{"schemaVersion":"agentdepot.eval_plan.v1","asOf":"2026-10-28","storesIpAddresses":false,"query":"security","matchMode":"matched_task","taskType":"security","label":"Adversarial Red-Team, Prompt-Injection & Safety Guardrail Evaluation","recommendedBenchmarks":[{"id":"tau-bench","name":"τ-bench (Tau-bench)","officialUrl":"https://github.com/sierra-research/tau-bench","graderType":"Final DB state + policy pass^k check"},{"id":"bfcl","name":"Berkeley Function-Calling Leaderboard (BFCL)","officialUrl":"https://gorilla.cs.berkeley.edu/leaderboard.html","graderType":"AST structural match + live execution check"}],"depotBenchmarkSuite":{"type":"security_guardrail","tasksApi":"/api/benchmark/tasks?type=security_guardrail","submitApi":"/api/benchmark/runs"},"stackPicks":{"evalHarness":{"jtbdId":"adversarial-red-team-evals","pickProviderId":"promptfoo","coPickProviderId":"patronus","runnersUpIds":["patronus","confident-ai"]},"sandboxOrRunner":{"jtbdId":"production-guardrail-evals","pickProviderId":"galileo","coPickProviderId":"braintrust","runnersUpIds":["braintrust","maxim"]},"ciGate":{"jtbdId":"continuous-eval-ci","pickProviderId":"braintrust","coPickProviderId":"promptfoo","runnersUpIds":["promptfoo","confident-ai"]},"judgeCalibration":{"jtbdId":"llm-as-judge-calibration","pickProviderId":"braintrust","coPickProviderId":"patronus","runnersUpIds":["patronus","ragas"]}},"targetMetrics":["attack block rate %","false-positive refusal rate on benign prompts","zero PII/secret exfiltration","guardrail p95 latency ms"],"antiContaminationRule":"Rotate adversarial attack templates each release and test benign lookalike prompts so over-refusal is penalized.","nextCalls":["/api/decide?jtbd=adversarial-red-team-evals","/api/benchmark/tasks?type=security_guardrail","/api/playbooks/agent-evals"]}