Claude Skills by h0x91b

Drive and visually QA the dev-3.0 UI in a real browser (headless Chromium via agent-browser). Use when verifying a UI/UX change, reproducing a visual bug, taking screenshots of the running app, or self-QA before review. Triggers — "check the UI", "screenshot the app", "does this render", "QA this screen", "verify the UI change in a browser", "drive the app".

Create the initial Product UX Bible for an existing web or full-screen web app by deeply auditing the repository, using sub-agents when available, and generating docs/ux manifests, schemas, budgets, surface maps, action taxonomy, and design-token policies from observed code.

Principal UX architect skill for deciding WHERE a UI feature belongs, before it is implemented. Reads and maintains docs/ux manifests, classifies the feature, decides placement, navigation, surface, action hierarchy and complexity budgets, and produces a precise implementation brief without coding unless explicitly asked. Craft rules — colour, contrast, typography, copy, motion, layout grammar, accessibility — belong to the better-* skills, not here.

How to test and verify work in the dev-3.0 repo — which vitest config covers what, how to write a test that fits the house style, mocking Electrobun RPC and i18n providers, what coverage is actually expected, and the browser QA hand-off. Use when writing or fixing tests, deciding what a change needs covered, hitting a failing or flaky suite, or preparing a change for review. Triggers — "write tests for this", "which config runs this", "how do I mock the RPC", "is this covered enough", "the su...

Add title generation instructions to the dev3 agent skill. After running `dev3 current` at session start, agents now synthesize a concise 5–8 word title and call `dev3 task update --title "..."` when the auto-generated title looks truncated or too long.

Added a managed `dev3-bug-hunter` skill that installs for all supported agents and gives each bug-hunting run a seeded starting area plus analysis style. The skill is explicitly read-only, now reports findings with a compact ASCII summary table plus detail sections for terminal readability, asks whether `critical` and `medium` findings should become separate dev3 tasks, and requires every follow-up task to validate and reproduce the bug before any fix is attempted. The shared `~/.agents/skill...

Cut the dev3 skill's token footprint: the Claude SKILL.md no longer duplicates the protocol already injected into the system prompt (a PROTOCOL.md fallback covers sessions started outside the launcher), the skill invoke now uses `dev3 current --brief` instead of re-printing the full help and task description, and the shared skill body was rewritten ~40% denser for all agents. A Claude session now carries ~4k tokens of dev3 protocol instead of ~13.5k.

Moved the testing detail out of `AGENTS.md` into a new repo-local `/verify-changes` skill (which vitest config covers what, what a change needs covered, house style, mocking Electrobun RPC and `useT()`, reproducing flakes under load, coverage expectations), leaving the two verification gates and the browser-QA mandate inline where they cannot be missed. The 1-5 tip coolness rubric now lives on a `TipScore` union type in `src/mainview/tips.ts` instead of being restated in prose, and the key-fi...

Added the ask-dev3 agent skill — a router over dev3's features and workflows, auto-installed into every supported agent's skill directory. When the user asks "how do I do X in dev3", the agent loads the skill's map (main flow, on-ramps, verifying work, the tmux terminal, running several agents at once, vocabulary, troubleshooting, standalone features) and teaches the intended flow instead of guessing, then offers to do it. The map is GUI-first, covers a curated core of the most common newcome...

Short: Share a report as a gist link dev3 now ships a `dev3-share-artifact` skill to every agent: ask for a link to a report and the agent folds the multi-file artifact into one self-contained HTML file, publishes it as a secret GitHub gist, and verifies the preview URL before handing it over. The folding is a new `dev3 inline-html` command that embeds local CSS, JS, images and fonts, leaves CDN links alone, and refuses to write a page carrying a credential-shaped string or a missing local as...

Users (and the agents serving them) don't remember every dev3 feature or the intended way to use it. We wanted a skill, modeled on Matt Pocock's `ask-matt` router, that answers "how is X done in dev3?" by teaching the intended flow instead of guessing. It ships to every dev3 install like the other managed skills.