interview-speclisted
Install: claude install-skill richfrem/agent-plugins-skills
# Interview Spec (`interview-spec`)
## Purpose
Acts as the universal front-door intake for non-trivial engineering tasks across all supported AI agent runtimes. Enforces Proposal Mode (strictly read-only) before code implementation:
1. **Native-First Deferral:** Detects active host runtime capabilities and defers to native environments while strictly enforcing conversational cadence.
2. **Socratic Pacing (ONE Question at a Time):** When interrogating requirements, the agent MUST ask only ONE question per turn with structured options and an explicit `[Recommended]` default. Never dump multiple questions simultaneously or answer on behalf of the user.
3. **Draft Spec & Implementation Plan Compilation:** Compiles the agreed requirements into a draft `TASK_SPEC.md` and `implementation_plan.md` in state `DRAFT_PLAN`.
4. **Multi-Agent Review Stage Gate (User-Controlled):** After draft spec compilation, explicitly asks the user whether they want to generate an external review bundle (via `context-bundler`) for multi-model critique in browser, or skip directly to approval.
## Stage-entry question contract
The authoritative stage-entry contracts are in `scripts/control_plane/transition_templates.yaml`
under `stages`. A transition template's `human_questions` are edge-approval questions; they
do not replace the questions for the state being entered.
When entering a state:
1. Load that state's `stages.<STATE>` contract before asking anything.
2. Ask every `entry_questions` item in