Compare
Specifically wshobson/agents — the collection everyone means when they say "just use the free one." They mostly have a point.
wshobson/agents is MIT-licensed, free, and huge: 203 agents, 175 skills, 109 commands, roughly 36.6k GitHub stars, and an active community filing issues and improvements. It is the dominant option in this category and it earned that position.
Here is the part a sales page is not supposed to say: if you enjoy curating — reading through 200 agents, picking the dozen that fit your stack, checking their model choices, and wiring in your own conventions — use it and pay nothing. That is the right call for a lot of engineers, and nothing we sell changes it. Agent prompt text is a commodity. Ours is not magically better.
The afternoon you would spend doing that curation, done carefully and kept current. Six small packs instead of a 200-item list. A CLAUDE.md scaffold so the packs work on day one. Per-agent model routing — haiku for mechanical work, sonnet for the bulk, opus for architecture and debugging — which is a real cost saving on real usage. And a filter the free collection has no reason to run: every agent is scored on public fixtures against the free baseline and against no agent at all, and anything the bare model matches gets cut. The methodology is on the benchmark page; no numbers exist yet and none appear here until the first public run completes.
| Agently | wshobson/agents | |
|---|---|---|
| Price | $69 one-time; $39 for the first 100 buyers | Free, MIT license |
| Content volume | 6 packs, kept deliberately small | 203 agents, 175 skills, 109 commands |
| Curation | Assembled per role; anything the bare model matches gets cut | You pick from the full list yourself |
| Quality signal | Scored on public fixtures against two controls; methodology published, first run in progress | Stars, issues, and community use |
| Model routing | Set per agent for cost: haiku for mechanical work, sonnet for the bulk, opus for architecture and debugging | Model choices vary by agent; you audit and tune them yourself |
| Onboarding scaffold | CLAUDE.md template wired to the packs | None; bring your own conventions |
| Updates | Lifetime repo access; optional $12/mo for monthly packs and benchmark refreshes on model releases | Community-driven, no schedule |
Counts and stars as of July 2026. Check the repository for current numbers.
Verdict
The free collection wins on volume, community, and price, and it always will — those are structural, not temporary. Agently wins only if the assembly is worth more to you than the time it replaces: a scoped set instead of a list to triage, routing you did not have to tune, a scaffold you did not have to write, and a published test for whether any of it beats the bare model. If $69 is more than that time is worth, take the free one and do not look back.