ai-shared

End-to-end local git and PR workflow: branching, committing, rebasing, conflict resolution, pushing, creating PRs, and processing review feedback. Use when creating a branch, writing commits, rebasing, resolving conflicts, opening a PR, or submitting changes for review. NOT FOR: GitHub API operations like listing issues, searching code, or reading PR reviews (github-mcp).

Write, rewrite, review, and improve Jira ticket content quality. USE FOR: crafting ticket titles, descriptions, acceptance criteria, and story structure. Use when user says 'write a Jira ticket', 'turn this into a ticket', 'review this ticket', 'is this ticket clear enough', or 'improve acceptance criteria'. NOT FOR: API operations like creating/searching/transitioning issues in Jira (atlassian-mcp). This skill handles content quality, not CRUD.

Web automation via Playwright MCP browser tools. USE FOR: filling forms, clicking buttons, navigating pages, interacting with web UIs, browser automation tasks. ALWAYS use when user asks to automate browser actions, fill out a form, click through a wizard, or interact with a website.

End-to-end local git and PR workflow: branching, committing, rebasing, conflict resolution, pushing, creating PRs, and processing review feedback. Use when creating a branch, writing commits, rebasing, resolving conflicts, opening a PR, or submitting changes for review. NOT FOR: GitHub API operations like listing issues, searching code, or reading PR reviews (github-mcp).

Manual QA execution workflow. USE FOR: creating and running QA checks from Jira tickets, PR diffs, local changes, preview URLs, or regression risk. Use when asked for manual QA, QA plan, QA checklist, test instructions, or functional/regression validation beyond automated tests.

Frontend UI validation methodology after implementation. USE FOR: verifying visual correctness, runtime health, design comparison, regression checks, and deciding pass/fail after code changes. ALWAYS use when a ticket has UI impact and needs browser-level validation before PR.

Decide how to package a unit of work into reviewable pull requests — one PR or a stack — and carry it out. USE FOR: should this be one PR or several, planning a ticket as multiple PRs, stacking PRs, splitting a PR into two, breaking up a large PR, extracting a refactor from a feature PR, making a big diff easier to review. Use when scoping a ticket's delivery, when a change grows too large, or when the user asks to split, stack, or break up a pull request.

Proactive code review using a repo-local review ledger, line-by-line changed-hunk inspection, file-type lenses, and a 4-layer heuristic: surface correctness, test coverage gaps, bounded refactors, and architecture attention signals. USE FOR: reviewing PRs, review this PR for missing tests, self-review before creating a PR, evaluating changed files. Use when user says "review this", "check this PR", "anything I missed", "review my changes", or asks for line-by-line code review.

Capture learnings and evolve skills from experience. USE FOR: end of complex tasks where something reusable was learned, when user says 'save this', 'remember this', or 'this should be a skill'. Use when a pattern repeats across sessions or a hard-won insight deserves persistence. NOT FOR: mid-task execution — finish the user's task first, then capture the learning.

Frontend UI validation methodology after implementation. USE FOR: verifying visual correctness, runtime health, design comparison, regression checks, and deciding pass/fail after code changes. ALWAYS use when a ticket has UI impact and needs browser-level validation before PR.

Build and review frontend UI with accessibility in mind. USE FOR: implementing components, forms, dialogs, menus, and flows that need good semantics, labels, keyboard support, focus management, error handling, and reduced-motion behavior. Use when implementing UI, running an a11y audit, checking keyboard/focus, or prioritizing accessibility issues. NOT FOR: backend code, API design, or non-visual logic.

Query and verify analytics data from Amplitude via the Amplitude MCP server. USE FOR: debugging tracking events, verifying analytics implementation, checking event payloads, reviewing user funnels. ALWAYS use when user asks about tracking data, analytics queries, or Amplitude events. NOT FOR: implementing tracking code in source files — that is regular coding with applying-coding-style.

Personal code writing standards for naming, comments, and cleanliness. USE FOR: writing new code, reviewing code style, refactoring for readability. ALWAYS apply when generating or editing TypeScript, Vue, or SCSS files. Use when user says "write clean code", "fix naming", "remove comments", or "code quality". NOT FOR: architecture decisions (reviewing-code layer 4), security patterns (security-hardening), or test strategy (test-driven-development).

Interact with Jira tickets and Confluence wiki pages via the Atlassian MCP server. USE FOR: reading/creating/updating Jira issues, searching Jira with JQL, reading/creating/updating Confluence pages, adding comments. ALWAYS use when user references a Jira ticket (e.g. DSC-1986, B2C-12345) or Confluence page.

Read content from Contentful CMS via the `contentful` CLI and CMA API. USE FOR: reading content types, entries, assets, spaces, and environments. ALWAYS use when user asks about Contentful content, product data, or CMS entries. READ-ONLY — never create, update, or delete Contentful data. NOT FOR: modifying CMS content or managing Contentful configuration.

Structured debugging workflow: reproduce, localize, reduce, fix, guard. USE FOR: test failures, build breaks, unexpected behavior, runtime errors. Use when quality gates fail, a bug is reported, or behavior does not match expectations. NOT FOR: writing new features test-first (test-driven-development), or reviewing code that already works (reviewing-code).

Write and maintain technical documentation: ADRs, READMEs, inline docs, and wiki pages. USE FOR: writing ADRs, updating READMEs, documenting architecture decisions, creating onboarding docs. Use when a decision needs recording, a module needs documenting, or docs are stale. NOT FOR: Jira ticket writing (jira-ticket), or code comments and naming conventions (applying-coding-style).

Read Figma designs and node context using Figma integrations before browser fallbacks. USE FOR: opening Figma design URLs, reading node context, extracting specs or copy from Figma, and checking whether a Figma link is readable. Use when the user shares a Figma URL, node-id, or asks to read a design.

Interact with GitHub via the `gh` CLI. USE FOR: reading/creating/updating pull requests, issues, branches, commits, reviews, code search, repository operations. ALWAYS use when user references a PR number, issue, or asks to perform GitHub API operations. NOT FOR: local git operations like committing, rebasing, or branch management (git-workflow).

Fetch Google Sheets, Docs, and other Drive content via CLI using existing OAuth credentials. USE FOR: reading spreadsheets, downloading CSV data, accessing Google Docs. ALWAYS use when user shares a Google Sheets/Docs link or asks to read spreadsheet data. READ-ONLY — does not modify Drive content.

Write, rewrite, review, and improve Jira ticket content quality. USE FOR: crafting ticket titles, descriptions, acceptance criteria, and story structure. Use when user says 'write a Jira ticket', 'turn this into a ticket', 'review this ticket', 'is this ticket clear enough', or 'improve acceptance criteria'. NOT FOR: API operations like creating/searching/transitioning issues in Jira (atlassian-mcp). This skill handles content quality, not CRUD.

Web automation via Playwright MCP browser tools. USE FOR: filling forms, clicking buttons, navigating pages, interacting with web UIs, browser automation tasks. ALWAYS use when user asks to automate browser actions, fill out a form, click through a wizard, or interact with a website.

Proactive code review using a repo-local review ledger, line-by-line changed-hunk inspection, file-type lenses, and a 4-layer heuristic: surface correctness, test coverage gaps, bounded refactors, and architecture attention signals. USE FOR: reviewing PRs, review this PR for missing tests, self-review before creating a PR, evaluating changed files. Use when user says "review this", "check this PR", "anything I missed", "review my changes", or asks for line-by-line code review.

Security review and hardening for frontend applications. OWASP Top 10 prevention, input validation, auth patterns, secrets management, dependency auditing. Use when handling user input, building auth flows, storing data, integrating with external services, or making security-sensitive changes. NOT FOR: general code style review (reviewing-code) — activate alongside reviewing-code when changes touch auth, input handling, or data boundaries.

Capture learnings and evolve skills from experience. USE FOR: end of complex tasks where something reusable was learned, when user says 'save this', 'remember this', or 'this should be a skill'. Use when a pattern repeats across sessions or a hard-won insight deserves persistence. NOT FOR: mid-task execution — finish the user's task first, then capture the learning.

Test-driven development: write failing test, make it pass, refactor. USE FOR: implementing new functions, fixing bugs with regression tests, adding behavior to existing code. Use when writing tests before code or when asked to TDD a feature. NOT FOR: debugging existing failures without writing new code (debugging), or reviewing test coverage in existing code (reviewing-code layer 2).

Manual QA execution workflow. USE FOR: creating and running QA checks from Jira tickets, PR diffs, local changes, preview URLs, or regression risk. Use when asked for manual QA, QA plan, QA checklist, test instructions, or functional/regression validation beyond automated tests.