DefaultPerson
UserSkills for AI coding agents
Categories
Indexed Skills (6)
extract
Use when a notes/plan file contains URLs (YouTube videos, Telegram posts, articles) whose content needs to be brought into the workspace before /cleanup or further analysis. Replaces each URL with a local pointer to extracted text. Tradeoff: YouTube subtitles via yt-dlp, public Telegram via embed-page scrape, generic HTML via pandoc/curl. Private or auth-required resources are out of scope. For 1-2 URLs, copy-paste is faster. Triggers: "extract", "/extract", "развернуть ссылки", "expand links", "fetch URLs", "извлеки контент".
clarify
Use when you have a clean spec/notes file that needs to be made implementation-ready: decomposed into atomic tasks with verifiable acceptance criteria (Given/When/Then + shell proof commands), constraints, edge cases, and risk surface. Output is suitable for human implementation, mattpocock:tdd, or Claude Code goal feature. Tradeoff: slow and thorough — overkill for tasks under 1 hour. For freeform PRD use mattpocock:to-prd instead. Triggers: "clarify", "/clarify", "уточни спеку", "enrich spec", "обогати спеку", "decompose spec".
cleanup
Use when you have a messy notes/plan/chat dump (50+ lines, mixed topics, no clear structure) that needs lossless reorganization into a clean sectioned markdown file. Slow and thorough — overkill for files under 30 lines or already-sectioned docs. Triggers: "cleanup", "/cleanup", "почисти", "реорганизуй", "clean up", "rewrite plan", "sort plan", "plan-rewrite", "/plan-rewrite".
deepen
Find shallow modules (interface nearly as complex as their implementation) and propose deepening refactors — bigger implementations behind smaller interfaces. Optionally explores 2-3 interface variants in parallel via the Agent tool ("Design It Twice"). Use for architectural friction: hard-to-test modules, tangled callers, refactors that feel obvious but the right shape isn't. Triggers: "deepen", "/deepen", "improve architecture", "refactor this module", "make X more testable".
diagnose
Disciplined debug loop for hard bugs and performance regressions: build a feedback loop FIRST, then reproduce → ranked hypotheses → instrument → fix → regression-test → cleanup. Use when a bug is non-trivial, intermittent, or has resisted ad-hoc inspection. For one-line typos or obvious errors — just fix it; this skill is overkill. Triggers: "diagnose", "/diagnose", "debug this", "why is this broken", "почини баг", "разберись", "performance regression".
svgl
Fetch SVG brand/tech logos from the svgl.app public API into a project — by name, by category, or just listing categories. Downloads the actual `.svg` file(s) locally (handling light/dark theme variants and optional wordmarks), or with `--json` returns metadata/URLs only. Tradeoff: depends on svgl.app's catalogue (~660 logos) and its undocumented rate limit; only public logos. For a single logo you already have a URL for, just curl it. Triggers: "svgl", "/svgl", "get logo", "fetch logo", "svg logo", "download logo", "логотип", "svg иконка", "достань логотип".
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.