claude-skills-tokens
CLAUDE.md1 starsChanged 3 months agoApache-2.0View on GitHub
Skill marketplace for Claude Code token optimization. Installable skills backed by research.
In this repository2 other files it gives its agents.
- token-hygieneSkill
- token-optimizerSkill
The file
We split this CLAUDE.md into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 46
Use subagents to read reports in parallel — don't load all 11 into one context
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Claude Code Token Skills
Skill marketplace for Claude Code token optimization. Installable skills backed by research.
Directory Layout
| Directory | Contents | Public |
skills/ | Installable skills (agentskills.io format) | Yes |
reports/ | Polished research guides | Yes |
research/ | Personal audit notes and handoff prompts | No (.gitignored) |
docs/ | Design specs and implementation plans | Yes |
Reports
| File | Topic |
reports/01-token-control-features.md | All token control levers (model, effort, caching, compact, tools) |
reports/02-pricing-and-model-comparison.md | Full pricing table, cache multipliers, subscription vs API |
reports/03-dynamic-model-switching.md | /model, OpusPlan, subagent models, cache invalidation |
reports/04-comprehensive-token-optimization.md | Combined checklist, anti-patterns, savings summary |
reports/05-mcp-tool-routing-and-tokens.md | Tool Search (85-95% savings), defer_loading, CLI vs MCP |
reports/06-cost-tracking-and-monitoring.md | /cost, /usage, budget flags, community tools |
reports/07-skill-loading-and-claudemd.md | Progressive disclosure (98% savings), path-scoped rules |
reports/08-context-management-deep-dive.md | /compact, auto-compaction, cache TTL, Edit vs Write |
reports/09-github-ecosystem-and-community.md | GitHub repos, tools, community findings |
reports/token-optimization-best-practices.md | Synthesized 10-section actionable guide |
reports/claude-code-token-optimization-guide.md | Full 31K end-to-end workflow patterns guide |
Key Findings
- CLAUDE.md bloat is the #1 controllable cost — 91.9% reduction possible with no quality loss
- Skills progressive disclosure saves up to 98% vs putting everything in CLAUDE.md
- Tool Search reduces MCP overhead 85-95% (enabled by default since v2.1.x)
- Prompt cache invalidation from model/effort switching is a silent cost killer
- Edit tool is 95% cheaper than Write for file modifications
- Subagents with Haiku are 10x cheaper for research tasks
- The 270s cache boundary matters — never sleep exactly 300s
- Skill invocations themselves can consume massive context (one skill consumed 31% of a 1M context window)
- VS Code autosave (
claudeCode.autosave: true) is a silent cache killer — causes full cache invalidation via LSP diagnostic mutation - Subagent caching is hardcoded off (
enablePromptCaching: false) — 1h cache TTL irrelevant for multi-agent workflows - Community tools (ecotokens, token-optimizer-mcp, foveance) offer 75-95% savings via hook-based filtering and caching
Working With These Reports
- Use subagents to read reports in parallel — don't load all 11 into one context
- The reports overlap intentionally (different agents researched from different angles)
- Cross-reference findings before acting — some community-sourced numbers need verification
- Pricing data is as of July 2026; Sonnet 5 intro pricing expires August 31, 2026
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.