メニュー
Conduct an inbound voice screen with a candidate via ACS / GPT-Realtime. Ask 4 calibrated rubric questions, abort on red-flag responses, score the transcript, return a shortlist signal. Per spec §4.5.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/arturcrmbot/zava-control-plane --skill voice-screenerコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Sub-skill of `compose-domain`. Writes ONE in-process Python synthetic MCP adapter to a sandbox path, shape-isomorphic to existing tools under `api/server/mcp_tools/`. The adapter is callable from a runtime SKILL via `allowed-tools` and emits OTEL spans + FleetEvents the same way real tools do. Returns deterministic synthetic data — no real backend. It is functional demonstration scaffolding: a real customer adapter replaces it at the same tool boundary without changing workflow, governance, or evidence contracts.
Current design-time meta-skill. Given a domain brief (YAML) or a free-text idea, produce a complete Durable-fidelity domain sandbox: orchestrator, per-phase graphs, validators, agent skills, synthetic MCP adapters, persona(e), + a graduate.sh script that mechanically wires everything into the live trees. The current contract encodes: durable.* event vocabulary, workflow_type stamping on every checkpoint, persona/external_event/context on every HITL gate, persona SKILL.md with executable decision_policy, and the per-domain ramp-loop registration. Every generated domain inherits the contract by construction. Sandbox-only; never touches real trees directly. Calls `author-runtime-skill`, `author-mcp-tool`, `author-persona`, and `author-durable-domain` as sub-skills.
Add a new business domain (workflow_type) to the Zava control-plane substrate. Triggered by requests like "add a new domain", "compose a domain", "design a new workflow", "let's build the X domain", or anything that proposes a new corporate-function automation that should land alongside expense-claim, hiring, vendor-kyc, etc. End-to-end recipe: brief → compose-domain sandbox → graduate.sh → validate active pack → VERTICAL-PROOF.md.
SOC 職業分類に基づく
SKILL.md を表示中
| name | voice-screener |
| description | Conduct an inbound voice screen with a candidate via ACS / GPT-Realtime. Ask 4 calibrated rubric questions, abort on red-flag responses, score the transcript, return a shortlist signal. Per spec §4.5. |
| allowed-tools | acs_dial, transcript_score |
You are the voice-screener step in the POC2 hiring orchestrator (Phase 6).
A candidate_id, the JD's key_competencies list, and a phone_number (from the candidate's application).
acs_dial(phone_number, prompt) where prompt is the 4-question rubric script (see below). The mock returns a transcript; the cloud target is GPT-Realtime over ACS.transcript_score(transcript, rubric) for a 0.0–1.0 score per question.{
"candidate_id": "C-001",
"completed": true | false,
"abort_reason": null | "comp_above_band" | "unprofessional" | "refusal",
"scores": {"motivation": 0.0, "technical": 0.0, "soft": 0.0, "comp_fit": 0.0},
"overall": 0.0,
"verdict": "advance" | "drop",
"transcript_url": "acs://transcripts/..."
}
verdict == "drop" short-circuits subsequent phases; the orchestrator records
this as a Phase-6 termination, not an auto-drop.