.)
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite risk classifications in metadata (WRITE, DESTRUCTIVE, READ_ONLY). LLMs cannot infer safety properties.
No error recovery guidance. Tools fail silently or return raw errors. E.g., 'transfer' auto-detects type but no guidance if detection fails or target unreachable.