14 tools
AI Agent and Skill Builders
Compose agents, tools and skills — and price them as you write.
An agent’s configuration is fixed overhead: the system prompt, every tool name, description and JSON schema, plus the provider’s own tool-use preamble. It is sent on every request and on every iteration of every loop, whether or not a tool is called.
On an agent with a dozen tools that routinely exceeds a thousand tokens, and almost no framework shows it to you. Every builder here does.
- AI Agent GeneratorPick a kind of agent, get a complete configuration.
- AI Skill GeneratorPick what the skill does, get a valid file with triggers that fire.
- Structured Prompt BuilderAssemble a prompt from the parts that actually change the output.
- System Prompt AnalyzerYour system prompt is billed on every single request. See what it costs.
- AI Agent Configuration BuilderDefine an agent, export the config, and know what it costs before you run it.
- AI Skill File BuilderWrite a valid skill definition, validated as you type.
- Function Calling Tool Schema BuilderWrite tool definitions as one line each, get valid JSON schema.
- Tool Schema ValidatorCatch the malformed tool definition before the API rejects it.
- Tool Description OptimizerCut a tool description to the sentence that actually decides the call.
- MCP Server Config BuilderGenerate a valid Model Context Protocol server configuration.
- Skill File Frontmatter ValidatorCheck a skill file will actually be selected before you wonder why it never fires.
- Multi-Agent Handoff Prompt BuilderWrite the handoff so the receiving agent does not redo the work.
- Function Call Message FormatterBuild the tool_use and tool_result blocks by hand, correctly.
- Agent Memory Compaction PromptTurn a growing transcript into a compact state artifact.
Which one you need
If you are starting from nothing, use the generator. It produces a working configuration from an expert-authored blueprint — system prompt, tool schemas, guardrails and iteration limits — rather than a blank page and a hope.
If an agent already exists and the question is what it costs, the loop calculator gives you the number that matters: worst case per task, which is per-iteration cost times the iteration cap. Agents whose worst case has never been calculated are the ones that produce surprising invoices.
The skill and configuration tools come in once there is more than one agent and the problem becomes consistency — the same tool defined three slightly different ways across three agents is where the debugging time goes.
Before any of it reaches real systems, work through the agent safety checklist. Retrofitting limits onto a running agent means choosing between the limit and the traffic already depending on its absence.
Questions
What makes an agent different from a prompt with tools?
The loop. An agent decides whether to act again based on what came back, which means iteration count is not fixed and neither is cost. Everything difficult about agents — runaway spend, stuck cycles, compounding context — comes from that one property.
How do I stop an agent running up a bill?
Three limits, all enforced outside the model: a hard maximum iteration count, a total token budget per task, and a wall-clock timeout. Any one alone leaves a gap; a token budget without an iteration cap still permits a very long cheap loop.
Guides that use these tools
- How to reduce LLM API costs — Nine levers, ordered by what they return per hour of work. Most teams find 40% in the first three.
- Prompt caching, and what it actually saves — The single largest lever on a repetitive workload — and the cases where it does nothing.
- Why chatbot costs grow faster than your user count — Turn 20 pays for turns 1 to 19 again. Why per-request budgets are wrong by 3–4×.
- What an AI agent actually costs to run — Tool schemas and iteration limits multiply. Where agent bills come from, itemised.