toolcalling.dev

Dashboard Tasks MCP servers Data Settings

decoys only

loading…

decoys only · no real tool is ever executed · $0 with the mock provider

Scenarios Runs Traces Exports

Trace 5nv9serg

Prompt_Injection_Defense · no_call · dotted·terse · attempt 2→ run inlgaccq

approvedG re-judge · L loop · E edit · J/K queue

lineage: bsb9db → v9serg

Transcript (3 turns)

system

You are a helpful assistant with access to Prompt_Injection_Defense tools. Error behavior: retry once on 5xx/timeout, then explain & stop

user

help me with my Prompt_Injection_Defense request

assistant

What specific aspect of prompt injection defense do you need help with - scanning input, reviewing detections, updating rules, or something else?

Hard validators ($0 gate)

schema ok
referential ok
grounding ok

Judge verdict

score 4

schema: 5
arguments: 5
coherence: 5
tool_choice: 4
plausibility: 4
error_recovery: 5

The schema is correct and the arguments are plausible. The assistant correctly asks for clarification on which tool to use, showing understanding of the available tools. However, a more proactive suggestion or a specific tool choice might have improved the trace.