Claude Skills by itsvedantkumar

Use when you need to screenshot or drive a dev server and the shared Chrome is unavailable or contended — e.g. parallel Conductor workspaces: headless per-workspace browser via npx agent-browser.

Use before shipping a risky change or reviewing a diff you don't trust: finds what it breaks elsewhere, beyond the diff, and proves the one safety fact by running real code instead of writing it up.

First, and only when the shape is undecided: explore intent and requirements for new capability before any plan or test. Skip it for a bugfix or a change whose shape is known.

Before writing a UI component in a React/Tailwind repo: pull a vetted accessible primitive from a shadcn-compatible registry (npx shadcn add). Hand-roll only what none has.

Nothing exists yet: this repo has no scripted proof its UI, CLI or service works. Generates the executable .claude/verify.sh gate, a verify-<app> skill, and a feature map.

A written plan or numbered task list already exists and the ask is to carry it out: work the tasks in order, verify each, pause at the review checkpoint between phases.

The user wants to DISCOVER or INSTALL a skill, not have the task done now: find a skill for X, is there a skill for X, what skills exist for X, install a skill. Searches the ecosystem.

Before the thing exists: stress-test a plan, design or decision the user is still forming, as a round-based interview. Fires on grill me, poke holes, or what am I missing.

While polishing UI that already renders, when visual quality is the bar: production-grade typography, motion, spacing, and brand-vs-product modes beyond the stock design level.

After the thing exists: parallel reviewers over a diff one person or nobody has read, especially auth, payments or agent-written code. Fires on tear this apart.

Something exists and has drifted: a repo's verify.sh gate, verification skill or feature map no longer matches the app. Audits every feature from source and live, ships proven fixes.

Apply when wiring input validation, error handling, or a framework/MCP/auth adapter. Concentrate guards at boundaries; trust internal types; keep business logic in pure functions.

Apply to non-trivial edits, migrations, analyses, or checks — not just bulk work. Build a codemod, script, generator, or skill instead of hand work or a one-off throwaway.

Apply when a correction, instruction, or fix repeats for the second time. Encode it as a lint, metadata flag, runtime check, or script — not more prose that will be missed again.

Apply when debugging or reaching for a try/except guard. Trace the symptom to its root cause; resist broad exception swallowing and nil-checks that silence crashes.

Apply when designing commands, cron/launchd jobs, or loops that may crash, restart, or retry mid-run. Converge to the same correct end state no matter where a prior run stopped.

Apply before declaring any task or fix done. A habit, not a generated gate: run the real artifact, read the actual value, inspect the diff. Never a self-report, proxy, or 'it compiles.'

Apply to sweeps, migrations, or stacked commits/PRs. Break work into small units that each end in a verified state before the next starts — never batch edits then verify once.

Apply when a struct, enum, or type can hold an invalid combination of fields that shouldn't compile. Redesign it so illegal states can't be represented — model the cases as separate variants.

Use after a session where you were corrected, hit dead ends before finding the path, or worked out a workflow worth keeping — mines the transcript and turns each lesson into a concrete skill edit.

Before unattended, overnight, or multi-phase work someone reviews after you step away, open an append-only TSV decision log: what, why, evidence, result, one row per decision.

Three or more independent units of work exist (files, packages, endpoints, rival approaches) and none needs another's result. Fans out N agents in ONE batched message, returns one table.

Before drafting a doc, RFC, README, PR description or commit message: pick the structure. Diataxis, Google developer style, STE instruction rules, Global English syntax.

About to write implementation code or a bugfix and no test fails for that behaviour yet: write the failing test first, watch it fail, then write the minimum that passes.

TypeScript idioms and type-safety patterns: discriminated unions, branded types, narrowing, boundary validation. Use when reading, writing, or reviewing any .ts or .tsx file.

After editing a UI file (.tsx/.jsx/.css/.vue/html) with a dev server running: screenshot 375/768/1440px, critique hierarchy, spacing and states, fix and re-shoot.

After prose is drafted, as the last pass over it: cut AI tells and restore human voice. Applies to every text you write, once there is text to edit.

Requirements are settled, nothing is written down about how, and the work is more than one edit: produce the plan document, file map, bite-sized tasks, one verification per task.