Public handling: selected full-text prompt with secret filtering.
Prompt Text
PLEASE IMPLEMENT THIS PLAN:# 立党老师 Public Ideas Corpus Ingest + Auto-Update## Summary- Build a `vipin wiki` corpus for 立党老师 / `lidangzzz`, optimized for high-frequency, low-to-medium-density public ideas rather than project-heavy sources.- Verified starting points: - YouTube RSS works for channel `UC4gzU_8MxRDiQrSFOiT79tQ`, title `立党 lidang`, currently exposes 15 recent videos. - X handle is treated as `@lidangzzz`; official X is canonical, third-party mirrors are auxiliary and clearly marked non-authoritative. - Existing wiki has no maintained 立党 corpus yet; only a separate local `lidang-perspective` skill exists.- Use batch aggregation by week/month plus high-signal single-item pages, not one wiki page per short post.## Key Changes- Add `scripts/ingest-lidang-public.ps1` with modes: - `-Mode Light`: daily capture of new metadata, links, hashes, mirrors, video RSS items, and candidate high-signal posts. - `-Mode Digest`: weekly clustering, de-duplication, taxonomy updates, wiki page generation, catalog/lint validation. - `-DryRun` and `-SkipValidation` for safe checks.- Store machine-readable state under `raw/lidang-public/manifest.json` and daily capture batches under `raw/lidang-public/inbox/`.- Create durable wiki surfaces: - `wiki/entities/lidang.md` - `wiki/topics/lidang-public-ideas.md` - `wiki/sources/2026-05-16-lidang-public-corpus.md` - `wiki/analyses/lidang-idea-taxonomy.md` - `wiki/analyses/lidang-weekly-digests.md` - `wiki/sources/lidang-public/` for weekly/monthly batches and high-signal single items only.- Public handling: - Official URLs, timestamps, metadata, short summaries, hashes, and source provenance are public wiki-safe. - Full X text from mirrors is local/cache-only unless redistribution is clearly allowed. - YouTube transcripts are fetched only when legally/technically available; otherwise store title, URL, metadata, and summary from accessible public text.## Classification And Data Model- Primary categories: - AI coding / agent workflows - 转码 / 职业路径 - 移民 / 身份 / 制度入口 - 创业 / Delaware / 公司工具 - 开源工程 / 工具链 - 教育 / 学科选择 / CS learning - 经济 / 房地产 / 社会观察 - 法律规则 / 治理 / rule-as-script - 日常启发式 / 判断框架 - 争议性观点 / 需复核观点- Manifest item fields: - `id`, `source_kind`, `canonical_url`, `mirror_urls`, `published_at`, `captured_at`, `content_hash`, `source_confidence`, `category`, `tags`, `engagement`, `importance_score`, `dedupe_key`, `wiki_page`, `public_handling`.- Wiki generation rule: - Daily short posts stay in weekly batch pages unless importance score crosses a threshold. - Single-item pages are reserved for long posts, high-engagement posts, videos, repeated themes, or ideas that materially update the taxonomy.## Automation- Create two Codex automations: - Daily light crawl: run `scripts/ingest-lidang-public.ps1 -Mode Light`, update raw candidate queue and manifest, avoid wiki churn, commit only scoped raw/manifest changes when new public metadata appears. - Weekly digest: run `scripts/ingest-lidang-public.ps1 -Mode Digest`, update wiki pages, taxonomy, index, log, catalog, validate, commit with `Update Lidang public ideas corpus`, and push `origin/main`.- Ignore unrelated local changes, especially existing untracked `GetPdf.pdf`.- If mirrors fail or X blocks access, record crawl errors in the manifest and keep the automation non-destructive.## Test Plan- Dry-run verifies: - YouTube RSS returns nonzero entries and channel title `立党 lidang`. - X canonical account URL and at least one auxiliary discovery source are reachable or errors are recorded. - manifest item IDs and hashes are stable across repeated runs.- Validation verifies: - generated pages have frontmatter, source URLs, category, summary, confidence, and public handling policy. - weekly batch pages link back to the topic hub, taxonomy, and source note. - high-frequency items do not flood `wiki/index.md`; index links only hubs, analyses, batch pages, and high-signal single pages. - `scripts/wiki-catalog.ps1`, `scripts/wiki-lint.ps1`, and `git diff --check` pass.- Acceptance: - first digest creates the corpus. - daily light runs capture new candidates without noisy wiki churn. - weekly digest clusters new public content into readable summaries and pushes scoped changes.
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.