Developer tools for your AI assistant

Show your remaining credits and what each tool costs.

Try: “How many JigWell credits do I have left?”

Produce a unified diff (like `git diff`, including the "\ No newline at end of file" marker) between two versions of a file or snippet. Returns {"output": the diff, "identical": true when there are no differences}.

Try: “Show me a unified diff between the old and new version of this function.”

Summarize source code: total, code, blank, and comment lines, TODO/FIXME/HACK/XXX markers, and the longest lines. A comment line is one that starts with the language's comment marker (with auto: #, //, /*, *, --, or <!--) or lies inside a multi-line /* ... */ or <!-- ... --> comment that starts at the beginning of a line. Docstrings and code with a trailing comment count as code.

Try: “How many lines, comments, and TODOs are in this file?”

Format code with Prettier, the most widely used code formatter. Supports JavaScript/JSX, TypeScript/TSX, JSON, CSS/SCSS/Less, HTML, Vue, Markdown, YAML, and GraphQL. Returns {"output"}; syntax errors are reported with their location.

Try: “Format this TypeScript file with Prettier using single quotes.”

Convert TypeScript (or TSX) to JavaScript by removing types. Returns {"ok", "code", "syntax_errors"}: on syntax errors ok is false, code is null, and each error has its line and column. Does not type-check; use typecheck_typescript for that.

Try: “Convert this TypeScript to plain JavaScript (ES2020, CommonJS).”

Type-check a self-contained TypeScript snippet with the real TypeScript compiler (ES2023 + DOM types).

Try: “Type-check this TypeScript snippet and explain each error.”

Minify JavaScript with Terser and report the size before and after, plus which mode was used.

Try: “Minify this script and tell me how much smaller it got.”

Run a regular expression against text and list every match with its position and capture groups (numbered and named). `start`/`end` count Unicode code points from 0 (Python indexing; JavaScript's UTF-16 indexes differ after emoji and other astral characters). Patterns that take over a second are stopped instead of hanging.

Try: “Test the regex (\d{4})-(\d{2}) against these log lines and show the groups.”

Run a JavaScript snippet in an isolated sandbox and return its console output and final value.

Try: “Run this JavaScript and show me the console output and result.”

Estimate how many tokens a text is for large language models (GPT, Claude, Llama) without any model's tokenizer. Returns {"estimated_tokens", "range": {"low", "high"}, "characters", "words", "lines", "method", "note"} and, when a price is given, "estimated_cost_usd". `words` counts words in spaced scripts and characters in Chinese, Japanese and Korean.

Try: “Roughly how many tokens is this prompt, and what would it cost at $3 per million?”

Validate a JSON Lines file (datasets, logs, fine-tuning files): every line is checked, and each problem is reported with its line number. Returns {"valid", "records", "blank_lines", "error_count", "errors": [{"line", "message"}], "format"} plus "roles" (messages per role) for format=chat. Blank lines are allowed and counted. At most 50 errors are listed; error_count has the total.

Try: “Check this fine-tuning dataset: is every line valid JSON in the chat format?”

Check a tool definition or its parameter schema the way AI platforms (MCP, OpenAI, Anthropic) and models see it. Errors mean a platform will reject it or a model can't fill it: invalid JSON Schema, a non-object root, `required` names that don't exist, a bad tool name, no description. Warnings mean the model will have to guess: parameters without a description or type, awkward names, deep nesting. Returns {"kind", "valid", "error_count", "warning_count", "problems": [{"severity", "path", "message"}], "property_count", "max_depth"}.

Try: “Check this MCP tool definition: will models use it reliably, and is it valid for strict structured output?”

Inspect a WebAssembly binary without running it: whether it's valid, its imports and exports (functions, memories, tables, globals), its size, and the toolchain recorded in its producers section (for example Rust, clang, wasm-bindgen). Returns {"valid", "size_bytes", "imports", "exports", "custom_sections", "producers"}. An invalid module is an error that quotes the engine's reason. Text-format .wat must be compiled first (wat2wasm).

Try: “What does this .wasm module import and export, and what was it compiled with?”

Validate and pretty-print JSON. Returns {"output"}. Numbers are kept exactly as written (no rounding, no 1e400 -> Infinity). Reports the line and column of any syntax error.

Try: “Is this JSON valid? Pretty-print it with sorted keys.”

Encode or decode text as base64, base64url, URL percent-encoding (url or form style), or hex. Returns {"encoding", "direction", "output", "warnings"}; decoded bytes that aren't valid UTF-8 and percent signs that don't start a valid escape are explained in `warnings` instead of being silently changed.

Try: “Base64-decode this string: aGVsbG8gd29ybGQ=”

Hash text (UTF-8) and return {"algorithm", "digest"} with the digest in lowercase hex.

Try: “What's the SHA-256 hash of 'hello world'?”

Decode a JWT's header and payload and explain its timestamps (iat, nbf, exp) as RFC 3339 UTC times with relative times, plus `expired`, `not_yet_valid` (nbf in the future), and `valid_now`, all judged at `as_of` (the server time, or the `now` you pass). `expired` and `not_yet_valid` are null when the token has no exp or nbf claim; `valid_now` is then true unless the other claim says otherwise.

Try: “Decode this JWT and tell me when it expires.”

Generate unique IDs (UUIDv4, time-ordered UUIDv7) or cryptographically secure tokens and passwords. Returns {"kind", "values": [...]}.

Try: “Generate 5 UUIDv7 IDs for my database seed file.”

Validate a JSON document against a JSON Schema and return every violation with its path and the failing keyword. Useful for checking AI structured output, API payloads, and config files.

Try: “Check this API response against our JSON Schema and list every problem.”

Generate a JSON Schema (draft 2020-12) from example JSON. Detects date-time, date, email, uri, and uuid strings. By default objects are closed (`additionalProperties: false`); set allow_additional_properties=true for an open schema.

Try: “Generate a JSON Schema from these example API records.”

Convert configuration or data files between JSON, YAML, and TOML. Returns {"output", "warnings"}.

Try: “Convert this docker-compose YAML to JSON.”

Look up an npm package: latest version, license, deprecation notice, weekly downloads, whether it ships TypeScript types, supported Node versions, and links.

Try: “Is the npm package request deprecated, and what should I use instead?”

Look up a Python package on PyPI: latest version, release date, supported Python versions, license, and links.

Try: “What's the latest version of requests on PyPI and which Python versions does it support?”

Check a package.json or requirements.txt against the live registry.

Try: “Which dependencies in this package.json are out of date?”

Check, compare, sort, and bump semantic versions using npm's exact rules: does 2.1.3 satisfy ^2.1.0, which is the newest version matching a range, is a change major/minor/patch, what comes after 1.4.9.

Try: “Which of these versions satisfy ^2.1.0: 2.0.9, 2.1.3, 2.9.0, 3.0.0?”

Fetch a public web page or API and return it as readable text, raw HTML, or parsed JSON.

Try: “Read https://example.com and summarize it.”

Convert a time between Unix timestamps, ISO 8601, and time zones, and say how far it is from now.

Try: “What time is 1700000000 in Tokyo and New York?”

Show when a cron schedule will next run in a given time zone (RFC 3339 times), plus `schedule`: what each field means in plain values (e.g. hours "9-17", days_of_week "MON-FRI").

Try: “When will '*/15 9-17 * * MON-FRI' run next in Berlin time?”