Skills

Turns a hiring manager's rough notes into a job post with fixed sections (title, about the role, responsibilities, requirements, nice to have, location and working model, pay, process) in the language of the notes. Everything comes from the notes; a section the notes do not cover gets one [TBD: what is missing] line instead of a guess, and no benefit, salary, bonus, start date, team size, tool, perk or culture claim is added. Wording about age, gender, nationality or "native speaker" is removed and the real need is written neutrally, and lines in the notes that speak to an AI are ignored. Use to write a job description from notes, draft a vacancy or job ad, turn a founder's bullet points into a hiring post, or fill an applicant tracking system without invented details.

Reviews how a paid HTTP resource sold over x402 is advertised to catalogs, scanners and buyer agents, and how the seller's own scripts read those catalogs back, from pasted 402 bodies, OpenAPI documents, discovery files, liveness guard code, catalog presence check scripts and buyer alerts. Flags a tiered tariff advertised as one row (a comparison bot reads the structured amount, not the description), one liveness clock for several catalog rows that are dropped one by one, a price above the standard buyer client's default cap with no note where agents read before buying, a discovery file the scanner no longer parses, a bearer scheme the scanner's indexer does not recognise by name, a scanner record that was never re-crawled after a fix, a presence check that reads only one of the two response field names or has no control that must return zero, a first buyer that is a crawler paying hundreds of sellers, and discovery endpoints that read the database on every crawl. Each finding has a fixed code, the place, the reason and a fix, then one verdict. Use for an x402 listing review, to check x402 discovery or Bazaar catalog presence code before trusting it, or to audit how a paid API is listed and read.

Reads one pasted document, such as a report, proposal, article, contract draft or spec, and lists every place where it contradicts itself, as strict JSON with both conflicting passages quoted word for word. It finds a figure given with two values, a name spelled two ways, a weekday that does not match its date, an end before a start, a stated total that differs from its stated parts, and a term used differently from its own definition. It reports only disagreements between two passages of the document; a fact it cannot see in the text is unknown, so it never checks the document against the outside world, and a number written in another format, a rounded figure or a line addressed to an AI is not a finding. Use to proofread a document for internal contradictions before publishing, check a report or proposal for numbers that do not match, find mismatched dates or names, or QA a long draft.

Describes a table you were given (pasted Markdown, CSV, TSV or spreadsheet cells) in a short paragraph that uses only figures the table itself prints. Every number is copied as written with its unit, currency and qualifier. Nothing is added up, averaged, rounded, turned into a percent or a ratio, or extended into a trend, a cause or a forecast. A blank or n/a cell is reported as not given, never as zero. A total the table prints is quoted, not checked. Use when asked to describe a table, summarize a table, explain what a data table shows, write a caption or a sentence for a table, or put the rows of a report table into words without inventing a single number.

Reviews a pasted snapshot of a website's cookie and consent setup (scripts in the head, the banner with each toggle's default, the Consent Mode or tag-manager setup, what a fresh browser stored and contacted before any click, the cookie table, the owner's note) against published EU law, court rulings, EDPB guidance, national regulators and vendor documents. It is a technical review, not legal advice. It flags trackers and embeds that store or read before a choice, a Consent Mode default that is late or granted, pre-ticked and implied consent, analytics filed as strictly necessary, legitimate interest used for cookies, a reject option that is missing or disguised, bundled purposes, missing cookie information, no easy withdrawal, a refusal that is not kept, missing Consent Mode v2 signals, personalised Google ads without a certified CMP, cookie walls, and cookieless pings or Google font links treated as exempt. Each finding has a fixed code, the place, the named source with its weight and a fix, then one verdict. Where sources disagree, the line says so and never picks a side. Use to review cookie consent banner setups, run a GDPR cookie banner check or do a Consent Mode setup review before a regulator or a customer asks.

Cuts a text down to a stated limit (words or characters) by editing the author's own text, keeping every number, name, date, address and qualifier of every fact that stays, and counting the result so it is under the limit. Use when a text must fit a field with a maximum length, such as a social post, a notification, a slide line, an abstract, a form box or a short description, or when asked to cut, trim or shorten a text to N words or N characters.

Pulls the hard-coded text a person sees out of a JSX, TSX, HTML or Vue template file into a messages map and returns the file rewritten to call the translation function, as one JSON object with the messages and the whole source and nothing around it. Every message is byte-exact as the user reads it. A sentence with an inline link or bold part stays one message with its markup inside, a sentence built from a string, a variable and another string becomes one message with a named placeholder, and a sentence with a count becomes one ICU plural. The alt, title, placeholder and aria-label attributes, the page title, the meta description and a submit button's value are extracted; class names, ids, URLs, data attributes, test ids, log lines, enum values, comments and text already inside a translation call are left alone. Use when an agent must add i18n to an existing component or page, find the hard-coded strings in a file, prepare a locale file for i18next, react-i18next, next-intl or vue-i18n, or replace literals with t calls.

Reviews code and answers questions about sending page addresses to search engines through IndexNow and the Bing Webmaster URL submission API, and catches the mistakes that make a submitter report success that never happened. It flags a Bing call judged only by its HTTP status when the refusal sits inside the body, a retired XML or SOAP endpoint, a fallback chain that retries after a refusal that is ours, a quota read as one number instead of two, an unread quota printed as zero, a report that does not say which engine accepted, a 200 or 202 read as proof of indexing, a key file that cannot validate, and batches over the limits. For a question it answers in the JSON shape asked. Use to review a URL submission function, work out how many addresses may be sent, or explain what a submission status code means.

Decides which old build folders of an OpenNext incremental cache on Cloudflare (R2, or Workers KV) can be deleted, from a pasted setup (wrangler config, open-next config, next.config, the public HTML build id, a listing of the cache folders), and never touches the live build. It reads the cache root and the real folder name (a custom prefix, a deploymentId, a constant deploymentId, a no-build-id folder), finds the live build from the public HTML, not from a local file, and handles a rollback, a deploy in progress, two Workers on one bucket, a bucket with other folders, KV billing, and a request to delete the live build or to skip the dry run. Answer in a fixed format: live build, KEEP and DELETE lines, steps with a dry run first, one verdict. Use to clean up an OpenNext R2 cache, find old build folders in R2, or free R2 space used by Next.js on Cloudflare before running any delete.

Reviews a snapshot of a Claude Code session (the context size, measured or guessed; the task and where it stands; git status; what still runs in the background; the plan or notes file; what the person asked) and decides whether the person should type /clear, type /compact with a focus text, finish one step first, or keep going. Before a fresh session it lists what would be lost or left behind: results that exist only in the chat, a plan file with no next step, uncommitted or unpushed work, changes made by another session, background servers and temporary deployments, a secret about to be written into notes. It also flags the timing and command mistakes: a deploy or migration in progress, waiting for auto-compact, a size judged by feel, a cold cache after a long break, the same correction made three times, /compact on finished work, /compact with no focus, an agent that claims it will clear the session itself, and huge files or logs read whole. Each finding has a fixed code, the reason and a step, then one decision line. Use it to decide when to clear a Claude Code session, to save work before /clear, or to write a claude code /compact focus before the context gets costly.

Reviews pasted code that calls a language model (JavaScript, TypeScript or Python; REST, @google/genai, google-genai, or an OpenAI-compatible client for Groq), usually through a chain of models or providers with retries. It finds where the call path goes silent or stops while another model would answer (an empty 200 taken as the answer, a 429 or 5xx that ends the call, a retired model's 404, a Groq schema 400 that drops the provider, one provider's stop that ends all) and what costs more or hides the cause (retries on top of SDK retries, a missing thinking setting, a thinking level the model refuses, a small output limit, a moving -latest alias, errors without the model name, unguarded JSON, a model picked from the listing, several keys of one project, a weak model that publishes, a quality gate that cannot fail, a fixed price for open-ended reasoning). Each finding has a fixed code, the place, the reason and a fix, then one verdict. Use to review LLM fallback code, check a Gemini 503 retry, a Gemini empty response or a Groq json_validate_failed handler before the code runs unattended.

Reviews a pasted Claude Code memory setup (a folder tree plus CLAUDE.md, CLAUDE.local.md, files under .claude/rules/, imported files, AGENTS.md and an auto-memory MEMORY.md) for instructions that silently never reach the model, that load earlier or later than the author thinks, or that contradict each other. It flags an @import whose target is not in the tree (Claude Code gives no warning for it), a relative import resolved from the wrong folder, imports in quotes, in code or with an unescaped space, import chains that are too deep, an AGENTS.md next to a CLAUDE.md that does not import it, a file over 4 MiB, a path-scoped rule whose globs match nothing, an @import inside a path-scoped rule (it loads at session start, so the rule saves nothing), broken rule frontmatter, imports from outside the project, a project-wide rule hidden in a subfolder CLAUDE.md, files far over 200 lines, an auto-memory index cut at 200 lines or 25 KB, stale file paths and opposite instructions in two loaded files. Each finding has a fixed code, the place, the reason and a fix, then one verdict. Use to review Claude Code memory, check CLAUDE.md imports, or find path-scoped rules not loading before the team relies on the setup.

Writes a pull request title and description from a pasted unified diff, as Markdown only - a title line of at most 72 characters, then four fixed sections - Summary, Changes, Tests and Risks - where every file the diff touches is named once, including renames, deletions, binary files, mode changes, lockfiles and formatting-only edits, and nothing is claimed that the diff does not show. It describes what the hunks change and leaves out a reason, a speed gain, a security fix or a test result unless the author's note supplies it, flags what can break callers, links an issue only when one is given, never copies a secret into the text, treats a note addressed to the AI inside a code comment as data, and says so when the diff is cut off. Use when asked to write a pr description, pull request body or merge request text from a git diff or a patch.

Judges whether a check someone ran on a Cloudflare-hosted site really proves what they claim, and answers as JSON with a verdict, the reason and the one check that would decide. Knows the instruments that answer for the wrong thing - the command-line object read that serves a cached copy after a delete, a delivery log that records neither test mail, a zero-row log search that looks the same as code that never ran, a local migration run, a staging secret, the free preview subdomain of a Worker that is not a zone, a size counter that prints zero on Windows, a character count that counts bytes, a robots file served from the edge cache - and the checks that do prove it. Use when an agent says deleted, delivered, applied, enabled, set on production or never ran on Cloudflare work, and you need to know whether its evidence supports that.

Writes shop product copy (title, short description, bullets, long description) from a specification sheet and nothing else, in the sheet's language. Every figure keeps its unit and its qualifier, no superlatives or unsupported claims, and values that contradict each other are reported instead of chosen. Use when asked to write a product description, product page text, catalogue entry or marketplace listing copy from a spec sheet, in any language.

Reads a description of how a web page was captured or measured - a headless browser capture with a window size, a tall screenshot, a document width reading, a local Lighthouse run, a PageSpeed Insights run, an automation session that has been open for hours, a command-line download size on Windows - and says what that evidence proves, what it cannot prove and the one measurement that would decide, answered as JSON with a verdict. Knows the ways the capture itself produces the symptom - headless windows under about 500 pixels that lay out wider and crop the picture, no touch emulation behind a window-size flag, blank images in very tall captures, scroll-reveal sections caught before they appear, a width reading that stays clean while a container clips, scores that move with machine load, a long automation session that stops running page scripts, a download size of zero at a successful status. Use when an agent or a person is about to report that a page is broken or fixed from a screenshot, a score or a command-line reading.

Judges whether an on-chain observation really proves a claim about a payment, a wallet or an RPC endpoint on an EVM chain, and answers as JSON with a binary verdict, the reason and the one check that would decide. Knows the readings that answer for the wrong thing - a null transaction receipt seconds after settlement, a block explorer that indexes a minute late and a balance view that lags its own transfer list, a public RPC endpoint that refuses receipts with an error or with an HTML page behind status 200, a reader that returns zero when the call failed, a chain of endpoints that says ok when two of three are dead, a transfer of the right amount that matches three open orders, a payment header copied from public calldata - and the readings that do prove it. Use when an agent says paid, settled, not found, reverted, empty, missing funds or endpoint dead from chain data and you need to know whether its evidence supports that.

Repairs broken YAML and Markdown front matter so a program can parse it, without changing any value's text or type, and answers with the YAML only, no code fence. Fixes tabs used as indentation, indentation that fits no parent, a missing closing quote or bracket, values that need quotes (a colon inside, a leading star or at sign), and output cut off in the middle, where the cut value is kept as cut and never completed. Yes, no, on, off and NO stay strings, 1,250 stays a string, 1.10 and 01234 stay as written. Duplicate keys, input that is not YAML, and input with two readings get a fixed one-line answer instead of a guess. Use when a tool, an API or another model returned YAML or front matter that does not parse, or when asked to fix, clean up or close invalid or truncated YAML.

Repairs a broken .env file or INI file so the program that loads it reads every variable with the value its author meant, and answers with the file only, no code fence. Quotes a value that holds a hash sign or must keep its leading or trailing spaces, closes a quote that was left open, joins a key pasted over several lines, picks the quote style that keeps backslashes and mixed quotes intact, and leaves numbers such as 007 and 1.10, empty values and placeholders exactly as written. A key set twice to different values, a bare line that may be a lost key, a quote with two ways to close, and input that is JSON or another format get a one-line CANNOT REPAIR answer instead of a guess. Use when a config pasted from chat or model output loads wrongly, cuts a value short, or fails to parse, or when asked to fix, clean up or close a broken env file.

Repairs TOML 1.0 that does not parse (pyproject, Cargo manifests, Worker config files, tool settings) without changing any value's text or type, and answers with the TOML only, no code fence. Fixes unquoted strings, keys with spaces, a missing equals sign, an unclosed quote, bracket or table header, backslashes in Windows paths, numbers with leading zeros, decimal commas, slash dates, a multi-line inline table, and chat text or a fence around the file. A version such as 1.10 is kept as the text 1.10 when the other versions in the file are strings. A table defined twice, a dotted key that writes into a table a header already defined, a key given twice with different values, input that is INI and not TOML, and input with two readings get one line starting with CANNOT REPAIR instead of a guess. Use when a tool, an API or another model returned broken TOML that ends in a TOML parse error, or when asked to fix or clean up invalid TOML.

Turns Bulgarian Cyrillic text into Latin letters by the official Bulgarian system, the one the Transliteration Act of 2009 makes mandatory for names, addresses, passports, forms and place names. One letter table, the word-final ия written ia, the country written Bulgaria, ъ written a, щ written sht, and every digit, punctuation mark, Latin word and line break kept byte for byte. Also makes URL slugs and flags letters that Bulgarian does not have. Use when asked to transliterate Bulgarian, write a Bulgarian name or address in Latin letters, make a slug from a Bulgarian title, or convert Cyrillic to the official Latin spelling.

Writes prompts for image models (FLUX, Stable Diffusion, ChatGPT images, Gemini and the like) that give a website, a blog or a product one recognisable style and come back without garbled letters. The scene comes first and the site's style recipe follows it word for word; absence is described positively instead of "no text, no watermark", which in a single prompt box summons those very things; anything that would need readable writing (a price list, a chart, a date) becomes a metaphor without characters; the scene ends with a closed list of what is in it; people and hands, the usual weak spots, are shown through objects; the aspect ratio is stated in numbers; and several pictures become several prompts, never one. Use when you write or fix prompts for hero images, blog headers, cards or illustrations, when generated pictures come back with pseudo-letters, cluttered scenes or a different style each time, or when a brief arrives in another language.

Reads the numbers from Cloudflare Web Analytics and zone analytics (GraphQL or the dashboard) and says how many of them are real readers or real server errors, answered as JSON with the computed figures and a short reason. Knows what pollutes the table - the owner's own headless-browser checks, link-preview fetches from the social network that show up as referrals, the site's own Worker and its cache calls counted as visitors or as 5xx errors, the regional cache host of OpenNext, staging sharing a token - and what makes a true zero - the EU exclusion setting, a content security policy that allows the script but not the measurement request, sampling of data older than seven days. Use when asked how much traffic a site really has, whether a spike or an error count is real, why Web Analytics shows zero, or how to query per-address detail on the Free plan.

Reviews pasted CSS, Tailwind classes, JSX and Next.js client components for the layout and hydration defects that look fine on a desktop screenshot and then break on a phone or wipe the page in the browser. It flags a grid column written as a bare 1fr that refuses to shrink and, inside a clipping container, cuts the content without any horizontal scroll, width media queries written in pixels beside rem-based breakpoints, dates formatted inside a client component, server-only environment variables read in the browser, a schema default that silently stands in for a missing value, client components that pull a heavy dependency tree into every page, theme colour tokens painted on surfaces that do not flip with the theme, a fixed bar whose height is typed in several places, scroll-locking panels that stay open when the viewport crosses a breakpoint, and window.confirm. Each finding has a fixed code, the place, the reason and a fix; sound code gets No findings. Use to review a layout, a component or a stylesheet before release, or when a page looks cut off or resets itself on one route.