skills/llm-wiki/SKILL.md
This skill should be used when the user says "build a wiki", "maintain a wiki", "ingest docs into wiki", "query my wiki", "set up llm wiki", "wiki-based knowledge base", "stop re-reading docs every session", "persistent knowledge base", or "compress my docs into wiki pages".
npx skillsauth add kyuna0312/kyuna_token_saver LLM WikiInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Build and maintain a persistent wiki Claude can reference instead of re-ingesting raw documents each session. Derived from Karpathy's LLM Wiki pattern.
Why it works: Raw docs reloaded every session = wasted tokens. A wiki is a distilled, pre-summarized artifact — same knowledge, 10x lower token cost.
raw-sources/ (immutable originals)
└── docs, PDFs, changelogs, README files
wiki/ (LLM-maintained markdown pages)
├── index.md ← content catalog
├── log.md ← append-only ingest/query log
└── [topic].md ← one page per concept
CLAUDE.md (tells Claude: "reference wiki/ instead of raw-sources/")
Create wiki directory at project root or ~/.claude/wiki/ for global:
mkdir -p wiki
echo "# Wiki Index\n\n## Pages\n" > wiki/index.md
echo "# Ingest Log\n" > wiki/log.md
Add to CLAUDE.md:
## Knowledge Base
Wiki at `wiki/`. Reference wiki pages instead of raw source files.
Do not re-read raw docs unless wiki page is missing or stale.
When user provides a doc, URL, or file to ingest:
wiki/[topic].md per concept (one page per topic)wiki/index.md with new pageswiki/log.md: [DATE] INGEST: [source] → [pages updated]Per wiki page format:
# [Topic]
_Last updated: [DATE] | Source: [origin]_
## Summary
[2-3 sentence distillation]
## Key Points
- [point 1]
- [point 2]
## Related
- [[other-wiki-page]]
When user asks a question:
wiki/index.md for relevant pages(wiki/[page].md)wiki/log.md: [DATE] QUERY: [question] → [pages used]Periodically check wiki integrity:
Contradictions: pages with conflicting claims
Stale pages: last updated > 90 days AND source has changed
Orphaned pages: not referenced in index.md
Missing links: [[references]] with no matching file
Report format:
WIKI HEALTH
===========
Pages: [N]
Last ingest: [DATE]
Issues found:
- [issue 1]
- [issue 2]
Recommendation: [action]
| Approach | Tokens per session | |----------|--------------------| | Re-read raw 50-page doc | ~15,000 tokens | | Reference 3 wiki pages | ~600 tokens | | Savings | ~96% |
/estimate-tokens after wiki setup to measure baselinewiki/ reference to CLAUDE.md using /optimize-claudemdreferences/wiki-patterns.md — Page templates and multi-project wiki setupstools
This skill should be used when the user asks to "show recent changes", "show forge changelog", "show changelog for <project>", "what did I add this week", "what files did I touch today", "what packages keep recurring", "what drifts from the template", "review template suggestions", "show pending forge suggestions", "sync forge templates", "apply template suggestion", or "compute back-mapping suggestions". Provides read, drift-discovery, and template back-mapping workflows that wrap the `forge-db` MCP tools (`get_changelog`, `compute_suggestions`, `apply_suggestion`).
testing
Use this skill when the user says "/task_brain", "break down this task", "plan this", "decompose this problem", or when a task has multiple moving parts, unclear dependencies, or high ambiguity.
tools
This skill should be used when the user says "optimize settings.json", "tune settings for low tokens", "settings.json for performance", "disable auto memory", "configure for token savings", "minimal context window settings", or "settings taking too many tokens".
tools
This skill should be used when the user asks to "add token counter to status line", "show context usage in status bar", "visualize token usage", "set up token status line", "live token counter", "show context window percentage", or wants to see token burn in the terminal status bar.