A collection of agent skills maintained by @mizchi, distributed via APM (Agent Package Manager).
Each directory is a standalone skill following the agentskills.io open standard.
Install an individual skill (global / user scope):
apm install -g mizchi/skills/<skill-name>Or add to a project's apm.yml:
dependencies:
apm:
- mizchi/skills/<skill-name>Pin to a tag:
apm install -g mizchi/skills/<skill-name>#v0.1.0These follow the open Agent Skills format, so any
compatible agent (Claude Code, Codex, opencode, Gemini CLI, Cursor, …) can use
them without APM. Each skill's README shows a npx skills add command (the
skills CLI, installs into every
detected agent) and a raw npx degit folder copy as alternatives:
npx skills add mizchi/skills --skill <skill-name>Every skill directory carries its own README.md with copy-paste install
examples, generated from the skill's SKILL.md frontmatter. Regenerate them all
after editing any frontmatter:
pkf run gen:readme # write / sync every skill README
pkf run check:readme # CI: verify they are in syncA toolkit for frontend consulting engagements. Install skills from mizchi/skills/<skill-name>.
| Skill | Install path | Description |
|---|---|---|
| frontend-review-triage | frontend-review-triage |
Day 0 initial assessment — scorecard, top 3 risks, app classification. |
| frontend-review-ci | frontend-review-ci |
CI timing analysis — bottleneck identification, sharding/cache/concurrency recommendations. |
| frontend-review-hygiene | frontend-review-hygiene |
Code quality — TypeScript strictness, lint, dead code, duplication. |
| frontend-review-deps | frontend-review-deps |
Dependency health — freshness, CVE triage with attack-vector weighting, Tier 1/2/3 library detection. |
| frontend-review-testing | frontend-review-testing |
Testing posture — vitest coverage, Playwright config, VRT setup. |
| frontend-review-security | frontend-review-security |
Security review — HTML sinks, auth/token storage, route guards, env var exposure, AI self-pentest. |
| frontend-review-state | frontend-review-state |
State management architecture — server/URL/form/UI classification, Jotai/Zustand/Redux anti-patterns. |
| frontend-review-performance | frontend-review-performance |
Rendering performance — profiler-first, memo correctness, virtual scroll, useTransition. |
| frontend-review-weekly | frontend-review-weekly |
Weekly orchestrator — runs all domain skills, diffs KPIs, files repeat-finding issues. |
Perspective sub-skills (invoked by frontend-review-weekly): frontend-expert, react-expert, performance-expert, security-expert, frontend-ops-expert.
| Skill | Install path | Description |
|---|---|---|
| node-sqlite-vec | node-sqlite-vec |
Combine Node 24+ node:sqlite with the sqlite-vec extension — extension load, vec0 BigInt rowids, why vitest fails, CLI shebang. |
| pi-coding-agent | pi-coding-agent |
Embed @mariozechner/pi-coding-agent as a coding-agent runtime in Node scripts, write pi extensions (plugins) with pi.registerTool / pi.on, package and pi install from npm/git, SDK vs pi --mode rpc. |
| Skill | Install path | Description |
|---|---|---|
| aws-ecs-codedeploy-blue-green | aws-ecs-codedeploy-blue-green |
ECS blue/green — recommends ALB-native weighted routing; covers CodeDeploy for teams already using it. |
| aws-ecs-service-connect-ipv6 | aws-ecs-service-connect-ipv6 |
Diagnose and work around ECS Service Connect DNS aliases returning IPv6 addresses to IPv4-only Fargate tasks. |
| aws-github-oidc-scoped-role | aws-github-oidc-scoped-role |
GitHub Actions OIDC trust pattern — job_workflow_ref scoping vs sub alone, Bedrock cross-region inference ARNs, required aws-marketplace permissions, ReadOnlyAccess + explicit Deny for AI agent roles. |
| aws-vault-mfa-iam | aws-vault-mfa-iam |
Configure aws-vault for IAM APIs blocked by MFA-required policies; FIDO2 passkey + virtual TOTP setup. |
| Skill | Install path | Description |
|---|---|---|
| void-app | void-app |
Build full-stack apps with Void and Vite+ — React Pages, typed loaders/actions/forms, D1 migrations, tests, and deployment. 日本語チートシート |
| cloudflare-deploy | cloudflare-deploy |
Deploy Workers and platform resources through cf and typed configuration. |
| cloudflare-access-app-setup | cloudflare-access-app-setup |
One-shot Cloudflare Access self-hosted application provisioning via the API — app + email allowlist policy + service token. |
| cloudflare-workers-otel-utels | cloudflare-workers-otel-utels |
Cloudflare Worker telemetry — OTLP traces / metrics / logs + utels error tracking + D1 Proxy slow-query warnings. |
| cloudflare-mbt-worker-bundle | cloudflare-mbt-worker-bundle |
Bundle a Cloudflare Worker that combines MoonBit core code with a TypeScript entry. |
| Skill | Install path | Description |
|---|---|---|
| sql-lint | sql-lint |
Static lint for sqlc-style SQL catalogs — duplicate query names, missing semicolons, SELECT *, double-wildcard LIKE. |
| sql-plan-audit | sql-plan-audit |
Run EXPLAIN QUERY PLAN against every query in a sqlc catalog and diff plans against baseline. |
| sql-schema-audit | sql-schema-audit |
Index coverage and N+1 review aids for SQLite/D1 schemas with a sqlc catalog. |
| sql-security | sql-security |
SQL injection screening for host code (MoonBit / TS / Rust) plus secretlint setup notes. |
| sqlc-gen-moonbit-safety | sqlc-gen-moonbit-safety |
Post-generation safety checks for sqlc-gen-moonbit + Cloudflare D1 — BigInt-bind hangs (D1 1101) and SQL placeholder mix. |
| Skill | Install path | Description |
|---|---|---|
| moonbit-practice | moonbit-practice |
MoonBit code generation best practices — syntax, tests, benchmarks, FFI, Nix, mbtx. |
| moonbit-js-binding | moonbit-js-binding |
Write MoonBit bindings to JavaScript with extern "js" (Promises, opaque types, esm/cjs/iife). |
| gleam-practice | gleam-practice |
Build and review Gleam projects on the Erlang target (Wisp + Mist, OTP, just, CI). |
| translate-programming-language | translate-programming-language |
Plan and execute language-to-language migrations with behavior parity — oracles, fixtures, parity tests, compatibility layers. |
| ts2moonbit-migration | ts2moonbit-migration |
Migrate TypeScript to MoonBit on the js target — JS-compatible exported boundary over an idiomatic MoonBit core, using mizchi/js, mizchi/x, mizchi/npm_typed; contract capture, type classification, .d.ts diff gate, parity verification. |
| Skill | Install path | Description |
|---|---|---|
| metal-practice | metal-practice |
Apple Metal rendering and compute — MetalKit, shader ABI, resource synchronization, Metal 4, GPU validation and profiling. |
| Skill | Install path | Description |
|---|---|---|
| playwright-cli | playwright-cli |
Run Playwright via terminal CLI (test runner, codegen, screenshot, CI sharding). |
| stryker-js | stryker-js |
Measure JavaScript/TypeScript test quality with StrykerJS; run Git diff mutations in CI and use surviving mutants to improve assertions locally. Includes setup snippets and AI-readable reports. |
| playwright-test | playwright-test |
Playwright Test (E2E) best practices — no fixed waits, network triggers, DnD, CI sharding/retry. |
| Skill | Install path | Description |
|---|---|---|
| review-image | review-image |
Review screenshots or generated images with OpenRouter vision models via Deno scripts; includes a strict CI `pass |
| vlmkit | vlmkit |
Entry-point for @mizchi/vlmkit — VLM-driven frontend toolkit (visual regression, markup synthesis from screenshots, design-token / theme / a11y / i18n audits, 2-stage VLM+LLM CSS auto-repair). Routes to 5 detailed sub-skills shipped under the vlmkit repo's .claude/skills/. |
| Skill | Install path | Description |
|---|---|---|
| formal-methods-reconciler | formal-methods-reconciler |
Reconcile specs, docs, code, tests, config, logs, and incidents with formal-methods checks; choose Z3 / Alloy / TLA+ / P / Dafny / MoonBit prove / Lean / Rocq and translate counterexamples into domain questions. |
| formal-methods-drift-guard | formal-methods-drift-guard |
Maintain existing formal models by checking spec/code/model drift, classifying stale claims, and translating verifier changes into domain questions. |
| Skill | Install path | Description |
|---|---|---|
| apm-usage | apm-usage |
Use APM (Agent Package Manager) to manage agent skills and dependencies. |
| ast-grep-practice | ast-grep-practice |
Run ast-grep as a project lint tool — sgconfig.yml, fix/rewrite rules, constraints, CI. |
| chezmoi-management | chezmoi-management |
mizchi's personal chezmoi dotfiles workflow (diff/apply, skill placement, APM vs chezmoi boundary). |
| conventional-changelog | conventional-changelog |
Conventional Commits + CHANGELOG generator comparison (release-please, changesets, git-cliff, towncrier). |
| dotenvx | dotenvx |
dotenvx reference for existing adoption or explicit user choice (encryption, multi-env, CI). |
| justfile | justfile |
just command runner reference with GitHub Actions examples. |
| nix-setup | nix-setup |
Requested or existing Nix environments — flakes, per-language templates, direnv, CI; devbox only when already adopted or explicitly selected. |
| upstream-fix-and-pin | upstream-fix-and-pin |
Upstream PR + temporary git-ref pin workflow — branch HEAD SHA → merge SHA, link: fallback, pnpm v10 build script gating. |
| maintainer-persona | maintainer-persona |
Measure a target repo's maintainers into an editable persona file, branch by OSS first-time / returning / internal team, draft in Japanese then translate and slop-check against their own PRs, gate the send. |
| utels-project-bootstrap | utels-project-bootstrap |
One-shot helper for registering a utels.dev project and writing the returned ingest token into a Worker secret through cf. |
| Skill | Install path | Description |
|---|---|---|
| k8s-crd-from-typed-schema | k8s-crd-from-typed-schema |
Generate Kubernetes CRDs from a typed schema source (zod / TypeBox / Valibot / json-schema) — Structural Schema dialect, /status subresource trap, metadata-prohibition rule. |
| Skill | Install path | Description |
|---|---|---|
| opentelemetry | opentelemetry |
Platform-agnostic OTel reference — signal selection, span design, context propagation, sampling, OTLP exporter config. |
| otel-node | otel-node |
Node.js OTel setup — SDK init, auto-instrumentation, and the esbuild ESM silent-failure gotcha. |
| gh-fix-ci | gh-fix-ci |
Debug and fix failing GitHub Actions PR checks via gh. |
| flaker-storage-cache-on-ci | flaker-storage-cache-on-ci |
Persist flaker's DuckDB storage across GitHub Actions via actions/cache@v6 with sliding key. |
| cloudflare-workers-cd-rollback | cloudflare-workers-cd-rollback |
GitHub Actions CD for a Cloudflare Worker with auto-rollback on smoke failure. |
| Skill | Install path | Description |
|---|---|---|
| publish-agent-skill | publish-agent-skill |
Publish and update portable skills through Claude Code marketplaces, APM and the skills CLI; verify discovery, installed resources and consumer updates. |
| multi-agent-orchestration | multi-agent-orchestration |
Decide whether to spawn agents and which topology to use — single-agent default, independently verifiable work, DAG + verifier. |
| empirical-prompt-tuning | empirical-prompt-tuning |
Iteratively evaluate and improve agent-facing text instructions using unbiased subagent executors. |
| retrospective-codify | retrospective-codify |
Convert trial-and-error lessons into ast-grep rules / skills / CLAUDE.md rules. |
| skill-selector | skill-selector |
Decide which skills to add to a project — Phase 1 picks from a curated catalog (Phase 2 escalates to skill-finder). |
| skill-finder | skill-finder |
Cross-source skill discovery (Anthropic official → claude-skill-registry → VoltAgent → ComposioHQ → Superpowers → GitHub topic) with a mandatory waxa eval gate before adoption. |
| waxa-eval | waxa-eval |
Operating manual for the waxa CLI — scenario authoring, grader selection, ledger schema, the four-stage iteration pattern, convergence rules. |
| extract-glossary | extract-glossary |
Extract domain-specific terms, repository implementation maps, and onboarding Mermaid diagrams from one or more repos / GitHub orgs. |
| optimizing-descriptions | optimizing-descriptions |
Audit and rewrite SKILL.md description fields per the agentskills.io optimizing-descriptions framework. |
| mizchi-blog-style | mizchi-blog-style |
Style guide and AI-smell detection for blog posts published as mizchi (zenn / dev.to). |
| tech-article-reproducibility | tech-article-reproducibility |
Simulate a first-time reader reproducing a technical article and list missing information. |
The 2026-11-16 skill freshness audit
inventories all 76 maintained entrypoints, separates confirmed stale guidance
from version-specific examples needing revalidation, and prioritizes update
detection. The upstream watchlist records
sources and validation gates for fifteen groups affecting 38 skills. The daily freshness check fetches current upstream evidence,
compares declared topics with jev-lint and fails on unreviewed upstream drift.
It reports candidates without rewriting skills or accepting a baseline automatically.
Run just --justfile skill-freshness.justfile check.
PR semantic review checks changed entrypoints and their
English/Japanese pairs using trusted tooling. Maintained Node examples use Node 24+.
pnpm links:install
just --justfile markdown-links.justfile check
just --justfile markdown-links.justfile onlinePinned lychee checks local files and heading fragments on every Markdown PR. External URL status checks run manually and weekly. See scope, reports, and configuration.
The repository uses pinned jev-lint tooling and the upstream jev-lint skill.
Custom rules check description/body agreement, conflicting instructions and
terminology across all maintained skill entrypoints. English/Japanese pairs also
receive explicit translation and activation-condition comparisons.
pnpm install --frozen-lockfile
apm install --frozen
just --justfile skill-lint.justfile plan
just --justfile skill-lint.justfile check
just --justfile skill-lint.justfile translationsLive checks read TYPESAFE_API_KEY from the environment. CI tests source selection
and replays calibrated rule fixtures without making API requests. See the
configuration and workflow and initial audit.
Lives at tools/waxa/ and is published as @mizchi/waxa. Used by the waxa-eval skill, and forms the adoption gate of skill-finder.
Run via npx (no global install needed):
npx @mizchi/waxa <eval.yaml>npx caches the package after the first run, so subsequent invocations are fast. Pin a version with npx @mizchi/waxa@0.1.1 when reproducibility matters (do not pin 0.1.0 — that release shipped a broken shebang and crashes before argv parsing).
Alternatives:
# install globally (when calling waxa frequently)
npm i -g @mizchi/waxa
# run from source (requires Deno 2+)
git clone https://github.com/mizchi/skills.git
cd skills/tools/waxa
deno task run -- path/to/eval.yamlRequirements: claude CLI on PATH and authenticated (OAuth login or ANTHROPIC_API_KEY).
Quick reference:
npx @mizchi/waxa init [--skill <name>] [--force] # scaffold <skill>/evals/ (skill-local)
npx @mizchi/waxa <skill>/evals/eval.yaml # single run
npx @mizchi/waxa <skill>/evals/eval.yaml --baseline # with_skill vs without_skill + delta
npx @mizchi/waxa iterate <skill>/evals/eval.yaml [--max N] # iteration loop with ledger
npx @mizchi/waxa compare <skill>/evals/eval.yaml --models <m1>,<m2> # multi-model comparison
npx @mizchi/waxa variant <skill>/evals/eval.yaml --base <skill-a> --candidate <skill-b> # A/BFrom 0.2.0 evals live at <skill>/evals/ (skill-local layout, agentskills.io-aligned). Workspace outputs go to <workspace-root>/results/<skill>/iteration-N/. The npm package bundles references/empirical-prompt-tuning.md (the methodology document), so iter / convergence semantics ship with the CLI.
Full reference: tools/waxa/README.md.
If you are setting up a new project that will produce or consume skills, this combination covers selection, discovery, and iteration:
# apm.yml
targets:
- claude
dependencies:
apm:
- mizchi/skills/apm-usage # APM manifest reference
- mizchi/skills/skill-selector # Phase 1: pick from curated catalog
- mizchi/skills/skill-finder # Phase 2: cross-source discovery + waxa eval gate
- mizchi/skills/waxa-eval # waxa CLI operating manual
- mizchi/skills/empirical-prompt-tuning # methodology / Iter 0 / [critical]-tag checklistThen:
apm install # installs the skills under .claude/skills/
npx @mizchi/waxa --help # cache the waxa CLI on first call
# (or `npm i -g @mizchi/waxa` if you call waxa often enough that startup matters)The skill-selection / discovery / evaluation flow then proceeds as: skill-selector (catalog pre-flight) → if no fit, skill-finder (Tier 1-4 sweep) → adoption gated by waxa-eval running waxa iterate against a evals/<skill>/ directory you author. empirical-prompt-tuning covers what waxa cannot reach (the Iter 0 description / body consistency check, tool_uses-based skill self-containment diagnosis, and the [critical]-tagged requirements checklist that gives binary success judgment).
All SKILL.md files are written in English. Skills that were originally written in Japanese preserve the original as SKILL-ja.md alongside the English version. mizchi-blog-style is the single exception and stays Japanese-only because the skill itself is about Japanese blog writing style.
These are maintained in their own repositories and installed separately:
- mizchi/flaker —
flaker-setup - mizchi/tui.mbt —
tuimbt-practice - moonbitlang/moonbit-agent-guide —
moonbit-agent-guide,moonbit-refactoring,moonbit-c-binding - ast-grep/agent-skill —
ast-grep
Each skill may carry its own license (see LICENSE.txt inside the skill directory). Skills without an explicit license default to MIT at the repository owner's discretion.