render-interview

Status: v1.2 (calibrated). Research-based calibration complete. RUBRIC.md ships with five cross-source-validated dimensions. A banned_phrases.json (25 entries, each backed by 2+ sources) and research-notes.md (7 cited sources) ship alongside the prompts. Pass/fail fixture corpus under tests/fixtures/interview/.

Use this when

• The user names an interview question and asks for an answer.
• The user pastes a job description and asks for talking points.
• The user says <!-- ko-example -->"STAR" or "behavioural" or "면접" or "기술 면접"<!-- /ko-example -->.

Do NOT use when

• The user wants a full resume → lockedin-render-resume-en.
• The user wants a Korean cover letter → lockedin-render-jaso.
• The vault has no relevant role / project / achievement nodes. Seed first via /lockedin init or by ingesting a resume.

Two-turn pattern

Writer turn produces the draft. Reviewer turn re-loads RUBRIC.md fresh in a separate Claude turn and emits a JSON score. Same as the other renderers; the split is load-bearing.

Output shape

A single markdown answer, no headers. STAR (Situation / Task / Action / Result) by default; PAR (Problem / Action / Result) when the question is incident-shaped. One experience per paragraph with explicit transitions, mirroring the policy in lockedin-render-resume-en and lockedin-render-jaso.

The answer pulls evidence from the vault using slug citation ([[type/slug]]); the slugs are resolved to natural language by lockedin/render/resolve_slugs.py before the artifact is shown to the user.

Files in this directory

SKILL.md             (this file)
research-notes.md    7 cited sources, cross-source analysis summary
banned_phrases.json  25 entries, severity-tagged, each backed by 2+ URLs
prompt-writer.md     writer-turn instruction
prompt-reviewer.md   reviewer-turn instruction (re-loads RUBRIC.md fresh)
RUBRIC.md            5-dimension scoring contract + score bands

Calibration status

v1.2 calibrated. The rubric dimensions (clarity, evidence_density, persona_fit, conciseness, tone) are grounded in cross-source public research from MIT CAPD, The Muse, Indeed, Harvard Business Review, Yale OCS, The Interview Guys, and Big Interview. The banned_phrases.json contains 25 entries across four categories (weak_ownership, trait_claim, rehearsed_non_answer, vague_filler), each backed by 2+ independent sources. Pass and fail fixture corpus is at tests/fixtures/interview/{pass,fail}/ (3 pass, 3 fail).