技能创建器
skill-creator
创建、改进和评估自定义技能,把专业流程转化为可重复使用的 AI 能力。
person-profile-writing【数据与研究】Research, draft, or revise source-grounded biographies and person-centered speaker, founder, or expert profiles. Use when identity resolution, career chronology, roles, projects, public ideas, or attribution are central to the deliverable. Pair with research-grounded-writing for evidence and swyx-writing for prose.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Produce an accurate account of a person and why their work matters. Apply
research-grounded-writing to the evidence and swyx-writing to the prose.
This skill owns the identity and biography-specific decisions.
Build an evidence packet when important facts remain uncertain:
Use smart-entity-resolution only when the identity is genuinely ambiguous.
For a visual profile page, qualify a portrait alongside research and writing; do not leave image discovery as a model suggestion or defer it until publication. Reuse the destination's existing photo adapters. Prefer an identity-matched organizer, employer or curated portrait, then a verified person-owned website or independently verified public account. Name similarity, a search thumbnail or face similarity alone does not establish identity. Do not generate a likeness to fill a gap.
Retain the exact image URL, identity source, image hash and qualification outcome. Check acquired image bytes and dimensions, and inspect the selected image. If no image qualifies within scope and budget, record the concrete unavailable reason; a missing portrait does not justify inventing identity or discarding useful prose.
Apply qualified portraits to fresh and cached accepted biographies as metadata updates without replaying approved writing. Preserve existing approved images and carry the image and provenance through snapshot assembly, compact indexes and public loaders. For publication, verify the selected image actually renders and loads on the affected desktop/mobile page; a biography checksum or initials placeholder is not portrait acceptance. Pure text-only writing needs no image work.
Lead with the strongest supported identity and contribution. The opening should explain who this particular person is and why their work matters. A conference, profile page, article, or citation should not become the main character.
Weave career transitions, projects, technical ideas, public writing, talks, and distinctive posts into a coherent account. Use transcripts for concrete insight instead of producing a sequence of presentation summaries. Preserve the distinctions between participation, authorship, ownership, employment, and claims about third-party products.
Choose chronology or another structure according to what best explains the person. Source-poor profiles should remain brief. Extensive evidence justifies more development only when it adds understanding.
For a generation pipeline, calibrate on a sample spanning evidence-rich and evidence-poor subjects, disputed identities, varied careers, different source types, and formatting edge cases.
Evaluate usefulness without requiring a site-specific reference: the opening identifies this particular person, supported contributions are explained concretely, and the account develops coherently without repetition or generic career padding. Meaningful links and navigation help readers investigate the work. Examples from other corpora illustrate those qualities, not a required style, length or outline.
When approved examples are supplied, compare complete profiles rather than their headings or length. Assess factual support and editorial usefulness separately: does the draft explain this person’s distinctive contributions and development, or merely recount the latest talk? Select useful mechanisms and examples without turning refinement into a transcript walkthrough. Source-poor brevity is valid; missing optional career details are not a reason to invent or demand a dossier. Repair substantive repetition, confusing selection or lost contributions locally; preferences about length or outline alone are not rejection grounds.
Check identity fidelity, supported chronology, ownership and attribution, meaningful links, private-source exclusion, and destination rendering. Persist accepted evidence and revisions so retries do not repeat successful paid work. Keep research, synthesis, editorial development, validation, and publication separate when those boundaries improve the pipeline.