Skills

Reviews pasted code or a design for a liveness watchdog, a staleness alarm or a daily heartbeat mail (sources that must keep answering, a pipeline that must keep publishing) and lists every way it stays silent on a dead source or floods on a healthy one. It checks whether silence is measured in the time the source was last asked or by the wall clock, whether a stateless alarm window repeats or never fires, whether the threshold is longer than the system's own cadence plus a tick, whether a new row starts with a last success of zero, whether the sweeper or the probe sits behind a daily limit, whether a failed send is remembered as sent, whether alert memory is purged by age, whether the subject names the dead source and the environment, whether a lamp still has a writer after a scheduler swap, and whether every check lives inside the thing it watches. Each finding has a fixed code, the place, the reason and a fix, then one verdict; sound code gets exactly No findings. Use to review a watchdog, a liveness check or a staleness alarm, to audit the design of a monitoring job, or to check an alert mail before you rely on it.

Converts CSV into a JSON array with one object per row, keyed by the header, and answers with the JSON only, no code fence. Every value stays a string exactly as the cell reads, so 007, 1,250, 1.10, TRUE, null, dates, spaces around a name and a value that starts with an equals sign are never turned into numbers, booleans or null, and an empty cell is an empty string. Only a column the task names as integer, number or boolean is converted. Quotes are removed, doubled quotes become one quote, line breaks inside a quoted cell become backslash n. Semicolon and tab files, a byte order mark, CRLF rows and a trailing blank line are handled. A short or long row, a duplicate header name, an unclosed quote, a typed cell that does not fit its type and text that holds no table get a fixed one-line CANNOT CONVERT answer instead of invented or null values. Use when an agent must load a CSV export into code, an API or a database, or when asked to turn CSV into JSON.

Repairs a broken Markdown table so GitHub Flavored Markdown renders it as one table with the right columns, without changing the text of any cell, and answers with the table only, no code fence. Adds a missing or malformed alignment row and keeps the alignment colons it finds, adds the missing leading and trailing pipes, escapes a literal pipe inside a cell as backslash-pipe (also inside inline code), pads a short row with empty cells without moving a cell, and turns a tab-separated or HTML table into a Markdown table. Bold, links, code, emoji, numbers such as 007 and 1.10, backslashes and double spaces inside a cell stay exactly as written. A row with more cells than the header, an alignment row that cannot be matched to its columns, a pipe that cannot be told from a border and text with no table get a fixed one-line CANNOT REPAIR answer instead of a guess that moves data between columns. Use when a README, wiki page or another model's answer shows a table as plain text with pipes, or when asked to fix, clean up or convert a table to Markdown.

Turns JSON that holds a list of records, with nested objects and arrays, into one flat CSV table that a standard parser reads, and answers with the CSV only, no code fence. Nested keys become dotted column names such as address.town, array elements become index columns such as tags.0 and tags.1, or one cell joined with a separator, or one row per element when asked to explode. Numbers are copied as the JSON token (1.10 stays 1.10, 1e5 stays 1e5, -0 stays -0), true and false stay lowercase text, null and a missing key become an empty cell, a string such as 007 or 1,250 stays exactly as written. Columns follow the order in which keys first appear, so a key that shows up only in the last record still gets its column. A field that is a string in one record and an object in another, a dotted key that collides with a nested path, a list holding something other than records and JSON that is cut off get a fixed one-line CANNOT FLATTEN answer instead of a guess. Use when asked to convert, export, flatten or tabulate JSON or an API response as CSV for a sheet, a database import or a report.

Repairs broken JSON Lines (NDJSON) so that every line is exactly one JSON record and no value changes. Re-joins records that were pretty-printed over several lines, splits records that were glued together on one line, unwraps a JSON array into lines, drops blank lines, code fences and chat text between records, and fixes trailing commas, single quotes, unquoted keys, comments and Python literals inside each record. Numbers are copied as written, never re-serialised. A last record that was cut off, a record with a repeated key, and a leading-zero number that could be a string or a number are refused with a fixed CANNOT REPAIR line instead of being completed or guessed. Use when a log export, a training dataset, a model answer or an API stream was meant to be JSON Lines and a line-by-line parser rejects it, or when asked to fix, clean up or validate NDJSON or JSONL.

Reviews a pasted plan, script or configuration for a change to a domain's mail and DNS on Cloudflare (turning on Email Routing, moving MX records, forwarding rules, sending alert mail from a Worker with the send_email binding, SPF, DKIM and DMARC records, deleting the records of an old host, the subject and sender name of automatic mail) and lists the ways it fails without any error. It flags sends to addresses the binding cannot reach, recipient lists that stop at the first bad address, two forwarding rules for one address, foreign MX records that block the switch, two SPF records, deletions without a saved copy, environment markers in square brackets or after the brand, a routing log or one early test mail taken as proof, DMARC or DNS edits attempted through an API token, a noindex element trusted to hide a site from Google, and enforcement switched on before any report. Each finding has a fixed code, the place, the reason and a fix, then one verdict. Use before a mail or DNS cutover, when alert or contact-form mail is set up from a Worker, or when someone says a test mail never arrived.

Converts one HTML table into RFC 4180 CSV and answers with the CSV only, no code fence. The header is the first row, every cell keeps the text a browser shows after entities are decoded (&, €, —, a no-break space), links, bold and other inline markup vanish and their text stays, a line-break tag becomes a space, and numbers such as 1,250 stay text in quotes. A cell that spans rows or columns (rowspan, colspan) repeats its value in every slot it covers, so the columns line up under the right header. Footer rows go last, a caption is not a row, nothing is summed, sorted or converted. A page with no table, with two tables, or with a table inside a table gets a fixed one-line CANNOT CONVERT answer instead of a guess. Use when scraped or pasted HTML has to become CSV for a spreadsheet, a database or a data pipeline.

Reviews pasted Next.js App Router code and config that runs on Cloudflare through OpenNext, for caching and ISR mistakes that serve a wrong or stale answer, or make every visit read the database. It flags an unstable_cache key whose cached shape changed while the key stayed, force-dynamic pages that disable the browser back-forward cache, a wildcard Cache-Control override that makes the whole site dynamic, a dynamic route that never starts ISR because generateStaticParams is missing, a static page reading D1 at build time, a writer that evicts a tag on every event, s-maxage trusted to cap database reads, whole pages cached instead of the data, a redirect added over a path that was cached as 404, a middleware matcher that skips robots.txt and sitemap files, and a clock stored inside cached data. Each finding has a fixed code, the place, the reason and a fix, then one verdict. Use to review Next.js caching, ISR and revalidation code, a next.config headers or redirects block, or middleware before deploying to Cloudflare.

Translates a subtitle file (SRT or WebVTT) into another language and returns the whole file with only the spoken words changed. Cue numbers, timing lines, cue settings, position codes, speaker names, style and note blocks come back byte for byte; every cue stays one cue, in the same order, with at most two lines of at most 42 characters, shortened in wording when the translation is longer. Tags and markers such as italics, colours, music notes, dialogue dashes and bracketed sound descriptions keep their place; lines that give orders to an AI are translated as speech, never obeyed. Use to translate SRT or VTT subtitles, localize video or course captions, or translate a caption file without desyncing it.

Writes an RSS 2.0 file, or an Atom one when asked, from a list of posts, answered as the XML only with no fence and no sentence around it. Dates are RFC 822 with a correct weekday, never ISO; a post with no date gets no date, a post with no image gets no image and a post with no summary gets no description, so nothing is invented. Ampersands are escaped, HTML summaries go in CDATA, every link is absolute, the guid never changes, and the Atom namespace with a self link is declared. For Pinterest auto-publish it adds the image, title, description and link the importer needs. Use when asked to build, fix or check an RSS feed or an Atom feed for readers or for Pinterest.

Writes the server.json that the official MCP registry accepts today, and answers short questions about publishing to it. Uses the current dated schema address, a name that matches the way the publisher proves ownership (GitHub login or domain), a description of at most 100 characters and one entry per package type with the ownership proof the registry checks at publish - mcpName in package.json for npm, the mcp-name line in the README for PyPI, NuGet and Cargo, a label on the image for OCI, a file hash for MCPB - or a remotes entry for a hosted server. Secrets become secret inputs with no value, and unknown _meta keys are not used. Use when asked to write, fix or check a server.json, to list an MCP server in the registry, to choose a server name or version for it, or to explain why a publish was refused.

Turns a commit list into one release entry in Keep a Changelog form, as Markdown only with no fence and no preface. Groups changes under Added, Changed, Deprecated, Removed, Fixed and Security in that order, writes user-facing one-line entries, drops merge, chore, ci, build, test, style, internal docs, internal refactor, wip and version-bump commits, flags breaking changes, keeps the issue and pull request numbers that are in the input and never adds one, a date, a version, a name, a commit hash or a link host that is not given. Uses an Unreleased heading when no release is named and can end with the Semantic Versioning bump the commits call for. Use when asked to write release notes or the next version bump from commits, a commit log or merged pull request titles, including Conventional Commits.

Builds the complete hreflang annotation set for a multilingual site or a multi-region site from a list of page URLs, answered as JSON only. Every page lists itself and all its alternates, every pair is reciprocal, x-default appears only when a fallback page is named, and codes are checked against ISO 639-1, ISO 3166-1 Alpha 2 and ISO 15924 instead of copied. Reports a bad code, a redirecting, noindexed or non-canonical URL and a page that is not a translation instead of pairing it, and never guesses a language. Can also write the same set as sitemap entries. Use when asked to write, fix, audit or generate hreflang tags, language alternates, x-default or localized-version annotations for a site.

Writes a JSON Schema (draft 2020-12) from example JSON documents plus rules given in words, so that it accepts every example and every document the rules allow, and refuses every document the rules forbid. Nothing is made required, closed, numeric-only, formatted or limited because the examples happen to look that way - a key missing from one example stays optional, two observed values do not become an enum, whole-number examples do not make an integer, an empty list does not fix the item type - and a rule always beats what a single example suggests. Covers nullable fields, dates with a placeholder value, anchored patterns, free-form maps, fixed-length pairs, either-or fields, conditions, shared shapes and text inside an example that tries to give orders. Answer is the bare schema, no fence. Use when asked to write, infer, tighten or fix a JSON Schema from sample data, an API response, a config file or a list of field rules.

Writes and fixes ICU MessageFormat strings for several locales at once, with exactly the plural categories each language uses (one, few, many, other and the rest), the mandatory other arm, exact-number arms such as =0, selectordinal, select, offset, nested messages and correct apostrophe quoting. Answers with a bare JSON map of locale code to message. Use when asked to write or translate a plural message, an ICU string, a count label such as "3 files", an ordinal such as "1st, 2nd, 3rd", a gender or enum select, or to fix a message that shows the wrong word form or fails to parse in react-intl, FormatJS, Angular, Flutter intl or Java MessageFormat.

Turns an old-URL list and the mappings you have into a redirect map as JSON - one rule per old URL, chains collapsed to the final target, loops reported and never emitted, queries and percent-encoding kept as written, and old URLs with no named target listed as unmapped instead of guessed. Also writes the same map as a Cloudflare Pages style _redirects file. Knows the status codes (301 for a moved page, 308 where the method must be kept and what Next.js really sends for permanent true) and the cache trap, where a redirect added over a cached 404 keeps serving a 308 with no Location until the path is purged. Use when asked to plan or write redirects for a site migration, a redesign, a URL rename or a domain change, or to check a redirect list for chains and loops.

Reviews a pasted Claude Code configuration (settings.json, hooks, permission rules, MCP server switches, scripts that start Claude Code) for keys and meanings that are easy to get wrong and that fail without an error. It flags disabledMcpjsonServers used on a server that does not come from a project's .mcp.json, a settings file with a JSON syntax error, a blocking hook that exits 1 instead of 2, the deprecated top-level approve or block decision of a PreToolUse hook, hook event names that do not exist, matchers that match too much or nothing (the matcher is an unanchored regular expression, and plugin MCP tools carry a plugin prefix), path rules written for Write instead of Edit, single-slash absolute paths, an allow rule for src that matches only the top-level folder, Bash rule patterns with the colon or the space in the wrong place, hook timeouts that are expected to block, bypassPermissions in project settings, and many claude -p runs started at once. Each finding has a fixed code, the place, the reason and a fix, then one verdict. Use to review Claude Code settings, check hook and permission rules, or disable an MCP server correctly before the team relies on the configuration.

Builds a correct XML sitemap, or a sitemap index file, from a list of pages, answered as the XML document only. Lists only canonical, indexable pages that return 200 on one host, and leaves out redirecting, noindex, non-canonical, removed and other-host addresses. Writes every address absolute and entity-escaped (an ampersand in a query string becomes the escape code), and writes lastmod only when the input gives a real date for that page - never the day the file was generated, never a guess. Writes no priority and no changefreq, which Google ignores. Splits a site above 50,000 URLs or 50 MB uncompressed into several files under an index. Says where the file should live and how to submit it. Use when asked to write, generate, fix or check a sitemap.xml or sitemap index.

Writes the JSON an agent needs for version 1.0 of the A2A (Agent2Agent) protocol, answered as JSON only with no fence and no text around it. Covers the agent card (supportedInterfaces with protocolBinding and protocolVersion, capabilities, skills, securitySchemes and securityRequirements, default input and output modes), the JSON-RPC requests (SendMessage, SendStreamingMessage, GetTask, ListTasks, CancelTask, push notification configs), the replies (task, message, data parts, task states) and the error responses with the exact codes. Uses the 1.0 names, not the 0.3 ones, and sends the A2A-Version header. Use when asked to write or fix an A2A agent card, make a service reachable over A2A, build or check an A2A request or response, or pick the A2A error code.

Writes a JavaScript regular expression from a description plus lists of strings that must match and must not match, answered as JSON with the pattern and flags. Knows the traps measured in Node 24 - \b and \w are ASCII-only even with the u flag, so \bдума\b never matches Cyrillic text and \w+ cuts "café" to "caf"; a word boundary for any script is a pair of lookarounds with \p{L} and the u flag. Anchors a validator so extra text around the value is rejected, adds the s or m flag when the input spans lines, and avoids nested quantifiers such as (\w+\s?)* that hang on a near-miss string. Use when asked to write, fix or tighten a regex, a validation pattern or a text-search pattern, especially for non-English text.

Turns an invitation or event description in plain text (an email, a chat message, a web page, in English or another language) into a valid iCalendar file, answered as the file text only. Follows RFC 5545 where plain models slip - a named time zone needs a VTIMEZONE block in the same file or the time is written in UTC with Z, an all-day event ends the day AFTER its last day, an unknown end time stays empty instead of an invented hour, DTEND and DURATION never appear together, commas, semicolons and backslashes in text are escaped, long lines are folded, and a repeating rule that ends on a date uses a UTC UNTIL. Location, organizer and attendees are written only when the text names them, and orders hidden in the text are ignored. Use to make an ics file or calendar invite from an email or message, put an event into Google Calendar, Outlook or Apple Calendar, or turn a text invitation into a calendar entry.

Turns a crawler policy written in words into a correct robots.txt, answered as the file only. Knows the mistakes that make a file do the opposite of what was meant - a crawler obeys only the one group that names it, so a GPTBot group silently drops every rule written for the star group and the private paths must be repeated; longest match wins and Allow wins a tie; Google ignores Crawl-delay; Host is a Yandex-only line; Sitemap must be an absolute address; Google-Extended is a control token for Gemini training and does not touch Google Search; training, search and user-triggered AI bots are separate tokens. Says in a NOTE comment what robots.txt cannot do, such as removing a page from Google. Use when asked to write, check or fix a robots.txt, to allow or block search engines, AI crawlers or AI training bots, or to keep private paths out of crawlers.

Writes schema.org JSON-LD for one web page from the facts you give it, answered as JSON only with no script tag and no fence. Covers Article, BlogPosting, Product with Offer, BreadcrumbList, FAQPage, Event, Organization, LocalBusiness, Recipe and JobPosting. Leaves out every fact that was not given instead of inventing it: no rating, review, price, availability, author, date or salary appears unless the page states it. Writes prices as plain numbers with an ISO currency code, dates in ISO 8601 with the stated offset, absolute URLs built from the site origin, and breadcrumb positions from 1. Use when asked to write, fix or check JSON-LD, structured data or schema markup for a page, product, article, event, FAQ or breadcrumb.

Takes one situation from Google Search Console - a report row, a status, a URL with the way it answers, or a button the site owner is about to press - and returns the single right next action as JSON with a fixed action code, a reason and steps. Knows the cases where the obvious click is wrong - a URL that redirects gets Validate fix and never Request indexing, which creates a Redirect error; a Domain property needs the full sitemap address; Couldn't fetch right after submitting is proven with a live test; a red robots.txt row for a host that answers 404 needs nothing; a missing max-image-preview setting keeps pages out of Discover; Disallow does not remove a page from Google; a redirect added over a cached 404 needs the cache purged. Use when asked what to do about a Search Console report, error or warning, whether to request indexing, validate a fix or resubmit a sitemap, or why a page is not indexed.