Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+22 more

    leakhound

    1

    Local token forensics for Claude Code to find waste, audit MCP servers, and optimize model routing.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    leakhound

    leakhound

    Example session with this skill installed

    I'm hitting my usage limits too fast today. Run leakhound waste to see what's going on.

    • Read your context and instructions
    • Compiled the leakhound

    I have analyzed your project transcripts to find token leaks. The newest session is still active and growing.

    output ~142,000 tokens, cache writes ~85,000, cache reads ~310,000
    🟡 context pressure: recent input averaging ~45,000 tokens — compaction likely soon; /clear between tasks helps.

    - cache-churn     ████████████████████  240,000
- giant-read      ███░░░░░░░░░░░░░░░░░   38,000
- verbose-output  █░░░░░░░░░░░░░░░░░░░   12,500

    cache-churn

    • now: /clear
    • say: "Summarize the last 3 turns then clear context."
    • adopt: Break large feature additions into smaller PRs.

    giant-read

    • now: add vendor/ to .claudeignore
    • say: "Stop reading the vendor directory."
    • adopt: Use specific file paths instead of folder-wide searches.

    Do these, in order

    1. cache-churn (240,000 tokens): /clear
    2. giant-read (38,000 tokens): add vendor/ to .claudeignore
    3. verbose-output (12,500 tokens): check logger settings

    Token discipline (generated by leakhound)

    Paste this into the project's CLAUDE.md to make the fixes permanent:

    
## Token discipline
- Avoid re-reading large dependency folders (vendor/, node_modules/).
- Clear session context after completing a discrete sub-task.
- Prefer targeted file reads over directory-wide grep for known symbols.

    Estimated recoverable tokens = 290,500 (estimate).

    Next steps

    • Copy the Token Discipline block into your CLAUDE.md.
    • Run /leakhound mcp-audit to see if unused servers are adding to the pressure.
    • Use /clear now to reset the current session's churn.
    Paste this into the project's CLAUDE.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Stop bleeding tokens and hitting Claude rate limits. Leakhound provides a suite of local forensic tools to identify precisely where your token budget is being wasted and how to reclaim it.

    What it does

    • Waste forensics identifies high-churn files, redundant reads, and verbose outputs that inflate costs.
    • MCP server auditing flags connected MCP servers that are consuming context but never being called.
    • Plugin analysis detects idle plugins and calculates their per-session token overhead.
    • Model routing recommends when to use Haiku vs Sonnet based on task complexity and your cost preference.
    • Context monitoring provides a real-time statusline meter to track context pressure and compaction risks.

    How it works

    1. Use /leakhound waste to get a breakdown of token burn in your current project with specific fixes.
    2. Run /leakhound mcp-audit or plugin-audit to identify and remove dead-weight integrations.
    3. Apply suggested fixes to your CLAUDE.md to enforce long-term token discipline.
    4. Execute /leakhound trend to verify if your optimizations are actually reducing token growth over time.

    Frameworks & tools

    Built for Claude Code. Requires Node.js on the system PATH. Runs entirely local against transcript logs and configuration files.

    Why this beats prompting it yourself

    Generic prompts cannot see your actual usage data or local transcripts. This skill runs specialized scripts against your real history to find specific leaks that manual observation misses.

    Use cases

    • Troubleshooting "rate limit reached" errors by finding and pruning context-heavy files.
    • Optimizing expensive sessions by delegating mechanical tasks to cheaper models.
    • Auditing a cluttered workspace to disable unused MCP servers and plugins.
    • Monitoring real-time context growth to avoid unexpected automatic compactions.

    Known limitations

    Does not work with claude.ai or Cowork. Requires a local Node.js installation. It provides recommendations only and will not modify your MCP or plugin settings automatically.

    How to use

    Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill passes our trust checks below.

    Security scanned

    Passed our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scanned

    Verified clean 1 day ago

    • Passed all security checks, Safe to install

    Requires

    macOS
    Node.js

    Listed1 day ago

    What's inside

    Frequently Asked Questions