Superdialog Playbook Skill — Development skill for Claude Code

superdialog-playbook-skill

[](https://github.com/ankitai-s/superdialog-playbook-skill) [](https://github.com/ankitai-s/superdialog-playbook-skill) [](LICENSE) [](playbook/references/checklist.md)

A skill for building, reviewing, and fixing **SuperDialog voice-agent playbook YAMLs** — the `persona` / `llm` / `journeys` / `checkpoints` / `interrupts` / `policies` / `knowledge_base` schema that drives a Director+Talker voice agent.

Works identically in **Claude Code** and **Codex** — both load the same `SKILL.md` + `references/` format natively, no conversion needed.

---

How a playbook actually runs

Every turn, the **Director** (a fast/cheap LLM) reads the transcript and decides whether to advance the checkpoint, fire an interrupt, or stay — then the **Talker** (usually a stronger LLM) speaks from whatever checkpoint the Director landed on. Almost every hard bug in a playbook traces back to something left to Director/Talker *judgment* that should have been made *structural* instead.

sequenceDiagram
    autonumber
    actor Caller
    participant Runtime
    participant Director as Director LLM
(judge: llm | expr)
    participant Talker as Talker LLM

    Caller->>Runtime: speaks this turn
    Runtime->>Director: evaluate(transcript, current checkpoint)
    alt slot filled / expr condition met
        Director-->>Runtime: advance -> next checkpoint
    else caller said something matching an interrupt
        Director-->>Runtime: interrupt -> target checkpoint
    else nothing resolved yet
        Director-->>Runtime: stay
    end
    Runtime->>Talker: current checkpoint's guidance + slots
    Talker-->>Caller: spoken