Skill Architect
Turns a job you repeat into a SKILL.md with inputs, a method, checks, and a deliverable.
Output Marketplace listing and a SKILL.md you can install.
Public directory · no sign-in · free
50 operating procedures. The eight below are the ones people open first. Pick one, paste a real brief, and the page writes the deliverable.
Start here
Runs on this page
Turns a job you repeat into a SKILL.md with inputs, a method, checks, and a deliverable.
~/.grok/skills/skill-architect/SKILL.md
01
Paste the brief. The page writes the deliverable from that run.
02
Every skill has a docs page with inputs, steps, and checks.
03
Download SKILL.md into ~/.grok/skills and start a new session.
Save any file to ~/.grok/skills/<slug>/SKILL.md and start a new session.
050 showing
05
Turns a job you repeat into a SKILL.md with inputs, a method, checks, and a deliverable.
Output Marketplace listing and a SKILL.md you can install.
Maps a business process into steps, roles, and handoffs without inventing a system you do not run.
Output Workflow map, roles, and handoffs.
Lifts a messy prompt into an SOP made of the sentences you already wrote.
Output SOP, inputs, and a QA checklist.
Matches a task to a skill in this directory by overlapping words, and says when the match is weak.
Output Routing decision, confidence, and the next skill to open.
Marks an AI draft for unsupported claims, bare numbers, and unfetched links. It does not fetch them.
Output Error log, a plain quality note, and a marked-up version.
05
Turns a question into a research plan and attaches only the public pages this run retrieved.
Output Questions, starting sources, method, and the deliverable to write next.
Scores retrieved pages by source type. The score is a type weight, not a truth score.
Output Source scorecard and a usage note for each page.
Reads public pages on a competitor and refuses any claim those pages do not contain.
Output Competitive intelligence report tied to URLs.
Lays out a category using retrieved pages, and does not draw whitespace the pages never mentioned.
Output Landscape limited to retrieved pages, plus an explicit gap list.
Sorts notes you already have into a memo. It will not invent the recommendation.
Output Memo with recommendation, alternatives, risks, and actions — or a gap where you left one.
05
Builds a PRD from the goals, people, and acceptance lines you supply. Missing parts stay missing.
Output Goals, personas, user stories, and acceptance criteria.
Compares only the components and stack options you name, against constraints you name.
Output Components, data flow, options, and risks you identified.
Structures a bug you can already reproduce. Hypotheses are only the ones you wrote.
Output Reproduction steps, hypotheses, tests, and fix options you proposed.
Scans code you paste for a fixed set of reliability and safety patterns. It does not run the code.
Output Prioritised notes from a text scan, plus what was not reviewed.
Turns acceptance lines you wrote into test cases. It does not invent expected behaviour.
Output A test table whose expected column is your wording.
05
Orders topics you already have. Search volume stays blank unless you typed a number.
Output A topic table and a writing order.
Checks a page you paste against a fixed conversion list. It does not invent a conversion rate.
Output A presence audit and a revision brief that quotes you.
Builds a campaign skeleton from your audience, offer, and budget. It does not invent performance.
Output Audience, angles, ad slots, budget, and tests — filled only where you did.
Drafts a sequence that uses only your offer and your proof. Missing proof stays a blank.
Output A message map and emails that do not add proof.
Writes a positioning line from the audience, category, and difference you supply.
Output Positioning, pillars, and a proof plan that stays empty where you have no proof.
05
Builds an ICP from who you say buys, who you exclude, and the triggers you have actually seen.
Output ICP, exclusions, triggers, and personas you named.
Prepares a prospect brief from public pages retrieved in this run. People are not guessed.
Output Context, triggers, and hypotheses tied to retrieved URLs.
Writes a multi-touch sequence with blanks where a real account fact should go. It does not research the account.
Output Email, short note, and follow-up, with personalisation left blank.
Prepares discovery questions and uses only the qualification rules you wrote down.
Output Call plan, questions, and your qualification signals.
Turns an opportunity into scope, deliverables, and a price only if you supplied the price.
Output Scope, deliverables, pricing logic, and next steps.
05
Documents a procedure from the steps you listed, with owners left blank where you named none.
Output Purpose, steps, roles, exceptions, and a checklist.
Stress-tests a model using only the revenue, cost, and channel facts you enter.
Output Revenue, cost, channel, risk, and what cannot be judged yet.
Extracts decisions, actions, owners, and deadlines from notes. Unnamed owners stay unnamed.
Output Decisions, an action table, and items with no owner.
Builds a scorecard from the role and the criteria you list. Weights stay unset unless you set them.
Output Role brief, scorecard, and one evidence question per criterion.
Builds the weekly review from the updates you paste. KPIs appear only with numbers you typed.
Output KPIs you supplied, blockers, priorities, and actions.
05
Builds an outline and a draft from your notes only. Source gaps stay empty until you add URLs.
Output Outline, argument, a draft assembled from your notes, and source gaps.
Splits one source you provide into channel cuts. It does not write new claims for each channel.
Output A short post, an email, a script, and a snippet, all from the source.
Cleans notes into a post without adding a view the notes did not contain.
Output A cleaned piece plus a list of what was not added.
Turns your points into a hook, beats, and a script. Visuals stay blank unless you described them.
Output Hook, beat sheet, spoken lines, and visual notes.
Edits for length, weasel words, and bare numbers. The change log says exactly what was removed.
Output An edited version and a change log.
05
Computes margin only from numbers you type, and shows the formula beside the result.
Output An assumption table and the arithmetic those assumptions support.
Adds up a forecast you already wrote. It does not project the next month.
Output The months you entered, the sum, and the change between adjacent figures.
Compares price points you name. It will not guess willingness to pay.
Output An options table and a recommendation only when your own rule picks one.
Computes runway only when you supply cash and a monthly burn. Risks are the ones you listed.
Output A risk register, runway arithmetic, and mitigations you named.
Drafts an update from your notes and your KPI figures. It does not enliven the story.
Output Narrative, KPI table, asks, and milestones you supplied.
05
Checks a description or paste of an interface against usability heuristics. It does not open a URL.
Output A heuristic checklist with quotes, and items that were not assessable.
Maps steps you list, and records pain only where you described it. Drop-off is not estimated.
Output Journey table and a backlog of improvements you named.
Ranks features only when you score value and effort. Otherwise it refuses to rank them.
Output A table and, when scores exist, an order.
Converts your brief into the document a designer needs, without adding references.
Output Audience, job, references, deliverables, and constraints.
Reviews a first-run flow you describe. It will not invent an activation rate.
Output Activation definition, friction, and a revised order only where you asked for a change.
05
Builds a curriculum from the outcome and topics you name, and splits time only if you gave hours.
Output Topics, a schedule, and a practice rule.
Turns your material into recall questions. It does not write the answer key.
Output Questions, a practice task, and a revision order.
Prepares practice for a role using your stories. It does not look up the employer.
Output Question prompts, a story map, and gaps you named.
Compares the skills you have with the skills you say the next role needs. It does not rank you.
Output A gap list, portfolio pieces you proposed, and application steps you named.
Turns the note types you named into a taxonomy and a weekly pass. It does not reorganise a real vault.
Output Folders, a capture template, and a weekly workflow.