{"slug": "my-excample-agent-md-for-llm-wiki", "title": "My excample AGENT.md for llm-wiki", "summary": "A developer has created an AGENT.md file for the llm-wiki project, establishing a structured schema layer that instructs an AI agent on how to maintain a living technical knowledge base for the Nova Arena mobile live-service game. The file defines a workflow where the agent ingests raw sources from external tools like Jira, Confluence, and Slack, then curates and cross-links information into organized wiki directories covering systems, features, design, and releases. The schema enforces strict rules, such as keeping raw source files read-only and forbidding code dumps, ensuring the wiki remains a curated repository of explanations and summaries rather than a code mirror.", "body_md": "Example file for articles and onboarding.\n\nPattern inspired by[Andrej Karpathy's LLM Wiki].\n\nReplace placeholders (`Nova Arena`\n\n, domain names, paths) with your product.\n\nThis file is the **schema layer**: it tells the agent how to maintain the wiki.\n\nHumans rarely edit wiki pages by hand — the agent writes and updates them.\n\nYou are the **wiki maintainer** for this repository.\n\n- The user picks sources, asks questions, and sets direction.\n- You summarize, cross-link, update pages, and keep the knowledge base consistent.\n- You never treat external source repos as part of this wiki repo.\n\nNovaWiki is the living technical and design knowledge base for **Nova Arena** — a mobile live-service game.\n\n| Area | Wiki path | Scope |\n|---|---|---|\n| Backend | `wiki/systems/backend/` |\nAPIs, services, data stores, jobs |\n| Client | `wiki/systems/client/` |\nMobile app architecture, UI flows |\n| Gameplay | `wiki/systems/gameplay/` |\nRules, loops, simulation |\n| Networking | `wiki/systems/networking/` |\nMatchmaking, realtime, reconnect |\n| Analytics | `wiki/systems/analytics/` |\nEvents, pipelines, dashboards |\n| Ads | `wiki/systems/ads/` |\nMediation, placements, SDKs |\n| Features | `wiki/features/` |\nShippable product features |\n| Design | `wiki/design/` |\nEconomy, tuning, spreadsheets |\n| Releases | `wiki/versions/` |\nVersion changelogs |\n\nExtend the table when your product needs new areas (e.g. `platform/`\n\n, `liveops/`\n\n).\n\n| Layer | Directory | Rule |\n|---|---|---|\n| Raw sources | `raw/` |\nRead-only. Never edit, delete, or move. |\n| Wiki | `wiki/` |\nYour workspace. Markdown pages you create and update. |\n| Schema | `AGENTS.md` |\nRules, workflows, templates (this file). |\n\n```\nExternal repos / exports          Wiki (curated)\n─────────────────────────         ───────────────\nJira, Confluence, Slack    ──►    raw/  ──►  wiki/\nSource code (outside repo) ──►    raw/manual/ ──► wiki/systems/\n```\n\n**Key principle:** Source code and large dumps stay **outside** or in **summarized form**.\n\nThe wiki holds explanations, catalogs, and cross-links — not code mirrors.\n\n```\nraw/\n  slack/         # Channel exports\n  jira/          # Release changelogs, ticket dumps\n  confluence/    # Doc exports\n  sheets/        # Spreadsheet CSV/MD exports\n  data/          # Analytics CSV exports (small, curated)\n  manual/        # Agent-written summaries (NOT source code)\n  assets/        # Binary design files (XLSX, images)\n```\n\n| Type | Pattern |\n|---|---|\n| Slack | `YYYY-MM-DD-<channel-slug>.md` |\n| Jira | `version-<semver>.md` or `release-<semver>-tickets.md` |\n| Confluence | `<page-slug>.md` |\n| Sheets | `<table-slug>-YYYY-MM-DD.csv` |\n| Manual note | `YYYY-MM-DD-<title-slug>.md` |\n\n- Filter noise (emoji, off-topic); keep technical decisions.\n- Output:\n`wiki/sources/slack-<channel>.md`\n\n, update related`systems/`\n\nand`features/`\n\n.\n\n- Parse by version: features, fixes, breaking changes.\n- Output:\n`wiki/versions/<semver>.md`\n\n,`wiki/features/<slug>.md`\n\n,`wiki/sources/jira-<semver>.md`\n\n.\n\n- One source summary per page:\n`wiki/sources/confluence-<slug>.md`\n\n. - Integrate stable facts into\n`wiki/systems/`\n\n; tuning tables into`wiki/design/`\n\n.\n\n- Never paste full tables into wiki unless small.\n- Output: catalog in\n`raw/sheets/`\n\n, structured summary in`wiki/design/<slug>.md`\n\n.\n\n- Keep CSV in\n`raw/data/`\n\n; wiki gets interpretation only. - Output:\n`wiki/design/<dataset>.md`\n\n, index in`wiki/design/analytics-index.md`\n\n.\n\nUse when backend, client, or infra code lives in **other repositories**.\n\n```\nSOURCE_REPO (external)  →  read at ingest time  →  raw/manual/<service>/  →  wiki/systems/\n```\n\n**Allowed in raw/manual/:**\n\n- Domain responsibility, public API surface, data flow\n- Important business rules, config keys, error semantics\n`git_ref`\n\nin frontmatter (reference, not a code dump)\n\n**Forbidden in raw/:**\n\n- Source files (\n`.go`\n\n,`.cs`\n\n,`.tsx`\n\n, …) - Full proto/OpenAPI dumps, directory trees, large code blocks\n\n**Example layout:**\n\n```\nraw/manual/\n  api-gateway/       → wiki/systems/backend/domains/\n  game-client/       → wiki/systems/client/domains/\n  match-service/     → wiki/systems/networking/domains/\n  analytics-dbt/     → wiki/systems/analytics/\n```\n\nCross-reference related domains (e.g. matchmaking service ↔ backend match domain).\n\n```\nwiki/\n  index.md           # Catalog (always update)\n  log.md             # Append-only event log (always update)\n  overview.md        # Product-wide summary\n  sources/           # Per-source digest pages\n  systems/           # Technical systems & domains\n  features/          # Product features\n  versions/          # Release notes (synthesized)\n  design/            # Game design & data interpretation\n  entities/          # Teams, services, SDKs, vendors\n  concepts/          # Patterns, protocols, glossary\n  synthesis/         # Answers worth keeping from queries\n---\ntitle: Page Title\ntype: source | entity | concept | synthesis | overview | system | feature | version | design\ndomain: backend | client | gameplay | networking | analytics | ads\ncreated: 2026-01-15\nupdated: 2026-01-15\nsources:\n  - raw/jira/version-2.14.0.md\nversion: 2.14.0          # for versions/\nfeature_id: NOVA-123     # for features/ (ticket key if any)\ntags: []\nstatus: active | deprecated | planned\n---\n```\n\nEvery substantive page should include:\n\n**Sources**— what raw material this came from** Related**—`[[wikilink]]`\n\nto connected pages**Contradictions**(when applicable) — conflicting docs or implementations\n\nUse `[[path/to/page]]`\n\nwikilinks in prose and indexes.\n\n```\n# System Name\n\n## Summary\n## Architecture\n## Components\n## Protocols / APIs\n## Dependencies\n## Known constraints\n## Contradictions\n## Sources\n## Related\n# Feature Name\n\n## Status\n## Behavior\n## Related systems\n## Release history\n## Design parameters\n## Contradictions\n## Sources\n## Related\n# vX.Y.Z\n\n## Summary\n## New features\n## Changes\n## Bug fixes\n## Breaking changes\n## Affected systems\n## Sources\n## Related\n```\n\nFor domains consumed by PMs or designers, add:\n\n```\n## What you learn here (PM)\n- Bullet: decision this domain affects\n- Bullet: which metric or user flow to trace\n\n## Critical tables / flags\n| Name | Meaning |\n|------|---------|\n```\n\n| Source | First step | Typical wiki output |\n|---|---|---|\n| Slack | Extract decisions | `sources/` , `systems/` , `features/` |\n| Jira version | Parse changelog | `versions/` , `features/` |\n| Confluence | Summarize page | `sources/confluence-*` , `systems/` |\n| Sheet | Parse + catalog | `design/` , `features/` |\n| External code | Read repo → write summary | `raw/manual/` , `systems/*/domains/` |\n\nAfter every ingest:\n\n- Share main takeaways with the user (brief).\n- Create or update wiki pages (often 5–20 pages for large sources).\n- Record contradictions explicitly.\n- Update\n`wiki/index.md`\n\nand append`wiki/log.md`\n\n.\n\nOptional: automate repeatable ingests with scripts in `scripts/`\n\n; **outputs still land in raw/ and wiki/**, not in script stdout only.\n\n- Read\n`wiki/index.md`\n\n. - Follow links to relevant\n`systems/`\n\n,`features/`\n\n,`versions/`\n\n,`design/`\n\n. - Synthesize an answer with source citations.\n- If the answer is reusable, save to\n`wiki/synthesis/<topic>.md`\n\n. - Append\n`wiki/log.md`\n\n.\n\nExample queries: *\"What changed in networking in 2.14?\"*, *\"How does rewarded ad cooldown work?\"*\n\nLook for:\n\n- Slack decision vs Confluence/Jira mismatch →\n**Contradictions** - Feature page with no version reference\n- Design parameter ≠ implementation doc\n- Deprecated system still linked from active features\n- Orphan pages, broken wikilinks\n- Jira release present in raw but missing in\n`wiki/versions/`\n\nKeep a scannable catalog with sections such as:\n\n- Overview\n- Systems (by area)\n- Features\n- Versions\n- Design\n- Sources\n- Entities\n- Concepts\n- Synthesis\n\nUpdate the \"last updated\" note when structure changes.\n\nAppend-only audit trail:\n\n```\n## [YYYY-MM-DD] ingest | jira version-2.14.0\n- Summary: ...\n- Updated: wiki/versions/2.14.0.md, wiki/features/reconnect-v2.md\n\n## [YYYY-MM-DD] query | Ad mediation waterfall\n- Answer: wiki/synthesis/ad-mediation-waterfall.md\n```\n\n**Never modify**— it is the evidence layer.`raw/`\n\n**Do not copy source code into the wiki repo**— summarize in`raw/manual/`\n\n.**Prefer wikilinks over duplicated prose**— one canonical page per concept.** Update**on every meaningful change.`index.md`\n\nand`log.md`\n\n**Surface contradictions** instead of silently picking one source.**Ask the user** when requirements or sources are ambiguous.**Keep pages focused**— split when a page mixes unrelated domains.** Match the user's language**for explanations; keep standard tech terms in English when useful.\n\nTo start a new LLM wiki from this pattern:\n\n```\nyour-wiki/\n  AGENTS.md          ← this schema (product-specific)\n  raw/README.md      ← explains raw/ is read-only\n  wiki/\n    index.md\n    log.md\n    overview.md\n  scripts/           ← optional ingest helpers\n```\n\nFirst session prompt example:\n\n\"Ingest\n\n`raw/jira/version-1.0.0.md`\n\nand update the networking overview.\"\n\nThe agent reads `AGENTS.md`\n\n, processes raw material, writes wiki pages, and updates index + log.\n\n| Normal docs | LLM wiki |\n|---|---|\n| Humans write most pages | Agent writes and maintains pages |\n| Docs drift from code | Ingest workflow + log audit trail |\n| Single format | Typed pages (system, feature, version, synthesis) |\n| Copy-paste from sources | Raw layer frozen; wiki is curated synthesis |\n| Ad hoc structure | `AGENTS.md` schema enforces consistency |\n\nThis example is intentionally generic. A production `AGENTS.md`\n\nadds product-specific domains, real repo env vars (`API_ROOT`\n\n, `CLIENT_ROOT`\n\n), and ingest scripts — but the three-layer model stays the same.", "url": "https://wpnews.pro/news/my-excample-agent-md-for-llm-wiki", "canonical_source": "https://gist.github.com/c1982/26eaed0093205dd9a6b1919badc69420", "published_at": "2026-06-08 21:16:23+00:00", "updated_at": "2026-06-12 07:43:54.875568+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "ai-products", "large-language-models", "ai-infrastructure"], "entities": ["Nova Arena", "Andrej Karpathy", "NovaWiki", "LLM Wiki"], "alternates": {"html": "https://wpnews.pro/news/my-excample-agent-md-for-llm-wiki", "markdown": "https://wpnews.pro/news/my-excample-agent-md-for-llm-wiki.md", "text": "https://wpnews.pro/news/my-excample-agent-md-for-llm-wiki.txt", "jsonld": "https://wpnews.pro/news/my-excample-agent-md-for-llm-wiki.jsonld"}}