render-ideas

Status: v1.2 (calibrated). Research-based calibration complete. RUBRIC.md ships with five cross-source-validated dimensions. A banned_phrases.json (27 entries, each backed by 2+ sources) and research-notes.md (7 cited sources) ship alongside the prompts. Pass/fail fixture corpus under tests/fixtures/ideas/.

Use this when

• The user asks "what should I work on next" or "ideas for side projects" or names a constraint and asks for proposals.
• The user wants to surface novel combinations of their existing skills and domains.
• The user asks for career-move pitches grounded in their actual experience.

Do NOT use when

• The user wants a resume or cover letter → render-resume-en / render-jaso.
• The user wants an interview answer → render-interview.
• The vault is sparse. Seed first via /lockedin init or ingestion.

Two-turn pattern

Writer turn surfaces 3 to 5 ideas. Reviewer turn re-loads RUBRIC.md fresh and scores the set on five dimensions.

Output shape

3 to 5 ideas, each one paragraph long, with a one-sentence pitch followed by two to three sentences of rationale that cite vault entities. Slug citations ([[type/slug]]) are resolved to natural language by the orchestrator before the user sees the artifact.

Files in this directory

SKILL.md             (this file)
research-notes.md    7 cited sources, cross-source analysis summary
banned_phrases.json  27 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 (feasibility, novelty, evidence_ground, scope_match, motivation_alignment) are grounded in cross-source public research from Atlassian, Inc., FasterCapital, Proposify, Built In, fundsforNGOs, and arXiv. The banned_phrases.json contains 27 entries across five categories (buzzword_opener, vague_enthusiasm, hedging_language, unsubstantiated_claim, comparison_shortcut), each backed by 2+ independent sources. Pass and fail fixture corpus is at tests/fixtures/ideas/{pass,fail}/ (3 pass, 3 fail).