Best Claude Skills for Efficiency (2026)

Most "best skills" lists are copied README summaries. This one isn't: we installed all 80 efficiency skills in our catalog, ran each against a baseline and scored four criteria. 36 passed cleanly, average score 8.3/10. The one to start with: Kill Dev Process.

Ranked by SkillProof Score · how we score

  1. 1

    Kill Dev Process

    Classifies and safely kills orphaned dev servers/browsers with a hard exclusion list for databases and the IDE.

    From our test: Ran the actual step-1 investigation commands on a live dev machine: they surfaced postgres on :5432 and Claude's own IDE Helper node processes mixed in with real dev servers - exactly the kind of…

    Tested · Works ★ 9.6/10 by nokonoko1203

  2. 2

    Clean Cache

    Tiered, confirmation-gated cache cleanup for Flutter/Android/iOS/Node workspaces — verified not to touch lockfiles or source.

    From our test: Actually ran the bundled scan-and-delete bash commands against a throwaway workspace with a fake node_modules, android build/, and .dart_tool/: L1 cleanup wiped only the build artifacts and left…

    Tested · Works ★ 9.6/10 by addxai

  3. 3

    Agent Prompt Engineering

    A five-part checklist (role, paths, deliverables, constraints, comms) for writing sub-agent prompts that need zero follow-up.

    From our test: Drafted a spawn prompt for a 3-agent marketing-site redesign both ways: the baseline copy-agent prompt was one vague sentence ('improve the copy'), the skill-directed version had absolute file paths…

    Tested · Works ★ 9.6/10 by hesreallyhim

  4. 4

    Tidy Skill

    Read-only hygiene audits that stop agents littering repos with plan.md/todo.md junk.

    From our test: Ran score_repo_hygiene.py on a throwaway repo with plan.md/todo.md/summary.md in root — it returned a real 57/100 'Needs tidy-up' score with a dimension breakdown and named the 3 suspicious files…

    Tested · Works ★ 9.6/10 by Phoenix0531-sudo

  5. 5

    ACM Config

    Interactive Q&A that writes a validated ~/.claude/claudikins-acm.conf for the Claudikins context-handoff tool.

    From our test: Fully self-contained single-file skill; its real value is enforcing an exact serialization contract (file path, variable names, and a minutes-to-seconds conversion for SNOOZE_DURATION) that the…

    Tested · Works ★ 9.6/10 by elb-pr

  6. 6

    Perfetto Performance Analysis

    Perfetto trace analysis with a pinned trace_processor and 235 query graphs

    From our test: Bootstrapped its own Perfetto v57.2 trace_processor and verified the binary's SHA-256 against the bundled lock file, then ran startup_analysis end to end on a real 2 MB Android warm-start trace in…

    Tested · Works ★ 9.6/10 by Gracker

  7. 7

    SkillOpt-Sleep

    Offline nightly sleep-cycle that mines past sessions and gates CLAUDE.md/SKILL.md edits.

    From our test: Compared a naive direct-rewrite approach against the skill's stage-then-gate-then-adopt flow with backups; reading the actual gate code confirmed the harmful-edit block is implemented exactly as…

    Tested · Works ★ 9.2/10 by microsoft

  8. 8

    Golang Benchmark

    Statistically rigorous Go benchmarking, pprof profiling, and benchstat regression workflow

    From our test: Installed Go 1.26.5 and benchstat (installation command from Dependencies worked verbatim) and ran the documented workflow on a real module: naive concatenation vs. strings.Builder, ReportAllocs +…

    Tested · Works ★ 9.2/10 by samber

  9. 9

    Checkpoint

    Writes a structured, resumable session-handoff snapshot with file:line pointers and a resume prompt.

    From our test: Fetched the repo tree via gh API and read the full SKILL.md at .claude/skills/checkpoint/SKILL.md; verified all 6 cross-referenced files (rules/session-logging.md, plan-first-workflow.md…

    Tested · Works ★ 9.2/10 by pedrohcgs

  10. 10

    Mcp Client

    CLI to list and call tools on any MCP server without loading all schemas into context

    From our test: Cloned into a temp HOME, installed mcp+fastmcp in a venv, and ran the client against a real stdio server (@modelcontextprotocol/server-sequential-thinking). All three subcommands worked: servers…

    Tested · Works ★ 9.2/10 by coleam00

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.