Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+22 moreClaude CodeCursorCodex CLIMuseDots
OpenClaw
Manus
Grok
Grok Bot+18 agentsToken Assassin: Reduce AI Agent Costs
Understand prompt and trace cost drivers, with explicit assumptions and regression guidance.
Free
Token Assassin: Reduce AI Agent Costs
Example session with this skill installed
Analyze this local agent prompt and trace. Identify exact duplicates, separate estimates from measured usage, and propose a diff preserving authorization rules. Do not send the files elsewhere.
- Read your context and instructions
- Compiled the token assassin
Illustrative example, not a real executed product run.
Summary: the agent receives a bounded input for Token Assassin and produces a reviewable draft. No deployment, external action or approved file modification is claimed.
Deliverable
- Scope: inspect only the supplied material and the requested workflow.
- Input inventory: list the files or requirements received, with any missing prerequisite.
- Evidence: cite the relevant source section for each finding or requirement. Mark anything that cannot be established from the input as requiring verification.
- Assumptions: record the reason, impact and validation needed for each assumption.
- Proposed work: produce the domain deliverable requested in the prompt, with dependencies, risks and an explicit distinction between recommendations and completed actions.
- Validation: provide acceptance checks, identify available tools and mark checks not executed.
Next steps: review assumptions, supply missing inputs and run the applicable checks before accepting the deliverable.
This is an illustrative output structure. Replace it with a real run of this skill before publishing.
Connects securely to your tools. The creator never sees your data.
About this skill
Token Assassin is a single instruction skill for an AI agent that can read SKILL.md and work with user-supplied context. Place the skill in your agent's documented skill location after checking its format requirements. Installation compatibility has not been tested.
The instructions guide the agent through a structured review or planning workflow. This draft package contains the skill document, not a standalone application. Its source marks the release as NOT READY FOR RELEASE.
The problem
Unstructured requests can hide missing requirements, weak evidence and unsafe assumptions. This skill specifies what the agent should examine, how to report uncertainty and what the output should contain.
What you get
| Part | Included |
|---|---|
| Instruction skill | 1 SKILL.md |
| Bundled scripts | 0 |
| Reference files | 0 |
| Plugin manifest | None found |
| License | Not supplied |
How it works
The agent follows the instructions using its own available tools. The package does not supply an executable engine. Every conclusion still needs evidence from the user's input.
Uses
Find exact duplicate instructions. Request a focused deliverable with evidence and stated limitations. Try: "Find exact duplicate instructions for the supplied input. State assumptions and identify what needs verification."
Review semantic redundancy. Request a focused deliverable with evidence and stated limitations. Try: "Review semantic redundancy for the supplied input. State assumptions and identify what needs verification."
Measure tokens with stated method. Request a focused deliverable with evidence and stated limitations. Try: "Measure tokens with stated method for the supplied input. State assumptions and identify what needs verification."
Model costs from supplied prices. Request a focused deliverable with evidence and stated limitations. Try: "Model costs from supplied prices for the supplied input. State assumptions and identify what needs verification."
Review repeated tool calls. Request a focused deliverable with evidence and stated limitations. Try: "Review repeated tool calls for the supplied input. State assumptions and identify what needs verification."
Identify oversized tool responses. Request a focused deliverable with evidence and stated limitations. Try: "Identify oversized tool responses for the supplied input. State assumptions and identify what needs verification."
Propose prompt diffs. Request a focused deliverable with evidence and stated limitations. Try: "Propose prompt diffs for the supplied input. State assumptions and identify what needs verification."
Plan regression checks. Request a focused deliverable with evidence and stated limitations. Try: "Plan regression checks for the supplied input. State assumptions and identify what needs verification."
Guardrails
Do not claim tests passed, changes were applied or external actions completed without evidence. Review permissions and uncertainty before acting. The underlying agent determines actual access and data handling.
Who it is for
People who want structured assistance for this domain and can review the result. The current draft is not for buyers expecting tested software, a complete command-line tool or verified platform integrations.
Safe handling
Only provide inputs you are authorized to process. The package has no bundled executable scripts, network client or credentials. This does not make the host AI agent offline: its provider and tool settings determine how data is processed.
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
Free forever
No account required to browse
Trust & safety
Security scanned
Verified clean 6 days ago
- Free to download with an account