stealth-engine
OrganizationAgent skills we use daily in the studio — reusable, on-demand context for Claude Code, Cursor & other skills.sh-compatible AI agents. Install: npx skills add stealth-engine/skills
Categories
Indexed Skills (17)
agents-md-setup
Set up a project's agent instructions as a single source of truth — AGENTS.md as the real file, with CLAUDE.md (and optionally other agents' files) as symlinks to it. Use when starting/bootstrapping a repo, when asked to "add a CLAUDE.md / AGENTS.md", "set up project conventions / agent instructions", "make Claude use AGENTS.md", or when a repo has a standalone CLAUDE.md that should become cross-tool. Also use to fix duplicated/drifting CLAUDE.md + AGENTS.md, or to add per-workspace instructions in a monorepo (Turborepo / pnpm workspaces).
autonomous-pr-driver
Autonomously drive a pull request to merge-ready — opening or attaching to it, then resolving automated code review (triage findings, fix the valid, reject the invalid, push, loop) and pinging a human to merge. Knows when to STOP: at diminishing returns (niche/trivial/contradictory findings, severity trending down, or review budget accruing) it declares the PR good-to-merge on substance and pauses rather than auto-looping to chase a bot to zero comments — resuming only if the user insists or a genuinely important finding appears. Use when asked to 'drive / ship / land this PR', 'get the PR green', 'resolve the PR review comments', 'address the CodeRabbit / Cursor / Bugbot / Codex findings', 'fix the code review and push', 'stop over-fixing / merge it', or to loop on PR reviews until checks pass.
code-complexity-stats-pr
Add a GitHub Actions workflow that posts code-size and cost statistics — lines of code, per-language breakdown, complexity, and a COCOMO "what would this cost to build" estimate — as a sticky comment on every pull request, using scc. Use when you want a per-PR code-stats or "worth of the codebase" comment, to show LOC / language breakdown or an estimated development cost/effort on PRs, to set up scc (Sloc Cloc and Code) in CI, to post or upsert a single self-updating bot comment from GitHub Actions, or to change the COCOMO salary/currency. Covers fork-PR token limits, runner architecture, version pinning, and comment pagination.
conventional-commits
The Conventional Commits format — also called "semantic commits" / semantic commit messages — a `type(scope): description` header (`feat`, `fix`, …) plus an optional `BREAKING CHANGE` footer, and how it makes history machine-readable to drive automated version bumps and changelogs. Use when writing a commit message or PR title, deciding the type/scope/bump for a change, setting up or fixing a repo's commit convention, making commits parseable by semantic-release / changelog tooling, validating PR titles, or when a release didn't bump or the changelog came out blank because a commit wasn't conventional. Covers the type→semver mapping, breaking changes, scopes, monorepo scopes, and squash-merge PR titles.
dual-agent-review
Review a change set with two or more independent AI reviewers in parallel — a Sonnet subagent and the CodeRabbit CLI — then merge + dedupe their findings, triage valid vs invalid, fix (directly or via parallel fix-subagents), and hand back a report of what was fixed/how and what was ignored/why. Use when asked to 'review my changes with two/multiple agents', 'dual-agent / multi-reviewer review', 'get a second opinion from Sonnet and CodeRabbit', 'review the diff and fix the valid issues', 'run parallel reviewers and fix in parallel', or to cross-check one reviewer against another before committing. Diff-scoped by default (branch-vs-base or the PR diff), like CodeRabbit — NOT a whole-repo audit.
fable-style-reasoning
Reasoning and communication discipline modelled on Claude Fable 5's default behaviours, for Claude Opus 4.8 (or any Claude 4.x model): calibrate effort, verify before asserting, separate what you know from what you infer, and report plainly. Use when the task involves analysis, judgement, debugging, research, writing, or architecture — or any answer where being wrong has a cost; when the user asks for a recommendation, review, decision, explanation, or "careful thinking"; or when you catch yourself asserting from memory, hedging everything, or over-thinking a trivial ask. Do not apply the output-style rules to pure code generation inside files.
git-trunk-branch-and-pr-automation
Trunk-based Git workflow with enforced branch naming and squash-merge PR titles. Use when setting up or standardising a branch/PR workflow, naming branches (feature/ fix/ hotfix/ and AI-agent prefixes claude/ cursor/ codex/ copilot/ codegen-bot/ dependabot/), configuring squash-only merges where the PR title becomes the commit and the body is the concatenated commits, making the PR title a valid Conventional Commit, adding GitHub Actions that validate branch names or auto-normalise PR titles, fixing a PR-title bot that loops, or deciding trunk vs release branches. Covers the GitHub repo settings, the validation/normalisation workflows, and how it feeds semantic-release.
meaningful-linear-issues
Create complete, analytics-ready Linear issues in ONE pass — resolve the right project (never guess it), and proactively fill the metadata that's technically optional but matters for planning/analytics: priority, estimate, labels (a Type + an Area from the team's existing taxonomy), milestone, assignee, cycle, and relations. Use when asked to 'create/file/open a Linear issue', 'log a ticket', 'add this to Linear', 'make an issue for X', or before any Linear `save_issue` that **creates** an issue or **backfills** missing metadata on a bare one — so the user doesn't have to come back and ask for the labels/estimate/project a second time, and so an issue doesn't land in the wrong project. NOT for routine field edits (status changes, reassignments) — those are also `save_issue` but don't need this.
multi-agent-codebase-audit
Audit an ENTIRE codebase with multiple agents in parallel — map the repo, partition it into review slices, fan out one (or more, multi-lens) reviewer subagent per slice, reduce with a cross-cutting/architecture critic + a completeness check, then triage, fix, and report with an honest coverage ledger. Use when asked to 'audit the whole codebase', 'full security/quality review of the repo', 'review the entire project', 'do a deep/comprehensive code audit', 'scan everything for bugs or vulnerabilities', when onboarding/inheriting an unfamiliar repo, or for a periodic deep sweep. Whole-repo scoped and scales agent count to repo size — NOT a diff review (for changed lines use dual-agent-review).
nextjs-locale-monorepo
Add locale-prefixed i18n routing across a Next.js monorepo (Turborepo / pnpm workspaces) by extracting the engine into a shared workspace package that every app consumes — locale detection + a middleware/proxy factory + a `use client` LocaleProvider/hooks, plus a shared supportedLanguages config. Each app redirects `/` and unprefixed paths to `/<locale>/…` from the toggle's last choice (the NEXT_LOCALE cookie) then the browser's Accept-Language, wiring only a thin middleware + `[locale]` layout. Use when several apps in one repo need consistent `/en-hk/…` `/zh-hk/…` routing, sharing locale logic via a workspace package instead of copy-paste, dual ESM/CJS build of a package that ships a client provider, Turbo build-ordering so the lib builds before the apps, or per-app locale wiring. For a single standalone site, use nextjs-locale-standalone instead.
nextjs-locale-standalone
Add locale-prefixed i18n routing to a single (non-monorepo) Next.js App Router site — a middleware/proxy that redirects `/` and any unprefixed path to `/<locale>/…` using the locale toggle's last choice (the NEXT_LOCALE cookie) then the browser's Accept-Language, a `[locale]` layout with a LocaleProvider + hooks, and a LocaleToggle that persists the choice. Use when adding bilingual/multilingual routing to a standalone Next.js site, building `/en-hk/…` `/zh-hk/…` URL namespaces, redirecting the root to a default-or-remembered locale, persisting a language switch across visits, or detecting browser language in middleware. For a monorepo that shares this logic across several apps via a workspace package, use nextjs-locale-monorepo instead.
pooled-release
Cut fewer, batched releases (a 'release train') instead of one per merge — by triggering semantic-release on demand or on a cadence rather than on every push to main. Use when releasing on every merge is too noisy, when you want one larger readable changelog per release, to add a manual 'cut a release' button (workflow_dispatch) or a scheduled/weekly release, to set up prerelease channels (next/beta → promote to stable), or when deciding whether you need a release branch.
resolve-merge-conflicts
Resolve git merge/rebase conflicts non-destructively — preserving the intent of BOTH sides — and hand off to a human when a conflict can't be resolved safely. Use when a branch or PR has conflicts with its base ('this branch has conflicts that must be resolved', GitHub mergeable=CONFLICTING / mergeStateStatus=DIRTY), when a rebase / merge / cherry-pick / stash / revert stops with conflict markers (<<<<<<< ======= >>>>>>>), when a PR is behind base and needs updating, when syncing a long-lived branch or back-merging a hotfix, or when deciding rebase vs merge. Covers the safe-resolution rules (no blind -X ours/theirs, hunk-by-hunk with diff3, abort/reflog escape hatches, verify-after), the rebase ours/theirs inversion, lockfile/rename/semantic conflicts, and the escalate-when-stuck criteria.
safari26-liquid-glass
How iOS 26 / iPadOS 26 Safari's "Liquid Glass" translucent status & address bars interact with web content, the viewport/keyboard facts behind them, and what to watch for when a design gets creative (immersive/edge-to-edge layouts, custom drawers/modals, gesture panning, themeable backgrounds, canvas). Use when an iPhone/iPad web page shows black/grey bars, content "cut off" at the bar edge, cropped shadows, the top status bar turning opaque/tinted under a sticky or fixed header, a drawer/modal that breaks the layout, an inner scroll that won't scroll, the page jumping after the soft keyboard closes, or a canvas blur that won't render on iOS — or before building any full-screen/immersive iOS web UI.
semantic-release-automation
Automate versioning, changelog, tags, GitHub Releases, and npm publishing from Conventional Commits with semantic-release. Use when setting up or debugging automated releases, wiring a `.releaserc` / `release` config and the plugin pipeline (commit-analyzer, release-notes-generator, changelog, npm, git, github), making `main` cut a version on merge, generating CHANGELOG.md, publishing to npm or creating a GitHub Release per release, doing per-package releases in a monorepo (per-package tags + paths-filter matrix), pooling commits into a less-frequent release, or fixing a release that didn't fire / a CI loop from the release commit. Covers the single-package and monorepo flavors and the GitHub Actions workflow.
skill-publishing
How to author and publish agent skills for the skills.sh / `npx skills` ecosystem — the SKILL.md format, why the frontmatter description is the trigger, multi-skill repo layout, the `skills` CLI (init/add/list/find/update/remove with flags), publishing to GitHub with SEO metadata, and public vs private repos (it git-clones, so private works with auth). Use when asked to "create/write/author a skill", "publish a skill", "set up a skills repo", "make a skill installable/discoverable", "use npx skills", or "can I use a private skills repo".
vercel-deployment-password-gate
A free DIY reimplementation of Vercel's $150/mo Advanced Deployment Protection add-on — all three features (Password Protection, private/prod deployments, Deployment Protection Exceptions) plus named automation bypass tokens — a middleware gate for ANY framework on Vercel (Next.js proxy, or SvelteKit/Nuxt/Astro/Remix/static via Routing Middleware). Gates previews by default, production opt-in; fully branded unlock page; zero prod cost. Use when asked to password-protect or basic-auth a preview/staging URL, avoid or cancel that add-on, password-protect on a Hobby plan, brand/white-label a password wall, add a login that "shows once and stays unlocked", set or ROTATE the password, add/remove bypass tokens for CI or automation (Lighthouse, uptime), make one preview domain public, protect previews on an app with NO existing middleware, gate a production or pre-launch site with a shared password (coming-soon, client demo, internal tool), or choose between a DIY gate and Vercel Authentication (free team SSO).
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.