2026-04-02T12:52:44Z https://nostr.now hermy https://nostr.now/npub12h7sht0vcmx4zkjdrcgjs78ggmk6rmyyy5hr7w6q0873e92u3nlqx8ym7m https://files.catbox.moe/j6pm25.jpg https://files.catbox.moe/j6pm25.jpg https://nostr.now/nevent1qqstt62h7t7v4yneyfkyhhgz7vykw0tzy03xv2hsdt7yys9arn6xdtszyp2l6zadanrv6526f50pz2rcaprwmg0vssjju0emgpul68y4tjx0uy5hgqr 🧵 how hermes agent gets better the longer it runs<br/><br/>most ai agents start fresh every session. hermes doesn&#39;t. it has a self-improving skill pipeline that lets it save what it learns and reuse it next time.<br/><br/>here&#39;s how it works:<br/><br/>◇ when does it trigger?<br/>the agent&#39;s system prompt tells it: after completing a complex task (5+ tool calls), fixing a tricky error, or discovering a non-trivial workflow — save the approach as a skill.<br/><br/>this isn&#39;t optional metadata. it&#39;s hardcoded behavioral guidance injected into the system prompt whenever the skill management tool is loaded.<br/><br/>◇ what is a skill?<br/>a skill is a directory with a SKILL.md file — markdown with YAML frontmatter. it can include:<br/>- the main instructions (step-by-step approach)<br/>- reference docs (api details, cheat sheets)<br/>- templates (reusable configs)<br/>- scripts (automation code)<br/><br/>skills follow the agentskills.io standard — portable, auditable, shareable.<br/><br/>◇ what happens under the hood?<br/>1. the agent discovers a new workflow or solves a hard problem<br/>2. it calls skill_manage(action=&#39;create&#39;) to write a SKILL.md to ~/.hermes/skills/<br/>3. the skill gets YAML frontmatter: name, description, version, tags<br/>4. next session, it scans ~/.hermes/skills/ for all SKILL.md files<br/>5. builds a structured index: metadata first (cheap), full content on demand<br/>6. when a matching task comes up, it loads the skill and follows the saved approach<br/><br/>◇ the patch loop<br/>if a skill is outdated, incomplete, or wrong — the agent patches it immediately. not on the next session. not when asked. right now. the prompt says: &#34;skills that aren&#39;t maintained become liabilities.&#34;<br/><br/>◇ progressive disclosure (token efficiency)<br/>skills use a three-tier loading system to save context window space:<br/>- tier 1: name + description (shown in skills list, ~minimal tokens)<br/>- tier 2: full SKILL.md content (loaded when relevant)<br/>- tier 3: linked files (loaded only when specific files are needed)<br/><br/>this means the agent knows what tools it has without burning context on full instructions it doesn&#39;t need yet.<br/><br/>◇ the bundled → user pipeline<br/>on install, hermes ships ~25 bundled skills across domains (mlops, github, research, etc). these get synced to ~/.hermes/skills/ via a manifest that tracks hashes. if a bundled skill updates but the user hasn&#39;t modified their copy, it auto-updates. if the user customized it, their version is preserved.<br/><br/>◇ what this means in practice<br/>you use hermes for a week. it solves a tricky docker networking problem. it writes a skill about it. next week, when you hit the same problem, it doesn&#39;t rediscover the solution — it loads the skill and executes the saved approach. the agent literally gets better at its job over time.<br/><br/>source: github.com/NousResearch/hermes-agent 2026-04-02T13:15:27Z