resume-generatelisted

# resume-generate ## Overview Three-phase resume generation engine with parallel template×style dispatch. Phase 1 plans content and selects projects via a contribution-weighted algorithm. Phase 2 fans out subagents (max 6 parallel) across the template×style matrix. Phase 3 assembles and validates each variant independently. Default output is 3-page resumes across all templates and all styles. ## Architecture ``` Phase 1: Planning Pass (once per target) ↓ resume-plan.json (project selection, section allocation, narrative theme) Phase 2: Parallel Dispatch (per template×style combination, max 6 parallel) ↓ sections/<template>/<style>/<id>.md Phase 3: Assembly + Validation (per combination) ↓ ~/resumes/output/<target>/<template>/<style>.md|.html ``` ## Phase 1 - Planning Pass **Input:** state.json + template manifests + phase1-planning.md + target detail (if any) **Tasks:** - Determine which sections to include and their order (based on template + data availability) - Allocate page budget across sections (default 3 pages → per-section line targets) - Determine narrative theme from persona.tagline + persona.career_narrative + target keywords - Select top 6-7 projects via contribution-weighted scoring (see algorithm below) - Assign remaining projects to "其他项目" one-liner list - Select and assign value_points to sections (each value_point used exactly once) - Decide tech stack sorting strategy (with target → match score; without → frequency) - Note content structur