메뉴
Use when the user wants to audit, validate, or make a Cognigy voice agent production-ready against the Voice Go-Live Checklist, or asks what audit_voice_agent checks and which items are auto-fixable vs manual.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when adding custom logic inside a Cognigy tool branch with manage_flow_nodes — supported node types, config schemas, placement rules, and the tool-first workflow.
Use when the user wants to add xApps to a Cognigy flow — interactive HTML / Adaptive Card mini-apps rendered to the user, with the submitted data flowing back into the conversation. Covers the xApp flow nodes, the Init Session rule, and accessing xApp data.
Add a new MCP tool (or extend an existing one) in the NiCE Cognigy Plugin's MCP server
Use when the user wants to build, create, or set up a new Cognigy AI Agent from scratch — covers listing projects, ensuring an LLM exists, creating the agent, testing it, and refining persona/job fields.
Use when the user wants to add knowledge, RAG, or a knowledge store/source to a Cognigy agent — covers embedding vs Knowledge Search models, Knowledge AI settings, ingesting sources, and attaching knowledge as a tool.
Use when configuring or choosing an LLM for a Cognigy agent — valid provider names (openAI, anthropic, azureOpenAI, google, mistral), model strings, connection types, and credential resolution.
| name | voice-go-live-checklist |
| description | Use when the user wants to audit, validate, or make a Cognigy voice agent production-ready against the Voice Go-Live Checklist, or asks what audit_voice_agent checks and which items are auto-fixable vs manual. |
audit_voice_agent checks a voice AI agent against the deterministic subset of the Voice AI guidebook Go-Live Checklist (section 3) and can apply safe fixes. It reads the flow chart (and optionally the endpoint and LLM) over the Cognigy REST API, so it only covers what the API can both read and change. Speech-provider status, failover behaviour, and release-readiness items live in the external Voice Gateway app or are runtime/operational — those are manual and out of scope.
1. Dry-run report (no changes):
audit_voice_agent { aiAgentId }
audit_voice_agent { aiAgentId, endpointId, projectId } // also checks Output Transformer + fallback LLM
2. Apply all safe fixes:
audit_voice_agent { aiAgentId, apply: true }
3. Apply a subset:
audit_voice_agent { aiAgentId, apply: true, only: ["vg.barge-in-off", "agent.stream-output"] }
aiAgentId (the flow is resolved for you) or flowId directly.endpointId enables the Output Transformer check. projectId enables the fallback-LLM check.apply: true.aiAgentId (same resolution as the other agent tools) or the flowId you pass.setSessionConfig.aiAgentJob.endpointId / projectId are supplied.| Status | Meaning |
|---|---|
pass | Meets the recommended setting. |
fail | Deviates from the recommended default. Usually auto-fixable. |
warn | Advisory — needs human judgement (e.g. hint content), not auto-fixed. |
na | Not applicable / not enough data (e.g. no Set Session Config node yet, or no endpointId). |
Each fixable check carries a proposedFix in the dry-run report. summary counts pass/fail/warn/na and how many are fixable.
apply: true)| ID | Section | Checks | Fix |
|---|---|---|---|
vg.session-config-first | 3.1 | First node is a Set Session Config node | Creates a setSessionConfig node with recommended defaults, prepended before the AI Agent node |
vg.barge-in-off | 3.1 | bargeInOnSpeech and bargeInOnDtmf off | Sets both to false |
vg.continuous-asr-off | 3.1 | asrEnabled off (for Flux) | Sets asrEnabled = false |
vg.user-input-timeout | 3.1 | userNoInputTimeout 5000–7000 ms, userNoInputRetries ≥ 5 | Sets enable=true, 6000 ms, 5 retries |
vg.flow-input-timeout | 3.1 | flowNoInputTimeoutEnable on, ~1500 ms, filler phrase set | Enables, 1500 ms, default filler if missing |
vg.flow-fails-on-error-off | 3.1 | flowNoInputFail off | Sets flowNoInputFail = false |
agent.stream-output | 3.1 | AI Agent storeLocation = "stream" | Sets storeLocation = "stream" |
agent.fails-on-error-off | 3.1 | AI Agent errorHandling ≠ "stop" | Sets errorHandling = "continue" |
agent.error-message | 3.2 | AI Agent errorMessage non-empty | Sets a default voice error message |
agent.log-llm-latency | 3.2 | AI Agent debugLogLLMLatency on | Sets debugLogLLMLatency = true |
vg.silence-overlay-delay | 3.6 | silenceOverlayDelay = 0 when overlay configured | Sets delay to 0 (only when overlay is in use) |
The Set Session Config value checks report na until a Set Session Config node exists. Creating it with vg.session-config-first seeds recommended defaults that already satisfy them.
| ID | Section | Why manual |
|---|---|---|
vg.stt-hints | 3.3 | Cannot invent brand/domain terms; flags when sttHints is empty. |
endpoint.output-transformer | 3.7 | Confirms the transformer is enabled and non-empty; the markdown-stripping code is not auto-generated. |
llm.fallback | 3.2 | Adding a fallback requires choosing a specific model. |
These checklist items are not reachable through the Cognigy REST API and must be handled manually:
apply: true returns appliedFixes plus a fresh re-audit. Always verify the flow in the UI — particularly node ordering when a Set Session Config node was created (it is prepended before the AI Agent node on the flow's next chain).