LLM-maintained knowledge base for ps1raf
🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> |
||
|---|---|---|
| conventions | ||
| corrections | ||
| external | ||
| infrastructure | ||
| lessons | ||
| services | ||
| tools | ||
| CHANGELOG.md | ||
| index.md | ||
| LLMWIKI.md | ||
| README.md | ||
llmwiki — ps1raf
An LLM-maintained markdown knowledge base for the ps1raf homelab server. Agents (primarily Claude Code) read and extend this wiki as they work; humans can read it too.
- For agents: see LLMWIKI.md for the agent contract (how to read, when to extend, correction flow).
- For humans: start at index.md to browse.
- Recent activity: see CHANGELOG.md.
Design
Based on Karpathy's LLM Wiki pattern: a structured markdown directory that an agent builds and maintains as it works, so facts compound across sessions instead of decaying inside conversation context. Adapted for sysadmin/ops context rather than research.
Scope
What IS in scope:
- Services deployed on ps1raf (Kanidm, Forgejo, Zabbix, …)
- Host-level conventions (git flow, container style, secret storage, notifications)
- Tool knowledge (step-ca CLI, myntfy, podman quirks)
- External references (Forgejo URL, Kanidm URL, registered domains)
- Corrections & lessons (what went wrong, how to avoid repeating)
What is NOT in scope:
- User preferences (→
~/.claude/projects/-home-raf/memory/) - Task state (→
loop.md,loop.log) - Harness config (→
CLAUDE.md,~/.claude/settings.json) - Secrets (→
~/.env)