Works with the AI Agents you already use
Prompt Token Cost Audit
Audit LLM prompt files and tool schemas for token waste and enforce monthly budgets.
$2
Prompt Token Cost Audit
Example session with this skill installed
Audit our CLAUDE.md and weather.tools.json files for GPT-4o. We run about 5,000 requests a day with 200 output tokens per call. Check for waste.
- Read your context and instructions
- Compiled the prompt token cost
I have audited the specified files using the latest pricing for gpt-4o. The audit identified a significant opportunity to reduce costs by minifying the tool schema and reordering the system instructions to support better caching.
Prompt Audit Results (gpt-4o)
| File | Current Tokens | Monthly Cost (Est.) | Savings Potential | Recommendation |
|---|---|---|---|---|
weather.tools.json | 850 | $750.00 | 45% | Minify schema and remove redundant descriptions. |
CLAUDE.md | 1,200 | $1,050.00 | 15% | Move volatile session data to the end for caching. |
| Total | 2,050 | $1,800.00 | $495.00 |
The weather.tools.json file contains verbose property descriptions that account for 40% of its token weight. Additionally, CLAUDE.md currently places the date and time at the top of the file, which breaks prompt caching on every request.
Next steps
- Run
npx savedyouatoken@latest weather.tools.json --fixto apply lossless schema minification. - Manually move the
{{current_time}}variable inCLAUDE.mdto the bottom of the instruction block. - Generate a baseline file to ensure these optimizations are maintained in future PRs.
Connects securely to your tools. The creator never sees your data.
About this skill
Large language model bills often spiral due to invisible token waste in system prompts, bloated tool schemas, and inefficient few-shot examples. This skill audits your project files to identify exactly where money is being spent and provides tools to reduce costs without sacrificing model performance.
What it does
- Token Discovery identifies agent instruction files, tool schemas, and prompt templates across your repository.
- Cost Attribution calculates monthly spend per file based on real-time model pricing and request volume.
- Lossless Optimization applies rewrites to prompts and minifies tool schemas to reduce input token overhead.
- Caching Strategy identifies blocks that should be moved below static content to improve cache hit rates.
- Budget Enforcement generates CI/CD policies to block pull requests that exceed defined token limits or monthly budgets.
How it works
- Discover prompt-bearing files including
AGENTS.md,CLAUDE.md, and*.tools.jsonusing the integrated CLI. - Measure the baseline by selecting a specific model ID and defining request volume and average output tokens.
- Audit each file to rank opportunities, distinguishing between prompt overhead and response costs.
- Fix files via lossless rewrites and structural changes to ensure meaning is preserved while reducing weight.
- Protect against regression by creating baselines and policies that can be checked in automated workflows.
Frameworks & tools
This skill utilizes npx savedyouatoken for real-time model pricing and token counting. It supports instruction files for Claude, Gemini, and general LLM agents, as well as JSON tool schemas and MCP definitions.
Why this beats prompting it yourself
Generic LLMs cannot accurately count their own tokens or know current market pricing for specific provider models. This tool uses a live CLI to provide precise financial impact reports and structural optimizations that static prompts miss.
Use cases
- Reducing the overhead of large tool schemas in MCP-based agents.
- Trimming system instructions that are sent on every API request.
- Setting up a CI gate to prevent prompt bloat during team development.
- Justifying model switches by comparing projected costs across different providers.
Known limitations
Does not automatically pick up .cursor/rules or SKILL.md files unless explicitly passed. It cannot audit dynamic prompts built in code at runtime without the @savedyouatoken/sdk.
Requires manual identification of .cursor/rules and SKILL.md files. Runtime prompts built in application code require the SDK for accurate auditing.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 19 days ago
Requires