AI & Agent Tools

Building with an agent means writing a lot of small configuration files that fail quietly. A skill whose description never matches a task; a permission rule that does not cover the command you meant; a hook whose matcher names a tool that does not exist; a tool schema that validates and still gets called wrongly. These tools check that work before it costs you a debugging session. Write and lint skill files, subagent definitions and slash commands. Generate MCP server entries, hooks and plugin manifests, and test permission rules against real commands to see which one wins. On the prompt side, fill a template over a table of rows, diff two versions word by word with the token delta, lint a system prompt for contradictions, and build few-shot examples that cannot break out of their own tags. For the data around it: read and validate JSONL, check a fine-tuning dataset, reassemble a streaming response, and split a document into chunks that break on real boundaries. And for what it all costs: budget a context window, order a request so the cacheable prefix is as long as possible, and cost an agent loop turn by turn with the history it re-sends. Everything runs on your device — prompts, datasets and configuration files are never uploaded.

• Skill Builder — Write a SKILL.md with the frontmatter checked as you type.
• Skill Linter — Paste a SKILL.md and see what stops it being matched to a task.
• Subagent Builder — Build a subagent definition file with its tools and model set.
• Slash Command Builder — Build a custom slash command, with arguments and a tool allow-list.
• Project Instructions Builder — Draft a CLAUDE.md that earns its place in every session.
• Plugin Manifest Builder — Generate a plugin.json and see the folder layout it expects.
• Hooks Builder — Build settings.json hooks and catch the matchers that never fire.
• Permission Rule Tester — Test allow, ask and deny rules against real commands before you trust them.
• MCP Server Config Builder — Generate an MCP server entry for stdio, HTTP or SSE — and the CLI one-liner.
• MCP Tool Schema Builder — Define a tool and get the JSON Schema, with strict mode checked.
• Tool Schema Checker — Paste a tool definition and find out why it is rejected or mis-called.
• Prompt Template Filler — Expand one prompt over a table of rows and see what is missing.
• System Prompt Linter — Find contradictions, dead instructions and unfilled placeholders in a prompt.
• Prompt Diff — Compare two prompt versions word by word, with the token delta.
• Few-Shot Example Builder — Turn input/output pairs into tagged, JSON or chat-format examples.
• XML Prompt Section Builder — Wrap prompt sections in tags that content cannot break out of.
• JSONL Viewer & Validator — Read, filter and check a JSONL file line by line.
• Chat Transcript Converter — Turn a messages array into a readable transcript, and back again.
• Fine-Tuning Dataset Validator — Check a chat JSONL dataset for the errors that fail a training run.
• SSE Stream Viewer — Paste a raw streaming response and get the text, tool calls and usage back.
• Context Window Planner — Budget a context window across prompt, tools, history and the reply.
• Prompt Cache Planner — Order the blocks of a request so the cacheable prefix is as long as possible.
• Text Chunker — Split a document into overlapping chunks that break on real boundaries.
• Agent Loop Cost Estimator — Cost an agent loop turn by turn, with history re-sent and caching applied.
• Batch Request Builder — Turn a table of rows into a batch job, with unique ids per request.

Frequently asked questions

What are these AI tools for?

Building and debugging the things around a model rather than calling one: skill and subagent files, MCP server entries, hooks, permission rules, prompts, tool schemas, datasets and cost estimates. They are the checks worth running before a configuration file fails silently in a live session.

Do any of them send my prompts anywhere?

No. Every tool in this category is text processing that runs in your browser. Prompts, datasets, transcripts and configuration files stay on your device, which matters because these files routinely contain internal instructions and customer data.

Are the model prices and context limits current?

They carry the date they were last checked, shown next to every figure that uses them, and the rates are editable in the tools that spend them. Published prices move, so treat the built-in table as a starting point and confirm against the provider's own pricing page before you budget from it.

Do these only work with one provider?

The file formats — skills, subagents, slash commands, hooks, MCP — follow the conventions those formats use. Everything else is provider-neutral: JSONL, chat transcripts, streaming events, chunking, token budgets and prompt structure all work the same whichever model you send them to.

Why check a skill or tool description so carefully?

Because that one field decides whether the thing is ever used. A skill with a vague description sits unread; a tool with a thin description gets called with the wrong arguments. Both fail without an error message, which is exactly the kind of problem a linter is good at catching.