AGENT SKILL

profile-generate

Scaffold compact PROF-P entity profiles for unprofiled hub authors — ranks corpus REFs by citation in-degree, takes each top REF's primary author, skips institutional/group names and already-profiled people. Runs via `aiwg corpus profile-generate`.

License 许可
MIT
所属仓库
jmagly/aiwg
最近更新
2026-11-12
Stars
★ 229
Forks
36

01 前言 Front Matter

name
profile-generate
namespace
aiwg
platforms
[ "all" ]
commandHint
{ "model": "sonnet", "category": "research-profiles", "modelRole": "coding", "modelTier": "standard", "allowedTools": "Read, Bash, Write", "argumentHint": "[--limit N] [--scan N] [--write]" }
description
Scaffold compact PROF-P entity profiles for unprofiled hub authors — ranks corpus REFs by citation in-degree, takes each top REF's primary author, skips institutional/group names and already-profiled people. Runs via `aiwg corpus profile-generate`.

03 SKILL.md 正文

Profile Generate

Generate Tier-1 PROF-P entity profiles for the corpus's most-cited authors who don't have a profile yet — the bulk profile-induction pass.

How to run

aiwg corpus profile-generate                 # dry-run: list the PROF-P profiles it would create
aiwg corpus profile-generate --write         # write them (skips existing)
aiwg corpus profile-generate --limit 10 --write
aiwg corpus profile-generate --scan 100 --limit 40 --write
aiwg corpus profile-generate --fm --write    # FM-author PROF-P + group PROF-G from fm-config.yaml

--fm — foundation-model author/group profiles

--fm runs the FM pass (port of build_fm_profiles.py): it reads a corpus-local documentation/profiles/fm-config.yaml and scaffolds PROF-P profiles for the top-N authors of each listed FM paper plus PROF-G group profiles for team-authored releases. The FM-paper list and group specs are corpus data, not built-in:

# documentation/profiles/fm-config.yaml
fm-papers:
  REF-052: { model: "GPT-3", top-authors: 5 }
  REF-835: { model: "Llama 3", group: PROF-G-llama-team }
groups:
  PROF-G-llama-team:
    name: "Llama Team — AI @ Meta"
    parent-org: "Meta AI Research"
    parent-slug: PROF-O-meta-fair
    refs: [REF-835]

Absent file → no-op. Institutional/team author names are skipped from PROF-P.

  • Dry-run by default. --write creates files; existing profiles are skipped.
  • Ranks REFs by corpus in-degree (citations within the corpus, derived from the citation sidecars), takes each top REF's primary author, and skips institutional/group "authors" (team, labs, foundation, consortium, …) — those get group profiles, not PROF-P.
  • --scan N = how many top-in-degree REFs to consider (default 60); --limit N = max profiles to generate (default 25).
  • Generates a compact 5-section profile (summary, corpus presence, focus, co-author network, notes). affiliation is left blank for manual enrichment; corpus-refs is written as a list of REF-id strings.

Triggers

  • "generate profiles for top corpus authors"
  • "scaffold tier-1 hub-author profiles"
  • "profile the unprofiled hubs"
  • /profile-generate

References

  • @$AIWG_ROOT/src/artifacts/corpus-tools/profile-generate.ts — implementation
  • profile-status — entity-profile staleness; radar-init — the radar-sidecar analogue

04 同仓库更多 Skills