add: AGENTS.md schema, seed files, sample topic page

This commit is contained in:
truetype committed 2026-05-10 17:36:15 +02:00
1 parent 2326fc758f
commit 125f74c563
2 files changed
+139

No files matched your search

+104
View File
@@ -0,0 +1,104 @@
# LangCode Agent Wiki — Schema
## Purpose
Persistent knowledge base for the LangCode Agent project. Maintained entirely by an LLM agent running on Ollama (Gemma 2 9B). This wiki sits between raw source documents and your questions — knowledge is compiled once here, then kept current.
## Directory Structure
\`\`\`
llmwiki/
├── AGENTS.md ← this file (schema)
├── raw/ ← immutable source documents
│ ├── articles/ ← web articles, blog posts
│ ├── papers/ ← research papers, technical docs
│ ├── repos/ ← code repository notes
│ └── assets/ ← images, downloads, attachments
└── wiki/ ← LLM-generated, interlinked pages
├── index.md ← master catalog (updated every ingest)
├── log.md ← append-only changelog
├── overview.md ← wiki purpose & architecture
├── concepts/ ← one file per key concept
├── entities/ ← one file per person, org, tool
├── sources/ ← one file per source document
└── comparisons/ ← comparative analysis pages
\`\`\`
## Conventions
### Filenames
- lowercase, hyphens instead of spaces (e.g. \`self-attention.md\`)
- no special characters, no camelCase
- prefix with date for time-sensitive entries: \`2026-05-10-some-event.md\`
### Page Structure
Every wiki page must follow this template:
\`\`\`markdown
# Title
> One-sentence summary of this page.
## Tags
\`concept\` \`entity\` \`source-name\` \`date\`
## Body
[Content here — use ## and ### for sections]
## Related
- [[page-name]] — brief description of connection
- [[another-page]] — brief description
## Sources
- [Source Title](raw/sources/filename) — relevant passage or page range
\`\`\`
## Ingest Procedure
When a new source is added to \`raw/\`:
1. Read the source end-to-end
2. Identify key concepts, entities, claims, and data points
3. For each: create or update the relevant wiki page in \`wiki/\`
4. Add/update cross-references using \`[[wikilinks]]\`
5. Update \`wiki/index.md\` with new/modified pages
6. Append entry to \`wiki/log.md\` with format:
\`\`\`
## YYYY-MM-DD HH:MM — [source filename]
- Created: [new pages]
- Updated: [modified pages]
- Key takeaways: [bullet summary]
\`\`\`
7. Run the lint checklist (below)
## Query Procedure
When asked a question:
1. Read \`wiki/index.md\` first to identify relevant pages
2. Load only those pages (not the entire wiki at scale)
3. Synthesize answer from compiled wiki knowledge
4. Cite specific wiki pages: \`[[page-name]]\`
5. If the answer is valuable and reusable, file it as a new wiki page
## Lint Checklist
Run weekly or after every 5-10 ingests:
- [ ] Every page in \`wiki/\` appears in \`index.md\`
- [ ] Every \`[[wikilink]]\` resolves to an existing file
- [ ] No two pages cover the same concept (merge duplicates)
- [ ] Conflicting claims between pages are flagged, not silently overwritten
- [ ] No orphan pages (pages with zero inbound links)
- [ ] All pages follow the template structure above
- [ ] Tags are consistent across related pages
- [ ] Log entries are present and properly formatted
- [ ] Sources in \`raw/\` that have been updated are reflected in wiki
## Conflict Resolution
When sources disagree:
1. Note both claims explicitly on the relevant page
2. Cite both sources
3. Add a "Status" field: \`verified\` / \`disputed\` / \`unverified\`
4. Do NOT silently overwrite — preserve the history of claims
## Naming Conventions
- Page titles: Title Case (e.g. "Self-Attention Mechanism")
- Filenames: lowercase-hyphens (e.g. \`self-attention-mechanism.md\`)
- Tags: lowercase, hyphens (e.g. \`#attention-mechanism\`)
## Growth Targets
- After 10-20 sources: start querying the wiki for answers
- After 50+ sources: consider adding qmd for enhanced search
- After 100+ sources: review and potentially restructure wiki subdirectories
+35
View File
@@ -0,0 +1,35 @@
# LLM Wiki Pattern (Karpathy)
> A method for building persistent, structured knowledge bases using LLM agents — compiled once, kept current, queried efficiently.
## Tags
`concept` `methodology` `knowledge-management` `2026-05-10`
## Overview
Andrej Karpathy proposed the "LLM Wiki" pattern as an alternative to RAG (Retrieval-Augmented Generation). Instead of searching raw documents on every query, an LLM agent incrementally builds and maintains a wiki of interlinked markdown pages.
## Core Architecture
### Three Layers
1. **Raw Sources** (`raw/`) — Original documents, immutable. Never modified by the LLM.
2. **Wiki** (`wiki/`) — Structured, interlinked markdown pages compiled by the LLM.
3. **Schema** (`AGENTS.md`) — Rules, conventions, and workflows. The "operating system" of the wiki.
### Workflows
- **Ingest**: Drop source in `raw/` → LLM reads, extracts, creates/updates ~10-15 wiki pages per source
- **Query**: Ask question → LLM reads `index.md` → drills into relevant pages → synthesizes answer with citations
- **Lint**: Periodically audit wiki for contradictions, orphans, stale claims
## Key Principles
- One concept = one file
- Cross-reference everything with `[[wikilinks]]`
- The LLM owns the wiki directory; humans only edit `raw/` and the schema
- Good answers get filed back into the wiki as new pages (compounding knowledge)
- Works with any LLM that can read/write files (Claude Code, Codex, Cursor, local models via Ollama)
## Related
- [[AGENTS.md]] — Schema file for this wiki
- [[index]] — Master catalog
## Sources
- Karpathy's original idea file (GitHub Gist)