01Safety
0.0 / 30
What changed in the harness
Selection accuracy 95→100, token cost up 3%, 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
28.6 / 30
03Economics
19.8 / 20
04Discoverability
12.1 / 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 |
|---|---|---|
query |
no_return_description |
Execute a prompt via Claude Code CLI with optional file context and session resume, returning Claude's generated response text plus call metadata (sessionId, model used, turn count, and cumulative cost) that can be passed to a later call to continue the conversation. Claude is an AI coding agent that can generate, analyze, refactor, and explain code. Capabilities: code generation and refactoring, code analysis and explanation, file understanding (text and images), multi-turn conversations via sessionId. Cost: Default model is Sonnet (~$0.01-0.10/call). Use effort="low" for simple tasks, effort="high" + model="opus" for complex analysis. Set maxBudgetUsd to cap per-call cost (recommended for effort="max" or model="opus"). Tips: - Set workingDirectory to the target repo for project-aware responses. - Break complex tasks into focused prompts rather than one large request. - Resume multi-turn conversations with sessionId from a previous response's metadata. - Include relevant files via the files parameter for targeted context (text files inlined in prompt, images trigger allowed-tools mode). - Use noSessionPersistence=true for stateless one-shot calls. |
ping |
no_return_description |
Health check that verifies the Claude CLI is installed and authenticated, returning a status report with CLI and server versions, supported capabilities, and current configuration. No cost (local check only). |
Selection evidence
Confusable tool pairs.
0 pairs where similar names or overlapping descriptions may send an agent toward the wrong tool.
| Tool A | Tool B | Confidence | Why they collide |
|---|---|---|---|
| No confusable tool pairs were flagged in this assessment. | |||
Compare the field