01Safety
0.0 / 30
What changed in the harness
Selection accuracy 100→100, token cost up 12%, 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
27.6 / 30
03Economics
20.0 / 20
04Discoverability
16.5 / 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.
6 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 |
|---|---|---|
ai-safety-guard |
name_restates_behavior no_return_description |
Return caution instructions an AI agent should follow before calling another MCP. Given the type of MCP (email, slack, database, file, web, general), the operation type (read, write, execute, delete, send, query), and the data sensitivity level (public, internal, confidential, restricted), returns the relevant safety guidance to display to the agent. |
aim-text-guard |
name_restates_behavior no_return_description |
Analyze the given text for harmful content and return an assessment indicating whether harmful content was detected, along with the categories or severity of any flagged content. |
aim-security-prompt-tool |
name_restates_behavior no_return_description |
Return an enhanced version of the given user prompt with added security instructions. The security_level (basic, standard, strict) controls how strongly the security instructions are added, producing a modified prompt suitable for use with downstream models. |
prompt-injection-detector |
no_return_description |
Detect prompt injection attempts based on OWASP LLM01:2025 patterns in the given text, using the selected sensitivity level (low, medium, high), and return whether an injection was detected along with the flagged patterns. |
credential-scanner |
no_return_description |
Scan the given text for exposed credentials such as API keys, passwords, tokens, and SSH keys, and return a list of findings. When mask_findings is true, detected credential values are masked in the output. |
url-security-validator |
no_return_description |
Validate the given URL for safety, checking for phishing, malware, and HTTPS enforcement, and return a safety verdict with supporting reasons. When strict_mode is enabled, stricter security checks are applied. |
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 |
|---|---|---|---|
ai-safety-guard |
aim-text-guard |
low | Both names end in 'guard' and a vague task like 'guard against unsafe content' could surface both, but the descriptions are clearly distinct: aim-text-guard analyzes text for harmful content while ai-safety-guard issues MCP caution instructions and accepts an mcp_type/operation_type/sensitivity_level schema with no text parameter, which an agent inspecting inputs would see. Confusion is possible but unlikely. |
Compare the field