Skip to content

README.md

Latest commit

 

History

History
116 lines (91 loc) · 3.96 KB

README.md

File metadata and controls

116 lines (91 loc) · 3.96 KB
Raw

Interview prompt authoring

Interview packets are Markdown files in server/app/interviews/prompts/. Every packet uses the same comprehensive formula so the candidate experience, canvas context, pacing, evaluation, and safety boundaries stay consistent while the system-design content remains specific to the product.

Required front matter

---
display_name: My System Design
skill_level: Intern
---

Validation requires:

  • Front matter starts on the first line, contains exactly display_name and skill_level, and contains no extra fields.
  • The Markdown filename stem is the packet ID and must be unique lowercase kebab-case, such as my-system-design.md.
  • A packet with the tinyurl-system-design ID remains present.
  • display_name and the Markdown body are nonempty.
  • skill_level is Intern, Junior, or Senior.

Prompt formula and checklist

Use these sections in this order:

  1. Personality — Define Lyra's expertise and supportive, candidate-led posture.
  2. Environment — Preserve the shared Keyframe/ElevenLabs context and the identical React Flow canvas serialization semantics and connection syntax. A canvas update is background context, not a request to speak.
  3. Goal — State what the candidate is designing and what reasoning the interviewer evaluates.
  4. Interview flow — Include the topic's exact opening, a concise explanation for candidates unfamiliar with the product, candidate-led pacing, a topic-specific core/MVP completion definition, direct handling of candidate questions, and silence/vagueness/off-topic/solution/conflict/early-stop edge cases.
  5. Voice response style — Keep normal turns to one brief acknowledgment and one focused question; do not read lists aloud.
  6. Candidate-facing reference — Separate functional requirements, non-functional requirements, and concrete scale assumptions. Give this information directly when asked without revealing a solution.
  7. Interviewer question bank — Supply optional clarifying, topic, risk, and allowed-guidance-boundary questions tailored to this packet.
  8. Private interviewer reference — Document multiple plausible solution families and a strong design direction for evaluation only.
  9. Evaluation and closing — Name the relevant evaluation dimensions and a concise feedback/closing sequence grounded in the candidate's actual design.
  10. Guardrails and Critical reminder — Never reveal the private answer during the candidate-led portion or provide a mitigation before the candidate reasons about it.

All content must remain topic-specific. Do not copy TinyURL's SQL/caching probe, product facts, or exact NoSQL closing into another packet.

Hard 10-minute constraint

Every interview is a hard 10-minute session enforced by the application. The prompt must tell Lyra to prioritize the core/MVP, ask one focused question at a time, stop opening new topics as time runs out, and close promptly. Lyra cannot read or infer the UI countdown, so never instruct the model to announce exact time remaining. The UI starts its countdown only after the avatar first connects and disconnects Lyra at 00:00.

Template skeleton

# Personality

# Environment
## Canvas context

# Goal

# Interview flow
## Opening
## If the candidate is unfamiliar with the product
## Candidate-led exploration and 10-minute pacing
## Core/MVP completion and closing trigger
## Answering candidate questions
## Edge cases

# Voice response style

# Candidate-facing reference
## Functional requirements
## Non-functional requirements
## Scale assumptions

# Interviewer question bank
## Clarifying questions
## Topics to probe
## Risk-focused probes
## Examples of the allowed guidance boundary

# Private interviewer reference
## Possible solution families
## Strong design direction

# Evaluation and closing

# Guardrails

# Critical reminder

Validate all packets from the repository root:

pnpm interview:validate