hermes

Hermes

Use First

Use this skill for Hermes Agent self-hosting and day-to-day operation: install, setup, model/provider selection, gateway platforms, Slack channel setup, command discovery, updates, plugins, tools, skills, memory, profiles, MCP, cron, Kanban, delegate_task subagents, logs, dashboards, tutorials, and long-running tmux sessions.

Do not rely on stale remembered commands. Hermes changes quickly. At the start of each Hermes task, resolve scripts/hermes_check.py relative to this skill directory and run:

python3 <skill-dir>/scripts/hermes_check.py --json

If command behavior matters, also refresh the generated local command snapshot:

python3 <skill-dir>/scripts/hermes_check.py --write-help <skill-dir>/references/command-map.generated.md

Report update availability before changing configuration. Do not run hermes update without explicit user approval unless the user already asked to update.

Decision Flow

1. Verify local state with scripts/hermes_check.py.
2. Use hermes --help and hermes <command> --help as the source of truth for executable commands.
3. Load only the needed reference file:
   • references/setup-and-update.md for install, versioning, updates, config health, backups, and safe update policy.
   • references/getting-started-tutorials.md for first-run setup, self-hosting paths, guided usage flows, and user-facing tutorials.
   • references/command-catalog.md for command categories and common CLI patterns.
   • references/official-links.md for official docs, release, reference, and troubleshooting links.
   • references/slack-gateway.md for Slack app, channel, Socket Mode, manifest, gateway, and allowlist setup.
   • references/operations-runbook.md for tmux sessions, one-shot execution, profiles, cron, webhooks, memory, skills, and autonomy guardrails.
   • references/kanban-operations.md for durable task boards, profile workers, dependencies, isolated workspaces, dispatch, and task notifications.
   • references/delegation-vs-kanban.md for delegate_task, subagent delegation, and choosing between delegation, Kanban, tmux, cron, and background terminal work.
   • references/slash-commands.md for in-session /... commands, gateway slash commands, and when to prefer CLI commands.
   • references/providers-tools-security.md for key paths, config sections, providers, toolsets, approval modes, redaction, and privacy.
   • references/plugins-and-tools.md for user/operator plugin installation, tool enablement, MCP connections, and safe verification.
   • references/voice-and-media.md for STT, TTS, voice mode, image input, and media-related troubleshooting.
   • references/self-improvement.md for updating this skill after new Hermes behavior, command changes, or operational lessons are discovered.
   • references/troubleshooting.md for logs, Slack failures, token problems, stale skills, gateway issues, and community-known pitfalls.
   • references/command-map.generated.md only when exact local help output is needed.
4. For interactive Hermes or tmux work, read the current pane before sending keys. Explain prompts and recommend a choice before sending input.
5. Prefer deterministic commands and scripts for inspection. Keep non-deterministic guidance in reference docs or in the Hermes conversation itself.
6. At the end of each Hermes task, run the self-improvement review in references/self-improvement.md.

Common Request Routing

• "Install Hermes" or "self-host Hermes" -> getting-started-tutorials.md, setup-and-update.md, then official-links.md.
• "Set up Slack channel" -> slack-gateway.md, run scripts/hermes_slack_check.py, then troubleshooting.md.
• "Check if my Hermes is healthy" -> run scripts/hermes_ops_check.py, then setup-and-update.md.
• "Enable a plugin/tool/MCP server" -> plugins-and-tools.md, providers-tools-security.md.
• "Teach me how to use Hermes" -> getting-started-tutorials.md, slash-commands.md, command-catalog.md.
• "Use Kanban", "dispatch workers", or "coordinate profiles" -> kanban-operations.md, operations-runbook.md.
• "Use delegate_task", "delegate this", "spawn subagents", or "parallel subtasks" -> delegation-vs-kanban.md; use Kanban instead when the work must be durable, human-visible, restartable, or assigned to named profiles.

Current CLI Baseline

As of Hermes Agent v0.12.0, prefer:

hermes
hermes chat -q "single query"
hermes -z "script-friendly one-shot"
hermes setup
hermes model
hermes doctor
hermes config check
hermes update --check
hermes gateway setup
hermes gateway status
hermes slack manifest

Avoid old third-party examples built around hermes run; current Hermes does not expose run as a root command in v0.12.0.

Safety Rules

• Treat ~/.hermes/.env, Slack tokens, OAuth tokens, and ~/.hermes/auth.json as secrets. Never print full token values.
• Set or verify allowlists before enabling Slack or other gateways.
• Use hermes update --check before update decisions; use hermes backup or the update backup flags for risky changes.
• For automation, start with read-only/reporting workflows, add verification, then schedule with cron/webhooks only after repeated successful manual runs.
• Prefer files over chat for durable Hermes behavior: config.yaml, .env, SOUL.md, AGENTS.md, skills, and runbooks.