Claude Skills by ufo-ai

Load when a member asks to run or repair a supplied script, inspect or change files in /workspace, build or diagnose a project, execute a workspace command, or use `ufo llm` or `ufo tool`. Do not load for an unexecuted snippet.

Load when the member asks for ufo's house colours, type, spacing, radius, or other exact visual-system facts.

Load when asked to set up, change, or stop automatic review of new pull requests, or when a pull-request source alert wakes you. Not for one pull request reviewed now.

Load when asked to review, check, or look over a GitHub pull request, or when a coding reviewer's result for a pull-request review arrives. Load it before any other call, even when GitHub is not connected yet. Not for writing or fixing code, and not for automatic review.

Load when asked to inspect or change source code, fix bugs, produce patches or diffs, work in or clone repositories, use pinned commits, or check GitHub access. On repository tasks, load coding first and call no other tool until it returns. Not for reviewing a pull request; code-review covers that.

Load before visual choices without a full brand system or as fallback: color, typography, and visual hierarchy across any artifact (websites, slides, charts, documents).

Review documents for errors, inconsistencies, and factual accuracy. Load when the user uploads a document (PDF, DOCX, PPTX, or XLSX) and asks to review, check, audit, verify, proofread, fact-check, redline, critique, or find errors in it.

Load when producing or changing a Word document (.docx) — writing one from source notes, styling it, applying edits as tracked changes or comments, flattening a redline, or exporting its pages.

Load when creating, restyling, editing, or mining a PowerPoint deck (.pptx) — slides, layouts, templates, speaker notes, comments, and pulling text or data out of an existing deck.

Load when creating, editing, or analyzing Excel spreadsheets (.xlsx) — cells, formulas, recalculation, and tabular data.

Load before producing or transforming a .pdf. E.g. merge, rotate, or compress; OCR a scan and extract tables; repair or unlock a damaged file.

Load when defining or applying a visual theme to Office documents — colors, fonts, and styles for consistent docx/pptx/xlsx branding.

Load when a member asks you to write, draft, edit, or tighten prose meant for people to read (a post, announcement, email, changelog line, or doc), or asks to make writing sharper, more human, or less AI-sounding, or to check whether a draft reads as AI.

Load at first contact with an unfamiliar dataset (.csv, .parquet, database tables) including profiling structure, summary statistics, distributions, and missing values.

Load before writing or tuning SQL — .sql files, schema DDL, or queries against tables.

Load before drawing statistical conclusions. E.g. hypothesis tests, regression, correlation, significance, or confidence intervals.

Load when data or a computed result must be checked against rules that should already hold. E.g. validate rows against a schema; check uniqueness, ranges, and referential integrity across files; QA analysis output before delivery.

Load when visualizing data — choosing chart types and producing plots that communicate distributions, trends, or comparisons.

Load when a member asks what their scheduled reports said, or asks to catch up on, skim, or roll up recent reports. Not for writing a report and not for scheduling one.

Load when a member asks for recurring competitive analysis or intelligence: what competitors changed across pricing, launches, hiring, or press, or says \"Set up my first task: a daily competitive analysis.\" Not for a single page or price watch, a change to an existing task, or a one-off competitor question.

Load before performing multi-source web research. E.g. comparing options in a sourced table; build a bottom-up estimate; assemble an official timeline or ranking. Not for questions a few searches answer, nor for writing up existing findings.

Load before writing findings, notes, or sources up into a report or executive summary.

Load when proving a manifest-contributed skill reaches the registry, index, and sandbox with its bundled script.

Load when a member asks to create, change, pause, resume, or cancel a recurring task, notification, reminder, watch, or periodic check, or asks for a one-time reminder.

Load when a member asks for a page or site hosted outside ufo — a website, landing page, browser game, board, notes page, dashboard, or a build that runs its own server. Not an app or application, which is a ufo app, and not an app's homepage.

Load when building a full-stack browser application that needs server-side authentication, an API, a persistent database, or durable multi-user data. Do not load for a homepage, project board, informational site, browser game, or frontend-only interaction.

Load when a member asks to create, edit, repair, or save a reusable custom skill for the workspace.

The conformance sample pack's probe skill — proves a pack contributes a pack-level skill that parses into the registry and renders in the skill index alongside an extension's.