lockedin-render-resume-en

render-resume-en

Research-based calibration. Ships with full rubric, writer and reviewer prompts, and a banned-phrase regex list. Dimension definitions derived from cross-source consensus across 20+ US tech resume guides. See research-notes.md for citations.

Use this when

• User asks for an English resume targeting a tech / PM persona.
• User wants their existing resume "polished" against the rubric.

Do NOT use when

• User wants a Korean cover letter → render-jaso.
• The vault has no project / role / achievement nodes yet → seed first.

Required design constraints

• Metric-first bullets — every bullet contains a number (%, x, $, count, or duration). Rubric enforces ≥80% metric density via regex.
• XYZ or CAR per bullet — XYZ = "Accomplished X as measured by Y, by doing Z"; CAR = Challenge / Action / Result compressed to one bullet line. Active voice, quantified result. (STAR is the implicit story arc; XYZ/CAR is the bullet shape.)
• Active voice — banned: "was responsible for", "helped to", "worked on", "was involved in".
• No keyword stuffing — ATS-friendly via real verbs and metrics, not hidden keywords.
• Target persona — --target us-tech-senior, us-tech-mid, pm-product each have tuned tone and bullet density.

Two-turn pattern

Same writer/reviewer split as render-jaso:

1. Writer turn produces the resume markdown.
2. Reviewer turn re-loads RUBRIC.md fresh, runs the metric-density regex, scores action-verb diversity, ATS keyword coverage, vagueness banlist. Emits JSON.

Final checklist

• Metric-density regex passed (≥80% bullets contain a number).
• Reviewer turn was a separate Claude context with fresh RUBRIC.md load.
• Concrete ontology slugs quoted (project / role / achievement).
• Active voice; no banned phrases.

Files in this directory

SKILL.md
research-notes.md     citations with URL + ISO date + 2-sentence gloss
RUBRIC.md             5 dimensions; score bands; fixture authoring guide
prompt-writer.md      writer turn instruction
prompt-reviewer.md    reviewer turn instruction (separate Claude context)
banned_phrases.json   regex list of weak / vague / templated phrases