2026 Must-Have Hermes Agent Skills: The Most Complete List and Install Priority
If you already run Hermes Agent but feel lost in the Skills Hub, GitHub lists, and MCP plugins, here is the direct answer: do not install everything—keep bundled core skills, then add 3–5 high-value external skills by priority. This guide includes a four-tier decision matrix, scene bundles, a pre-install safety checklist, a Top 20 ranked table, and a seven-step path you can run today (CLI verified against the official Hermes Skills docs as of 2026-05-26).
1. Conclusion first: Hermes Agent Skills are not an install-everything buffet
In 2026, the easiest way to misconfigure Hermes is not the main agent—it is Skills: bundled defaults, the Skills Hub, GitHub community lists, Browser Use, Obsidian, Playwright, SEO/GEO, and Skill Factory all look equally “essential.”
A workable strategy uses four tiers. “Most complete list” here means coverage for decisions, not install them all:
| Tier | Action | Examples | Before install |
|---|---|---|---|
| ① Keep bundled | Do not delete; use reset if broken | plan, TDD, code-review, github-*, native-mcp | Verify against catalog |
| ② New env priority | Install 3–5 from Hub | Browser Use, YouTube, Obsidian, Playwright, Defuddle | Inspect first |
| ③ Scene-specific | Add when workflow is real | Composio, SEO/GEO, ComfyUI, Skill Factory | Inspect + sandbox |
| ④ Install carefully | Skip or read-only audit | Social posting, full mailbox, prod repo write, offensive red-team packs | Source review + least privilege |
Developers should keep planning, debugging, TDD, code review, and GitHub workflows first. Research and content users should add page cleaning, YouTube, Obsidian, and SEO/GEO. Automation users can layer Browser, Composio, MCP, and multi-agent orchestration only after that foundation is stable.
2. What Hermes Agent Skills actually are
Skills are not a flat IDE plugin list. Each one is a workflow knowledge pack loaded on demand: a SKILL.md file describes when to trigger, what steps to follow, and which tools to use. The agent reads the full skill via skill_view only when the task matches—so you do not stuff every instruction into the system prompt.
- Install location: bundled skills copy to
~/.hermes/skills/at install time; Hub skills use the same path (HERMES_HOMEcan override the root). - Discovery: bundled Skills Catalog, Skills Hub (
hermes skills browse), the agentskills.io ecosystem, and custom GitHub sources viahermes skills tap add. - vs Tools: tools are capabilities (read files, run shell); skills are playbooks for recurring task types.
More installed skills does not mean a stronger agent—unused skills stay out of the token budget, but misinstalled high-permission skills widen the blast radius when triggered.
3. Commands you need before installing anything
These subcommands were checked against official docs on 2026-05-26. If anything changes, use the Skills System page and hermes skills --help. In chat, prefix with /skills for equivalent slash commands.
| CLI command | Purpose | How to use in this guide |
|---|---|---|
hermes skills list | Show installed skills | Confirm bundled items were not removed |
hermes skills browse | Browse Hub (--source official optional) | Shortlist—no batch install |
hermes skills search QUERY | Keyword search | Find Playwright, SEO, etc. |
hermes skills inspect ID | Preview before install | Required for external skills |
hermes skills install ID | Install from Hub/URL (security scan) | Only after inspect passes |
hermes skills check | Check upstream updates | Weekly maintenance |
hermes skills update | Reinstall changed Hub skills | Back up before major upgrades |
hermes skills audit | Re-run security scan | After adding high-permission skills |
hermes skills reset NAME [--restore] | Restore bundled default | When bundled skill was edited |
hermes skills config | Enable/disable per platform | Limit skills on 24/7 gateways |
hermes skills uninstall NAME | Remove Hub skill | Drop unused high-permission items |
New installs take effect in new sessions; use /reset or --now on install to refresh the current session (higher token cost on the next turn).
4. Official bundled skills you should keep
Hermes copies the bundled library into ~/.hermes/skills/ at install; hermes update respects local deletes and user edits. The table below prioritizes engineering discipline—paths follow the official Bundled Catalog.
| Skill | Path | What it does | Best for |
|---|---|---|---|
| plan | software-development/plan | Plan mode; writes .hermes/plans/ | All developers |
| writing-plans | software-development/writing-plans | Break work into executable tasks | Larger features |
| systematic-debugging | software-development/systematic-debugging | Four-phase root-cause workflow | Heavy debugging |
| test-driven-development | software-development/test-driven-development | RED-GREEN-REFACTOR | Quality gates |
| requesting-code-review | software-development/requesting-code-review | Pre-PR safety and quality scan | PR workflows |
| subagent-driven-development | software-development/subagent-driven-development | Execute plans via sub-agents | Large repos |
| github-pr-workflow, etc. | github/* | PRs, issues, reviews | GitHub teams |
| native-mcp | mcp/native-mcp | Connect MCP servers | Toolchain extension |
| hermes-agent | autonomous-ai-agents/hermes-agent | Configure and troubleshoot Hermes | Ops / power users |
| obsidian / youtube-content | note-taking/obsidian, media/youtube-content | Vault and video scripts (bundled) | Research/content (compare Hub variants before duplicating) |
Office skills (google-workspace, notion) and research skills (arxiv) are enable when you have accounts; large creative packs can stay dormant until you need them.
5. External skills to prioritize on a fresh 2026 environment
Hub IDs, repos, and maintenance status change—this table gives decision data, not copy-paste install commands. Workflow: hermes skills search … → inspect → install.
| Pick | Priority | Problem solved | Audience | Pre-install check | Privilege |
|---|---|---|---|---|---|
| Obra Superpowers | P0 | Stronger plan/debug/TDD discipline pack | Developers | Overlap with bundled software-development? | Medium |
| Browser Use | P0 | Browser automation, forms, scraping | Ops / QA | Login state, CDP port; no unauthorized bulk actions | High |
| YouTube (Hub) | P1 | Captions → summary / blog draft | Content / research | Egress, copyright, attribution | Medium |
| Obsidian (Hub/community) | P1 | Read/write vault, search notes | Personal KB | Vault path scope, sync conflicts | Med–high |
| Defuddle | P1 | Clean article body from web pages | Research / SEO | Fetch rate, site ToS | Medium |
| Playwright | P1 | E2E tests, visual regression | Web QA | Browser binaries, CI resources | Med–high |
| Composio | P2 | SaaS integration orchestration | Office automation | OAuth scopes; minimal connectors | High |
| SEO / GEO | P2 | Site structure and discoverability | Content sites | Writes to production CMS? | Medium |
| Skill Factory | P2 | Generate custom skills from workflows | Long-term skill building | Can generated scripts run shell? | Med–high |
| SkillClaw / Reflexion | P3 | Skill discovery / self-improvement loops | Advanced users | Risk of auto-installing other skills | High |
6. Scene bundles: which group fits you
| Scene | Suggested stack (bundled + external) | Inspect focus |
|---|---|---|
| Developer | plan + TDD + code-review + github-* + subagent-driven-development; optional Obra Superpowers | Duplication vs bundled |
| SEO / content site | Defuddle + SEO/GEO + humanizer (bundled) + writing-plans | CMS / DB writes |
| Research | arxiv + blogwatcher + Defuddle + youtube-content | Scraping compliance |
| Personal knowledge base | obsidian + llm-wiki + ocr-and-documents | Vault path allowlist |
| Office automation | google-workspace / notion + Composio + native-mcp | Minimal OAuth scope |
| Web testing | Playwright + dogfood + systematic-debugging | Staging URLs, credentials |
| Design / creative | sketch + excalidraw + architecture-diagram | Output paths and size |
| Security audit (defensive) | requesting-code-review + codebase-inspection; avoid offensive red-team packs | Authorized targets only |
| Long-term skill building | hermes-agent-skill-authoring + Skill Factory + your own SKILL.md | Generated script permissions |
7. Skills you should not install without scrutiny
These belong in tier ④—not banned, but inspect, read source, and apply least privilege for authorized, compliant use only:
- Browser login state / Browser Use: may retain cookies; never run unauthorized bulk actions on banking, email, or social accounts.
- Cloud accounts & Composio: every connector expands leak surface—enable only what the task needs.
- X/Twitter (xurl), WhatsApp, full mailbox access: fine for low-frequency personal assist; not for spam or bypassing platform limits.
- Production repo write access: scope github-* workflows to branches and directories you trust.
- Anthropic-Cybersecurity-Skills, godmode, jailbreak packs: authorized test environments only—never abuse third-party services.
Pre-install safety checklist (8 items)
- Source: official Hub or trusted GitHub org—avoid random single-file URLs
- Read full
SKILL.mdand frontmatter - Check
scripts/for shell, curl, rm -rf - Network: are outbound domains necessary?
- Secrets: does it read
.env, Keychain, or plaintext tokens? - Filesystem: is access limited to intended directories?
- Maintenance: recent commits and open issues?
- Reversible: uninstall, reset, update supported?
8. My 2026 Top 20 ranked list
Order balances beginner safety, dev velocity, content/research, automation, and long-term upkeep. “Bundled” needs no Hub install; “External” means search → inspect → install.
| # | Skill | Type | Scene | Before install |
|---|---|---|---|---|
| 1 | plan | Bundled | Dev | Confirm present |
| 2 | writing-plans | Bundled | Dev | — |
| 3 | systematic-debugging | Bundled | Debug | — |
| 4 | test-driven-development | Bundled | Dev | — |
| 5 | requesting-code-review | Bundled | PR | — |
| 6 | subagent-driven-development | Bundled | Large tasks | — |
| 7 | github-pr-workflow | Bundled | GitHub | gh auth |
| 8 | native-mcp | Bundled | Extension | MCP config |
| 9 | hermes-agent | Bundled | Ops | — |
| 10 | Obra Superpowers | External | Dev boost | inspect |
| 11 | Browser Use | External | Automation | inspect + isolate |
| 12 | Defuddle | External | Research/SEO | inspect |
| 13 | youtube-content / Hub YouTube | Both | Content | Pick one; avoid duplicate |
| 14 | obsidian | Both | KB | Vault path |
| 15 | Playwright | External | Web QA | inspect |
| 16 | arxiv + blogwatcher | Bundled | Research | — |
| 17 | SEO / GEO | External | Content site | inspect |
| 18 | Composio | External | Office | Minimal connectors |
| 19 | hermes-agent-skill-authoring | Bundled | Custom | — |
| 20 | Skill Factory | External | Skill building | inspect scripts |
Reference numbers: the official bundled catalog lists 20+ categories and 100+ entries (verify on install day); Hub sources include official, skills-sh, well-known, and tap add custom repos. For a new machine, keep external installs ≤ 5, run stable for two weeks, then expand.
9. Why “install everything” fails in practice
- Constraint: big lists feel like mandatory installs. Community READMEs rank by GitHub stars, not whether you use Obsidian or browser automation daily.
- Hidden cost: high-permission skills expand audit surface. Every skill that can read mail, post socially, or drive a browser is a misfire risk—worse on a 24/7 gateway host.
- Stability: install once, never maintain. Skipping
check/update/auditleaves stale SKILL.md after Hermes upgrades.
10. Seven-step path you can run today
- Validate Hermes itself: if not installed yet, follow the install & config guide; run
hermes doctorwith no fatal errors. - List bundled skills:
hermes skills listvs section 4; restore gaps withhermes skills reset --restoreif needed. - Pick one scene bundle: choose one row from section 6 and name only 2–3 external skills to add.
- Inspect each candidate:
hermes skills search …thenhermes skills inspect <id>; pass the section 7 checklist. - Install and sandbox:
hermes skills install <id>; test in~/hermes-lab, not a production repo. - Verify in a fresh session:
/resetor a newhermes chat; confirm skills trigger as expected. - Calendar maintenance: weekly
skills check, monthlyskills audit; align with the long-term maintenance guide backup policy.
11. Post-install maintenance
- Update on a schedule:
hermes skills check→hermes skills update(runhermes backupbefore major Hermes upgrades). - Reset when bundled skills break:
hermes skills reset NAME --restore. - Remove unused high-permission skills:
hermes skills uninstallplusskills configto disable on gateways. - Capture real workflows: use
hermes-agent-skill-authoringor Skill Factory to author your own skills instead of hoarding Hub trends.
12. Run Hermes Skills on Mac mini with less friction
Browser Use, Playwright, Obsidian, local MCP, and macos-computer-use depend on a native macOS stack—no WSL, and Homebrew plus Unix tooling are ready out of the box. Mac mini M4’s unified memory handles Hermes gateways, browser automation, and light local models comfortably; ~4W idle power makes it a practical 24/7 skills sandbox without tying up your daily laptop.
Gatekeeper, SIP, and FileVault give clearer boundaries around ~/.hermes/ than a typical Windows workstation. Combined with the inspect workflow above, a dedicated mini is a safer place to validate high-permission skills before rolling them to a team gateway.
If you want a quiet, stable host for layered Hermes Skills, Mac mini M4 is one of the best value starting points—run your prioritized stack on a dedicated node instead of experimenting on a full-permission daily driver.
Run your Skills workflow on Mac mini
Remote macOS nodes with low power draw—ideal for Hermes gateways and isolated browser-skill trials.