Public handling: selected full-text prompt with secret filtering.
Prompt Text
PLEASE IMPLEMENT THIS PLAN:# 姚顺宇 / 姚顺雨 Public Corpus Ingest + Auto-Update## Summary- Build two separate public corpora in `vipin wiki`: - **姚顺雨 / Shunyu Yao / `ysymyth`**: OpenAI researcher focused on language agents, ReAct, ToT, SWE-agent, SWE-bench, tau-bench, CUA, Deep Research. - **姚顺宇 / Dr. Shunyu Yao / `alfredyao.github.io`**: physics-to-AI researcher, Google DeepMind / Anthropic, quantum physics, RL, agentic coding, non-Hermitian skin effect, scramblon/quantum gravity work.- Do not mix the two people. Use distinct entity IDs, source directories, manifests, and automation prompts.- Use **safe public indexing**: commit metadata, summaries, links, hashes, categories, and license notes; avoid committing unlicensed full PDFs/code/text unless clearly redistributable.- Add a durable “public corpus ingest workflow” so future agents recognize this repeated prompt pattern and know how to implement it.## Key Changes- Add one script: `scripts/ingest-shunyu-yao-public.ps1` with `-Person ysymyth|alfredyao|All`, `-DryRun`, and `-SkipValidation`.- Store machine-readable manifests: - `raw/yao-shunyu-ysymyth/manifest.json` - `raw/yao-shunyu-alfred/manifest.json`- Create durable wiki pages: - `wiki/entities/yao-shunyu-ysymyth.md` for 姚顺雨. - `wiki/entities/yao-shunyu-alfred.md` for 姚顺宇. - `wiki/topics/shunyu-yao-public-corpora.md` as the hub that explicitly separates the two. - `wiki/sources/2026-05-16-yao-shunyu-public-corpora.md` as the batch provenance note. - `wiki/analyses/shunyu-yao-project-taxonomy.md` for projects/repos/products. - `wiki/analyses/shunyu-yao-paper-map.md` for papers, theses, benchmarks, and research themes. - `wiki/analyses/public-corpus-ingest-workflow.md` documenting the reusable workflow/prompt pattern.- Update `AGENTS.md` with a short pointer: when a user asks for “完整 GitHub / 所有项目 / 公开发布信息 / 所有论文 / 定时更新”, use the public-corpus ingest workflow and first disambiguate same-name people.- Update `wiki/index.md`, `wiki/log.md`, and `wiki/catalog.json`.## Source Coverage- For **姚顺雨 / `ysymyth`**, crawl: - Homepage/blog: `https://ysymyth.github.io/` and `https://ysymyth.github.io/blog/` - GitHub API: `https://api.github.com/users/ysymyth`, repos, gists, releases, README/license metadata. - Linked project repos from homepage: ReAct, Tree of Thoughts, SWE-agent, SWE-bench, WebShop, tau-bench, InterCode, Reflexion, Cognitive Architectures, and future linked repos. - Papers from homepage links, arXiv IDs, thesis PDF metadata, project pages, talks/YouTube links.- For **姚顺宇 / `alfredyao`**, crawl: - Homepage: `https://alfredyao.github.io/` - CV PDF metadata: `https://alfredyao.github.io/Shunyu_Yao_CV.pdf` - Google Scholar link from CV as a pointer only; use arXiv/Crossref/Semantic Scholar/OpenAlex where available for machine-readable paper metadata. - Linked writing pages, LinkedIn pointer metadata, and any GitHub links found from official pages or verified paper/project pages. - Physics/AI paper metadata around non-Hermitian skin effect, quantum chaos/gravity, scramblons, RL numerics, and agentic coding contributions.## Manifest And Page Rules- Stable manifest IDs: - `person:ysymyth`, `person:alfredyao` - `github:<owner>/<repo>` - `paper:arxiv:<id>` or `paper:doi:<doi>` or `paper:title-hash:<hash>` - `post:<canonical-url-hash>` - `talk:<platform>:<id>` - `project:<canonical-url-hash>`- Each item records: `person_key`, canonical URL, source kind, title, authors/owners, venue/date, repo/path, license, content hash, semantic hash, first seen, last seen, last changed, wiki page, category, tags, public handling, crawl errors.- Per-item wiki pages are created for repos, major projects, papers, theses, talks, and high-signal posts. Short/low-signal updates stay in batch digest pages.- Public pages summarize and classify; they do not mirror unlicensed PDFs, full webpages, or source code.## Classification- Shared categories: - Language agents / agent architectures - Tool use / computer use / digital automation - Software engineering agents and benchmarks - Evaluation / benchmark design - Reinforcement learning and reasoning - Human-agent interaction / user simulation - Quantum chaos / quantum gravity - Non-Hermitian topology / condensed matter - Physics-to-AI research transition - Public essays / research philosophy- Every page includes a “do not confuse with” note in metadata or source note, but not over-emphasized in prose.## Automation- Create weekly Codex automation: `Update Shunyu Yao Public Corpora`.- Prompt: run `scripts/ingest-shunyu-yao-public.ps1 -Person All`, summarize additions/changes/removals/errors, run `scripts/wiki-catalog.ps1`, `scripts/wiki-lint.ps1`, and `git diff --check`, commit scoped changes with `Update Shunyu Yao public corpora`, and push `origin/main`.- No noisy commits: if stable IDs and semantic hashes are unchanged, report no changes and do not commit.- Ignore unrelated local changes, especially existing untracked `GetPdf.pdf`.## Test Plan- Dry-run verifies both people resolve to distinct canonical homes and manifests.- Validation checks: - no duplicate stable IDs; - every item has `person_key`, `source_kind`, canonical URL, category, semantic hash, first seen, and public handling; - all public pages have frontmatter, source URLs, summary, confidence labels, and no private path leaks; - `wiki-catalog.ps1`, `wiki-lint.ps1`, and `git diff --check` pass.- Repeat-run acceptance: - first run creates the corpora; - second run with no upstream changes is no-op; - a new paper/repo/post creates exactly one manifest entry, wiki update, log entry, commit, and push.## Assumptions- “姚顺宇” means the DeepMind/Anthropic physics-to-AI researcher at `alfredyao.github.io`, not the AIGC/product person at `shunyuyao.github.io`.- “姚顺雨” means `ysymyth.github.io` / GitHub `ysymyth`.- Public GitHub redistribution stays license-safe: summaries, metadata, hashes, and links are public; unclear-license full text/code/PDFs are not committed wholesale.
Reuse Notes
EXTRACTED: This is a selected Codex prompt or automation prompt from the local Codex corpus.
INFERRED: Future agents can reuse its structure, constraints, and acceptance criteria when creating similar Codex workflows.