Agent skill · alirezarezvani
interview
Phase 1 of building a Claude Managed Agent — interview the founder about the one job the agent should do, then produce a build sheet (CMA primitives table + v1/v2 deferrals + eval plan) WITHOUT needing their API key yet. Use when the user says "help me scope an agent", "I have an idea for an agent", "what should this agent be", or when the orchestrator routes phase=interview. Drives the six intake slots (job, trigger, inputs, actions, definition-of-done, recurrence) via AskUserQuestion, maps them to primitives with interview_planner.py, assembles build-sheet.json with build_sheet_builder.py, and validates limits with primitives_validator.py. Connectors are mockable in v0 (schema-true custom tools); real MCP servers become v1 deferrals. Distinct from stage-launch (which turns the sheet into payloads).
What it needs
About 2k tokens when loaded.
What this skill does
Phase 1 — Interview → Plan Open warmly with one or two examples from ../../references/examples-bank.md, then interview the founder into a build sheet. No API key needed in this phase — the output is a plan. The six intake slots (ask one at a time; use AskUserQuestion for choices) Slot Question Maps to --- --- --- Job "What one job should this agent do end-to-end?" agent.system + outcome description Trigger "What kicks it off — you ask it, an event, or a schedule?" on-demand / event / cron Inputs "What does it read?" (files, repo, memory, gmail/slack/github, web) resources / MCP servers / memory Actions "What does it do?" (draft, write, call APIs, run code) agent toolset / custom tools / MCP Done "How would you grade a good run?" outcome rubric (required) Recurrence "Once, on request, or on a cadence?" single-pass / grade-loop / cron-loop See ../../references/interview-to-config.md for the full mapping. Workflow 1. Interview. Walk the six slots. Capture the founder's own words — never invent specifics they didn't claim. 2. Map to primitives. MCP inputs become schema-true mock custom tools in v0 and a v1 deferral to wire the real server. Irreversible actions (send/publish) become v2 deferrals behind alwaysask. 3. Assemble the sheet. 4. Validate limits. FAIL blocks progress; fix and re-run. WARN is advisory (surface it). 5. Record the plan in the goal. goalstate.py set --phase stage-launch --artifact buildsheet=./my-agent/build-sheet.json, then advance. Hard rules v0 is the core job only. Everything else is a versioned deferral with a reason and an exact mechanism. Their problem, their words. No key yet. The interview produces a plan; the key is a Phase-2 concern. Forcing-question library (recommend + cite) 1. "What one job — singular?" Recommend: the most-repeated task. Cite: interview-to-config.md. Two jobs → two agents. 2. "Real integration or v0 mock?" Recommend: mock; wire MCP as v1. Cite: interview-to-config.md rule 1. 3. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills alirezarezvani/interview--3bb2ee