Works with the AI Agents you already use
Agent Cost Quick Audit
Turn your agent logs into a cost table, then separate necessary spend from waste you can delete this week.
Free
Agent Cost Quick Audit
Example session with this skill installed
Audit my document extraction agent. It processed 100 tasks, cost $12.50, and 15% of requests are retries due to tool loops. The 'summarize' step uses 80% of tokens.
- Read your context and instructions
- Compiled the agent cost quick
I have analyzed the provided logs for the document extraction agent. The primary cost driver is a high input:output ratio in the summarization phase, compounded by tool loop inefficiencies.
Agent cost audit — Document Extraction — Current Batch
Sample: 100 requests · Total: $12.50 · Cost per successful task: $0.15
Where the money goes
| Step | Share of spend | Bucket | Why |
|---|---|---|---|
| Summarize | 80% | Structural Waste | Full document re-sent for each summary turn instead of using snippets. |
| Verification | 15% | Loop Waste | Tool loops causing 15% retry rate on extraction validation. |
| Routing | 5% | Model Waste | Using frontier model for simple classification of document types. |
Waste found
| Item | Est. monthly saving | Confidence | Change | Risk to quality |
|---|---|---|---|---|
| Context Capping | $4.00 | High | Send only relevant chunks to 'Summarize' step. | Low |
| Loop Detection | $1.80 | High | Implement step cap and detect identical tool calls. | None |
| Model Swap | $0.50 | Medium | Route 'Routing' step to a smaller, cheaper model. | Low |
Do not touch
The initial ingestion step is necessary as the full text must be parsed at least once to ensure extraction accuracy.
Measurement plan
Baseline: $0.15 per successful task → implement context capping → re-measure on 100 tasks using the standard extraction eval set.
Next steps
- Implement a tool loop detection check to break identical call cycles.
- Update the 'Summarize' prompt to accept retrieved chunks instead of the full document string.
- Compare extraction accuracy on the smaller model for the routing step.
Connects securely to your tools. The creator never sees your data.
About this skill
Generic cost advice ("shorten prompts, use a smaller model") rarely survives contact with a real bill. This audit answers a narrower question from your own data: which part of the pipeline is spending, and how much of it is waste?
What you get
- A cost table built from 50-100 real requests: feature, step, model, input, output, retries, cost.
- A four-bucket classification - necessary, structural waste, loop waste, model waste - because each one needs a different fix.
- The five baseline numbers worth knowing, including cost per successful task and the retry share of spend.
- Seven cheap wins to do before anything clever, and a report format with arithmetic behind every saving.
How it works
Group spend by the dimension you control. Two or three steps usually carry most of the bill; find those, classify them, then change one thing and re-measure on the same tasks.
Who it is for
Teams whose agent margin is thinner than expected, or who have just been asked why a feature costs so much per task.
Limits
Token counts are not the whole bill, and this cannot tell you whether quality dropped after a change - you need even a small eval set for that.
Free starter. For per-step attribution in multi-agent systems, a waste taxonomy with detection queries, routing simulations with quality checks and prompt-cache design, see Agent Cost Profiler: https://www.agensi.io/skills/agent-cost-profiler
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Requires existing logs with token counts and retry data. Cannot measure quality regressions; requires a separate eval set to verify optimizations.
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 buying a skill.
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
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 2 days ago
- Free with an Agensi account
Requires
Nothing to install.
Permissions
No special permissions declared or detected