boot-tax-monitor
Skill0 starsChanged 7 months agoMITView on GitHub
Monitor and alert when Claude Code session startup overhead (plugins, skills, MCP servers, CLAUDE.md chain) exceeds a configurable token threshold. Prevents the silent context bloat that eats your working memory before you type a single prompt.
In this repository24 other files it gives its agents.
- aarSkill
- agent-cost-modelSkill
- banana-makerSkill
- bitter-lesson-scorecardSkill
- compensating-complexity-auditorSkill
- context-fork-guideSkill
- context-hygieneSkill
- counterargument-stress-testSkill
- dark-code-auditSkill
- executive-briefingSkill
- failure-asymmetrySkill
- failure-postmortemSkill
- file-intelSkill
- geopolitical-signal-enricherSkill
- get-api-docsSkill
- gh-reviewSkill
- model-routerSkill
- optimize-descriptionSkill
- prompt-rewriterSkill
- skill-auditSkill
- skill-maintenanceSkill
- spec-gap-detectorSkill
- token-burn-auditorSkill
- what-am-i-forgettingSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Use when the user says "boot tax", "startup overhead", "check my boot cost", "how heavy is my setup", "session startup cost", or wants to …
- Very specific: line 22
All skill description fields from SKILL.md frontmatter load into the system prompt.
- Distinctive: line 67
If HEAVY or CRITICAL, append the top 3 reduction opportunities with estimated savings.
- Distinctive: line 71
If the user runs this regularly, store results to show trends:
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill 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
Boot Tax Monitor
Checks your Claude Code environment's startup token cost and warns when it exceeds a safe threshold. Designed to run periodically or on-demand to catch creeping bloat.
Trigger
Use when the user says "boot tax", "startup overhead", "check my boot cost", "how heavy is my setup", "session startup cost", or wants to know how much context is consumed before they even start working.
Phase 1: Measure Boot Tax
Calculate the total token cost of everything that loads when a new Claude Code session starts:
1. CLAUDE.md Chain
Measure global CLAUDE.md, project-level CLAUDE.md, project .claude/CLAUDE.md, and all MEMORY.md files.
2. Skill Descriptions
All skill description fields from SKILL.md frontmatter load into the system prompt. Count total skills and estimate combined description token weight.
3. MCP Server Tool Schemas
Each MCP server registers tool schemas that consume tokens. Count servers and estimate ~500 tokens per tool per server (conservative).
4. Hooks Configuration
Count total hooks across all event types.
Phase 2: Score Against Threshold
Default thresholds (adjustable):
| Level | Token Range | Status |
| Lean | <30K | All good -- plenty of working memory |
| Normal | 30K-60K | Acceptable for power users |
| Heavy | 60K-90K | Warning -- consider trimming |
| Critical | >90K | Alert -- significant context waste |
Calculate: available_working_memory = context_limit - boot_tax
For a 200K context window:
- Lean setup: ~170K working memory
- Critical setup: ~110K or less working memory
Phase 3: Report
Output a single compact report:
Boot Tax Report
===============
CLAUDE.md chain: XXK tokens
Skill descriptions: XXK tokens (N skills)
MCP tool schemas: ~XXK tokens (N servers)
Hooks: XXK tokens (N hooks)
Memory files: XXK tokens
-------------------------------
Total boot tax: XXK tokens
Context limit: 200K tokens
Working memory: XXK tokens (XX%)
Status: [LEAN / NORMAL / HEAVY / CRITICAL]If HEAVY or CRITICAL, append the top 3 reduction opportunities with estimated savings.
Phase 4: Trend Tracking (Optional)
If the user runs this regularly, store results to show trends:
echo "$(date -I),BOOT_TAX_TOKENS" >> ~/.claude/boot-tax-log.csvOn subsequent runs, show: Last check: YYYY-MM-DD (XXK tokens) -> Today: XXK tokens (delta)
Integration Notes
- Pairs well with
token-burn-auditorfor deep-dive analysis - Can be automated via
/loopor/schedulefor periodic checks - Results can inform
compensating-complexity-auditorwhen evaluating whether skill/hook overhead is justified
Source Attribution
Technique derived from Nate's Newsletter (2026-04-02): "You're Loading 66,000 Tokens of Plugins Before You Even Type" -- the "66K token" problem of invisible startup overhead.
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.

