Agent
Verifies and maintains the integrity of a live Hermes editable-install deployment (the editable-install pattern: checked-out git branch IS the running code). Use PROACTIVELY before/after any restart, update, or dev work on the live tree, and for "is the deploy healthy", "did we drift off the integrated branch", "the…
Agent
Benchmarks and QAs a whole live Hermes Agent instance across capability categories (basic chat, tool use, weather, ops, plus whatever the instance is for) — fast and in parallel, with latency and regression tracking. Use PROACTIVELY for "run the QA pass", "is the live instance healthy", "smoke-test Hermes", "did that…
Agent
Engineers and extends the Hermes messaging gateway — platform adapters, lifecycle hooks, authorization, session routing, delivery, and provider routing. Use PROACTIVELY whenever the user wants to add or modify a gateway adapter or hook, change auth/pairing, fix session-key routing or message delivery, wire a new…
Agent
Tests and improves a Hermes gateway end-to-end — the serving path, authorization, session routing, slash-command dispatch, delivery, and agent-through-gateway behavior — then drives a failure-oriented improvement loop. Use PROACTIVELY after any gateway change, before a gateway rollout, or for "test the gateway", "is…
Agent
Authors and edits skills FOR a Hermes Agent install. Use PROACTIVELY whenever the user wants to create, write, draft, refactor, or fix a Hermes skill, wrap a CLI/API as a Hermes capability, or turn a repeated Hermes workflow into a reusable SKILL.md. MUST BE USED for any "make/build/improve a Hermes skill" task so the…
Agent
Builds and runs evaluation suites for an individual Hermes skill, scores the results, finds failure classes, and recommends concrete edits. Use PROACTIVELY after a Hermes skill is written or changed, or whenever the user asks "is this skill any good", "test this skill", "why isn't my skill triggering", or "make this…
Skill Claude CodeCodex
Systematically debug a misbehaving Hermes install: wrong/old code running after dev work, config changes that don't take effect, provider/model 404s, simple ops queries that explode into 6-17 delegated model calls, missing tools, "which city?" weather failures, the gateway looking dead after a restart, or runs that…
Skill Claude CodeCodex
The deployment invariant and recovery procedure for a Hermes editable install where the checked-out git branch IS the running code (editable-install pattern). Use this skill WHENEVER restarting, updating, or doing dev work near the live Hermes tree, when the gateway won't stop/status correctly after a dashboard…
Skill Claude CodeCodex
Maintain a personal Hermes fork: rebase feature branches onto upstream/main, rebuild an integration branch, resolve conflicts on the hot files (delegatetool.py, gateway/run.py, hermescli/config.py, apiserver.py), and enforce the editable-install invariant — the checked-out branch IS the running code, so the live tree…
Skill Claude CodeCodex
Authoritative map of how Hermes Agent (NousResearch) works under the hood: the four ways to drive it (CLI, the AIAgent Python library, the OpenAI-compatible API server, and the messaging gateway), which surface to use for which job, the AIAgent constructor knobs that control speed/cost, and the gateway's message flow…
Skill Claude CodeCodex
How a Hermes orchestrator should ROUTE work: match the execution mechanism to the task. Deterministic single-command ops (is X running, service status, disk, logs, docker ps, cluster health) go DIRECT to a cluster-ops tool — never delegate, never KB-search first. Reasoning/judgment work (debug+propose, plan…
Skill Claude CodeCodex
Tune a local-model Hermes for latency: run the warm gateway/interactive path not a cold one-shot CLI, trim toolsets per profile, pin frequently-used tools so they aren't deferred behind toolsearch, keep the model resident, weigh the 32k-vs-64k context VRAM tradeoff, offload compression to a fast cloud model, and pick…
Skill Claude CodeCodex
How to author and edit skills FOR Hermes Agent (NousResearch) — the Hermes SKILL.md frontmatter contract, body structure, conditional activation, env vars and credential files, template tokens, inline shell, media delivery, and publishing. Use this skill WHENEVER creating, editing, reviewing, or debugging a Hermes…