SkillProof tested Claude skills

Best Claude Skills for Coding (2026)

Most "best skills" lists are copied README summaries. This one isn't: we installed all 560 coding skills in our catalog, ran each against a baseline and scored four criteria. 306 passed cleanly, average score 8.8/10. The one to start with: Pinme Auth.

Ranked by SkillProof Score · how we score

  1. 1

    Pinme Auth

    TypeScript reference for PinMe Worker auth: create/verify/list Identity Platform users.

    From our test: An unguided client guessed a generic Bearer-token API with an invented pagination scheme; the skill reproduced the real API-key header, exact response envelope, and the 403-vs-401 distinction the API…

    Tested · Works ★ 10.0/10 by glitternetwork

  2. 2

    Code Health Check

    Read-only, sub-agent-forked code health scanner with a fixed severity-graded report format.

    From our test: Ran it against the bundled demo app (a deliberately buggy Express API shipped in the same repo) and it caught all 8 planted issues - the SQL injection, the eval() config parser, both hardcoded…

    Tested · Works ★ 10.0/10 by huangjia2019

  3. 3

    S&box

    Blocks Unity muscle memory when writing s&box (Source 2 / Facepunch) C# components, Razor UI, and networking.

    From our test: Asked for a pickup component in both modes: baseline reflexively wrote MonoBehaviour/transform.position/Destroy(gameObject) Unity code, the skill produced correct…

    Tested · Works ★ 10.0/10 by gavogavogavo

  4. 4

    Coding Agent Sessions

    Finds and reconstructs real coding-agent session transcripts across Claude, Codex, OpenCode, and more.

    From our test: Ran the bundled finder live against this machine's real ~/.claude/projects and ~/.codex/sessions stores with zero pip installs: it correctly listed actual Claude Code sessions (right cwd, timestamps…

    Tested · Works ★ 10.0/10 by code-yeongyu

  5. 5

    RouterOS App YAML

    Writes and validates MikroTik RouterOS /app YAML — a docker-compose lookalike with real differences

    From our test: Validated both outputs against the project's own published strict JSON Schema for real: the naive docker-compose-style baseline threw 2 hard schema errors (unexpected 'version', unexpected 'deploy')…

    Tested · Works ★ 10.0/10 by tikoci

  6. 6

    API Filter

    Current (4.4+) API Platform QueryParameter filter reference that correctly steers away from soon-removed classes

    From our test: Checked its two central claims against the live api-platform/core source on GitHub, not just the skill's word: SearchFilter.php line 136 and OrderFilter.php line 202 both carry the exact `@deprecated…

    Tested · Works ★ 10.0/10 by api-platform

  7. 7

    Better Route Audit Enrichment

    Configures Better Route's audit-log middleware so idempotency keys get hashed and logger failures never break the API response.

    From our test: Ran real request objects through the actual better-route PHP classes cloned from Lonsdale201/better-route: a naive baseline audit-logger logged the raw bearer token and raw Idempotency-Key and let a…

    Tested · Works ★ 10.0/10 by Lonsdale201

  8. 8

    Before You Refactor

    A pre-refactor checklist that catches untested behavior before you 'clean up' code.

    From our test: Its checklist caught a real regression a naive refactor introduced: an archived record silently 'claiming' its dedup id so a later active duplicate got dropped — the naive rewrite passed all 5…

    Tested · Works ★ 10.0/10 by oribarilan

  9. 9

    Pyxel Game Development

    Builds and verifies Pyxel games using real headless validate/run evidence, not vibes.

    From our test: Ran the real pyxel-mcp validate+run tools against two versions of a moving-square game: the naive baseline skipped pyxel.cls() and validate flagged it live, then a headless run proved the bug on…

    Tested · Works ★ 10.0/10 by kitao

  10. 10

    Compone

    Python component framework for type-safe HTML/XML/RSS generation.

    From our test: Ran a real A/B against the actual pip-installed compone 0.4.1: my unguided baseline guessed a dominate-style API and crashed immediately (TypeError: 'module' object is not callable), while the…

    Tested · Works ★ 10.0/10 by kissgyorgy

How this ranking works

Each skill gets a fresh install in Claude Code, three trigger phrasings, and the same task run with and without it. Scores: installation (5), trigger reliability (5), output vs baseline (10), documentation honesty (5) — normalized to a 10-point SkillProof Score. Full rubric on the methodology page. Queue-verdict skills sit at the bottom until testing completes.

New here? Start with how to install Claude skills — or grab the whole setup in one command with a curated pack ($10).

On a different AI tool? These tested instructions adapt: see our conversion guides for Cursor rules, Copilot instructions, Windsurf and Custom GPTs.