Hermes Curator Stops Auto-Consolidating Skills by Default
The Hermes Agent Curator will no longer automatically consolidate skills. Teknium changed the default behavior to prune-only on June 17, following reports that the consolidation pipeline was consuming tokens at scale on frontier models.
The change drew 94 likes and 12 replies in under two hours. Several users reported similar token burn - one noted their consolidation runs were hitting $3-4 per session on Claude Opus.
What Changed
Before: the Curator ran both prune (remove unused skills) and consolidate (merge similar skills) on every scheduled pass. Consolidation requires the model to read multiple skill files, compare their content, and rewrite the merged result - a multi-turn chain that grows linearly with skill count.
After: prune is still on by default. Consolidation is now opt-in via config.yaml or the dashboard.
The practical impact:
| Metric | Before | After |
|---|---|---|
| Weekly token cost (frontier models) | ~$16 | ~$2-4 |
| Curator passes per week | Varies | Same schedule |
| LLM calls per pass | O(n skills) | O(1) |
| Consolidation | On by default | Opt-in only |
Consolidation still exists for users who want it. The flag is curator.consolidate.enabled: true in ~/.hermes/config.yaml. The dashboard also exposes a toggle. The feature moved from default to opt-in.
Capafy Adds Hermes Support
The same day, Capafy announced Hermes Agent support for its skill marketplace platform.
The announcement drew 40 retweets and 51 likes. The model is straightforward: developers publish Hermes skills, keep them closed-source, and earn per invocation. Capafy handles hosting, billing, and distribution.
Community reaction focused on the pay-per-use model. @bigaiguy noted: "Honestly I've been looking for a way to get paid for my Hermes skills. And not just give em away for free lol." @thisdudelikesAI called it "a good business model." @VK_ROXy asked about the revenue split with skill creators, though Capafy has not disclosed specifics.
This is the first skill marketplace to explicitly support Hermes Agent. Prior ecosystem monetization was indirect - consulting, custom builds, or platform integrations. A direct pay-per-use marketplace changes the incentive structure for skill authors.
Teknium Recommends Hermes for Remote Claude Code
When Matt Shumer asked how to control Claude Code from a phone (his tweet drew 592 likes and 295 replies), Teknium's answer was direct:
The workflow: Hermes Agent acts as a bridge between mobile messaging (Telegram, Discord, or any chat interface) and Claude Code on the desktop. The /claude-code slash command routes prompts through Hermes to Claude Code with the host machine's full permissions.
This is the same pattern Hermes uses for all its delegate tools - /github-pr-workflow, /codex, /delegate-task - but applied to Claude Code specifically. The mobile interface handles auth and session management; the desktop agent executes.
Why It Matters
The curator change and the marketplace launch address opposite ends of the same problem: who pays for agent compute, and how much.
The curator default change cuts cost for everyone running Hermes on frontier models. The opt-in model preserves the feature for users who actively want consolidation - usually those maintaining 30+ skills where manual deduplication is impractical.
The Capafy integration creates a path for skill authors to offset their own compute costs. If you build a skill that others find useful, you can publish it and earn per invocation. The platform model outsources hosting and billing so authors focus on the skill itself.
Neither change adds new capabilities. Both change the economics of using them.
[^1]: Teknium. "Making a change to the Hermes Agent Curator." X. June 17, 2026.
[^2]: Capafy. "Get paid for your Hermes skills." X. June 17, 2026.
[^3]: Teknium. "Yes use hermes agent and then in telegram." X. June 17, 2026.
[^4]: Shumer, Matt. "Is there a good way to control Claude Code from your phone." X. June 16, 2026.