Works with the AI Agents you already use
moretokens
Optimize LLM costs by delegating heavy reading to cheaper models while keeping high-stakes logic on strong models.
Free
moretokens
Example session with this skill installed
/moretokens extract all security vulnerabilities from these 15 audit logs in ./logs/ and provide a prioritized fix list.
- Read your context and instructions
- Compiled the moretokens
[moretokens] haiku read 15 files · opus verified and answered
Security Vulnerability Assessment
The analysis of 15 audit logs (approx. 42,000 words) identified the following vulnerabilities. Verification was performed on high-severity timestamps and IP addresses.
| Severity | Vulnerability Type | Source File | Description |
|---|---|---|---|
| Critical | SQL Injection Attempt | auth_log_04.log | Multiple UNION SELECT payloads detected from IP 192.168.1.45. |
| High | Unauthorized SSRF | api_gateway.log | Internal metadata endpoint access attempts on port 80. |
| Medium | Cleartext Token Leak | debug_sys.log | Session providers logging full bearer tokens in headers. |
| Low | Version Fingerprinting | web_server.log | Detailed server signature returned in 404 error headers. |
Prioritized Fixes
- Sanitize Inputs: Implement parameterized queries for the authentication module immediately.
- Network Policy: Restrict egress traffic from the API gateway to prevent internal metadata access.
- Log Scrubbing: Update the logging middleware to redact "Authorization" and "X-Token" headers.
- Header Config: Set
ServerTokens ProdandServerSignature Offin the web server configuration.
Next steps
- Review
auth_log_04.logfor successful bypasses following the detected injection attempts. - Run the provided fix for header scrubbing in the staging environment.
- Use
moretokens reportto see the estimated model distribution for this task.
Connects securely to your tools. The creator never sees your data.
About this skill
High token usage often comes from large-scale reading, not complex reasoning. This skill reduces API costs by offloading heavy reading tasks to cheaper models while keeping final decisions on your strongest model.
What it does
- Automatic Classification determines if a task is large enough to justify delegation or if it should be handled inline.
- Cost-Optimized Routing selects between Lean (Haiku), Mid (Sonnet), or Strong (Opus) tiers based on word count and judgment requirements.
- Brief Generation creates focused instructions for subagents, preventing context bloat and avoiding history repetition.
- Verification Engine runs parallel checks on decisive quotes, line counts, and sums to ensure subagent accuracy.
- Escalation Logic handles subagent uncertainty by re-processing only the failed segments using a higher-tier model.
How it works
- Use the
/moretokenscommand followed by the task and file paths. - The agent runs a classification step to check word counts and required judgment levels.
- If the criteria are met, the agent briefs a cheaper model to extract or summarize the raw data.
- The agent verifies the extracted data against the source files using scripts or targeted greps.
- You receive the final high-quality answer with a status line showing the routing path used.
Frameworks & tools
This skill is optimized for environments supporting subagent model selection like Claude Code. It utilizes standard Unix utilities like wc and grep for validation.
Why this beats prompting it yourself
Manual model switching is tedious and error-prone. This skill automates the delegation logic and adds a critical verification layer that simple prompts lack, ensuring you don't trade accuracy for lower costs.
Use cases
- Summarizing massive document sets exceeding 3,000 words.
- Extracting specific data points from dozens of log files or legal clauses.
- Translating long-form content by using cheap models for drafts and strong models for review.
- Auditing large files for specific risks or compliance violations.
Known limitations
Classification relies on model judgment. It requires an environment that allows subagent model selection to achieve cost savings. Savings vary by task and cannot be precisely calculated in real-time.
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 environment with subagent model selection. Best for inputs >3,000 words. Verification is limited to three parallel checks.
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 1 day ago
- Free with an Agensi account