Agently

Agent packs for Claude Code

Stop collecting agents. Install the ones that score.

Six curated packs of agents, skills, and commands for Claude Code — assembled, model-routed, and measured on machine-checkable fixtures against the free baseline and against no agent at all.

Get the packs — $69 Read the methodology

6 packs · per-agent model routing · scored, not vibes

Why this exists

The content is a commodity. The confidence isn't.

More than 200 free agents for Claude Code already exist, and the biggest collection is excellent, MIT-licensed, and maintained. We are not going to out-write it, and we won't pretend to.

What nobody does is tell you which agents actually change outcomes — which ones beat the bare model, and which ones just add tokens. That's the gap. What you're buying here is assembly, taxonomy, and proof.

Curated

Six small packs, not one dump of 200. Each pack is a working set for one kind of job, small enough to read before you install it.

Routed

Every agent is pinned to haiku, sonnet, or opus — haiku for mechanical work, opus for architecture and debugging. Your token bill reflects it.

Scored

Every agent is measured on machine-checkable fixtures with two control arms — the free baseline and no agent at all. Numbers publish when the run completes.

Six packs

Small working sets, one job each.

  • eng-core

    Everyday engineering

    Review, debugging, refactoring, tests — the agents you reach for daily.

    agents · skills · commands · hooks
  • eng-quality

    Code health

    Linting flows, coverage, simplification — keeping a codebase honest.

    agents · skills · commands · hooks
  • eng-infra

    Infrastructure

    CI, deploys, docker, cloud config — the plumbing around the code.

    agents · skills · commands · hooks
  • eng-testing

    Test automation

    Test automation: browser, mobile, API, load, and accessibility — plus the flake hunting that keeps suites trustworthy.

    agents · skills · commands
  • marketing-core

    Positioning

    Positioning, landing copy, SEO audits — for engineers who also ship the site.

    agents · skills · commands · hooks
  • marketing-content

    Content

    Posts, docs, changelogs, launch notes — the writing that follows the release.

    agents · skills · commands · hooks
See every agent the full catalog — names, models, and dispatch rules 
agently/
├── plugins/
│   ├── eng-core/
│   │   ├── agents/        # model pinned per agent
│   │   ├── skills/
│   │   ├── commands/
│   │   └── hooks/
│   ├── eng-quality/
│   ├── eng-infra/
│   ├── eng-testing/
│   ├── marketing-core/
│   └── marketing-content/
├── templates/
│   └── CLAUDE.md          # onboarding scaffold
├── CREDITS.md             # upstream attribution
└── LICENSES/

Standard Claude Code layout, installs like any plugin.

Plain markdown in the standard plugin structure — nothing compiled, nothing proprietary. A CLAUDE.md scaffold gets a new repo onboarded, and CREDITS.md with a LICENSES/ directory keeps attribution for everything we built on.

Scored, not vibes

Every fixture runs three ways.

  1. 01 · arm one — our agent

    Our agent

    The pack agent runs the fixture with its pinned model, exactly as shipped.

  2. 02 · arm two — free baseline

    The free baseline

    The closest agent from the dominant free collection runs the same fixture.

  3. 03 · arm three — no agent

    No agent at all

    The bare model runs the fixture with no agent, so we know what agents add.

If the bare model matches the agent, we cut the agent.

Outcomes are machine-checkable wherever possible — a test passes, a bug is found at file:line, a build succeeds. Where they can't be, a rubric judges, and we disclose which is which for every fixture. The fixtures go public when the numbers do.

Full methodology

Pricing

One price, lifetime. Nothing free, nothing fake.

Pro

$69 one-time

first 100 buyers: $39 — real counter, not scarcity theater

Lifetime access to all six packs via private GitHub repo invite. 14-day refund, no questions.

See pricing

Current

$12/mo · optional

cancel anytime by email

New packs monthly, and the benchmark refreshed on every model release. Genuinely optional on top of Pro.

See pricing

Questions

The questions a skeptic would ask.

How is this different from the free 36k-star repo?

Honestly, the raw content is a commodity. wshobson/agents is free, MIT-licensed, and good, and we build on that ecosystem rather than pretend to out-write it. What you pay for is assembly: six small packs instead of one dump of 200, per-agent model routing, an onboarding scaffold, and a benchmark that scores every agent against that free baseline and against no agent at all.

What exactly do I get?

Six packs — eng-core, eng-quality, eng-infra, eng-testing, marketing-core, and marketing-content — in the standard Claude Code plugin layout, with per-agent model routing, a CLAUDE.md onboarding scaffold, and CREDITS.md plus LICENSES/ preserved from the sources we build on. Delivery is a private GitHub repo invite within 24 hours of purchase.

Where are the benchmark numbers?

The first public run is in progress. The methodology is live now, and the fixtures ship publicly alongside the results. Until the run completes we publish no scores, here or anywhere else.

Do you offer refunds?

Yes. 14 days on Pro, no questions asked. Email krishnacodes321@gmail.com and we refund the purchase.

Will it work with my stack?

The packs are plain markdown in the standard Claude Code plugin layout — no runtime, no lock-in. They work wherever Claude Code works, regardless of language or framework.

Is this affiliated with Anthropic?

No. Agently is an independent product, not affiliated with, endorsed by, or sponsored by Anthropic. Claude is a trademark of Anthropic, PBC. Read the full disclaimer .

The packs are assembled. The proof is being measured.

Get the packs

$69 lifetime · first 100 at $39 · 14-day refund