More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    token diet

    1

    Reduce AI agent context bloat by pruning redundant instructions and sharpening vague rules without losing behavior.

    Free

    1 useSecurity scanned
    token diet

    token diet

    Example session with this skill installed

    My .cursorrules file is over 4,000 tokens and the agent is starting to ignore the styling rules. Can you slim it down and make sure it keeps the tailwind and typescript constraints?

    • Read your context and instructions
    • Compiled the token diet

    I have audited your .cursorrules file. I found significant redundancy in the React component section and several generic "best practice" filler lines that the model already follows by default.

    TOKEN DIET REPORT

    Before: ~4,250 tokens -> After: ~2,880 tokens (-32%)

    What changed

    • Merged 4 duplicate rules regarding functional components.
    • Moved the 15-line "Git Commit Message" guide to a separate reference/git.md.
    • Cut 8 generic lines (e.g., "be polite", "write clean code").
    • Sharpened 3 vague rules about error handling into concrete try/catch requirements.
    Protected RuleStatusNew Location
    Tailwind-first stylingPreservedStyling Section, Line 4
    Strict TypeScript typesPreservedCore Logic, Line 12
    No external UI libsPreservedConstraints, Line 8

    Conflicts found: None.
    Risk: Low. I refactored the component boilerplate examples to be more concise; verify the syntax matches your preferred snippets.

    Optimized .cursorrules

    
    # Core Constraints
    - Use TypeScript for all files; no `any` allowed.
    - Use Tailwind CSS for all styling. Do not install or use external UI libraries like Shadcn or MUI.
    - Prefer functional components and hooks over class components.
    
    # React Patterns
    - One component per file.
    - Export as default.
    - Use arrow functions for component definitions.
    
    # Styling Rules
    - Use utility classes only.
    - Group Tailwind classes by category (layout, spacing, color).
    - Mobile-first approach for all responsive design.
    
    # Error Handling
    - Wrap all async data fetches in try/catch blocks.
    - Log errors to the internal monitoring service using `logError()`.
    

    Next steps

    1. Review the sharpened styling rules to ensure the specific Tailwind grouping meets your needs.
    2. Replace your current .cursorrules with the optimized version above.
    3. Move the Git guide to a reference file if you still need the agent to see it occasionally.

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

    About this skill

    Stop paying for context bloat and sluggish agent performance. Large instruction files like .cursorrules, CLAUDE.md, and system prompts consume expensive tokens and cause models to lose focus. This skill audits your agent's configuration files to prune generic filler, merge duplicates, and sharpen vague rules while strictly preserving your core logic.

    What it does

    • Instruction classification categorizes every rule into keep, merge, move, cut, or sharpen buckets.
    • Deduplication finds redundant imperatives spread across multiple files or sections.
    • Context compression rewrites verbose sentences into concise, imperative triggers without losing meaning.
    • Conflict detection identifies contradictory instructions that confuse the agent.
    • Token estimation provides before and after approximations to quantify context savings.

    How it works

    1. Initial Audit scans the target files to measure character counts and estimate current token usage.
    2. Rule Mapping identifies every unique instruction and labels "protected" rules that must remain untouched.
    3. Refactoring merges duplicates, cuts fluff like "be helpful," and sharpens vague advice into concrete triggers.
    4. Validation generates a behavior-preservation report to ensure all protected logic is present in the new version.

    Frameworks & tools

    Works with any text-based instruction format including .cursorrules, CLAUDE.md, AGENTS.md, SKILL.md, MCP tool descriptions, and custom system prompts.

    Why this beats prompting it yourself

    Generic "make this shorter" prompts often hallucinate new rules or accidentally delete critical safety constraints. This skill uses a systematic classification workflow and a mandatory preservation check to ensure your agent's behavior remains identical while its footprint shrinks.

    Use cases

    • Fixing "Agent Amnesia" when long prompts cause the model to forget middle-of-file instructions.
    • Reducing API costs by trimming unnecessary filler from every message sent to the LLM.
    • Cleaning up tech debt in older instruction files that have grown messy over multiple iterations.
    • Standardizing instructions across a team of different agents for better consistency.

    Known limitations

    Token counts are estimates based on character ratios. It will flag but not automatically remove sensitive data like API keys. Files under 1,500 tokens may not see significant benefits.

    Token counts are approximations. Does not automatically fix secrets or API keys. Files already under 1,500 tokens are usually skipped to avoid over-optimization.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Permissions

    No special permissions declared or detected

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Token Diet specializes in structured instruction files like .cursorrules, CLAUDE.md, and custom system prompts, but it can optimize any text-based configuration or Markdown-heavy agent guide.

    The skill uses a strict 'Rule Mapping' process that identifies protected logic before optimization, ensuring that while the phrasing becomes more concise, the actual constraints and behaviors remain unchanged.

    While results vary based on the original file's verbosity, most users see a 20-40% reduction in token usage, which helps prevent 'middle-of-the-prompt' amnesia and lowers per-request API costs.

    Yes, you will receive a validation report that highlights merged rules, deleted filler, and a side-by-side token estimation to verify the optimization before you deploy the new instructions.

    Optimization is most effective for complex instruction sets; files under 1,500 tokens generally lack the redundancy required for the skill to provide a meaningful performance boost.

    New in Prompt & Skill Engineering