菜单
Fetch and download RAW sources (statements, invoices, documents) from a single Google Drive folder into the local raw cache, without versioning PII in git. Centralizes raw in Drive and keeps the repo lightweight.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Convert a new source into deterministic manifest, extracted text/chunks, normalized event, contextual LLM plan, and PR-ready ingestion proposal without writing canonical memory directly.
Single entry skill to set up AND operate a Markdown/Git living operational wiki (the "wiki viva kit") — configure wiki.config.yaml, scaffold or adopt the tree, ingest sources through the deterministic pipeline, run the delegated LLM deep read, keep the honesty gates green, compile the daily cockpit, and ship changes through the PR gate. Install this one skill to adopt and run the whole system; it points to the deeper per-step playbooks when you need them.
Generic router for Markdown/Git living-wiki repos that use memories/, docs/references/, wiki.config.yaml, wiki_core, PR gates, source manifests, normalized ingestion events, and operation cockpit pages.
Run the deep contextual reading (LLM pass) over the chunks selected by wiki_core, write the result to the cache, and then consolidate and INTEGRATE what was read into the target wiki pages (ingesting = integrating). The intelligence lives in the agent that runs the repo (Claude/Codex/Gemini), not in a Python LLM client.
Compile or review the living-wiki operation cockpit with current Git state, pending decisions, agent actions, source queues, stale contexts and validation links.
Audit source traceability, manifests, extracted text, chunks, indexes, LLM cache metadata, Markdown links and access-secret exclusions in a portable wiki repo.
| name | wiki-raw-drive |
| description | Fetch and download RAW sources (statements, invoices, documents) from a single Google Drive folder into the local raw cache, without versioning PII in git. Centralizes raw in Drive and keeps the repo lightweight. |
Raw sources (bank statements, invoices, PDFs with PII) must NOT be versioned in git: they leak sensitive data into history and bloat the repo. They live in a single Drive folder and are downloaded on demand into the local raw cache (data/raw/, which is gitignored). This separation follows the kit's architecture: the deterministic code processes; the agent that runs the repo uses the Drive connector to gather the sources.
The filename -> drive_file_id map lives in
data/raw/_drive_manifest.json (this JSON is
versioned; it contains no PII, only identifiers). The canonical Drive folder is
configured by the repo owner and referenced by folder id in the manifest.
drive_file_id in the manifest (or
search the folder via the Drive connector: search_files with parentId = '<folder_id>').download_file_content /
read_file_content) into the raw cache (data/raw/,
gitignored).wiki-llm-context-agent).filename -> drive_file_id in the manifest.