Top 10 Hermes Agent Skills for Personal Assistants

Best for: Initial Hermes assistant setup, environment tuning, and orchestrating assistant capabilities

hermes-agent acts as the central onboarding and configuration router for the Hermes assistant runtime. It guides initial assistant setup, personality parameters, operational modes, and tool bindings across supported platforms.

This skill is an onboarding and configuration router—not a generic memory layer, dependency manager, or model gateway. Use it to inspect current configuration files, establish operational boundaries, and verify active capabilities before modifying defaults.

• Onboarding router and assistant configuration
• Cross-platform support (Linux, macOS, Windows)
• Inspect configuration before altering defaults
• Distinct from memory stores or model gateways

Best for: Synthesizing structured long-term research notes and concept webs independently of external apps

llm-wiki implements Andrej Karpathy's LLM Wiki paradigm, allowing the personal assistant to build and maintain an interlinked Markdown knowledge base. Raw notes, meeting takeaways, and research findings are organized into topic pages with cross-references and bidirectional links.

This is a self-contained, distinct knowledge-base method, separate from external desktop note applications like Obsidian or Apple Notes. It writes clean, portable Markdown files that the assistant can query and incrementally update during research sessions.

• Karpathy LLM Wiki interlinked markdown pattern
• Self-contained knowledge base independent of desktop apps
• Incremental topic synthesis and cross-referencing
• Plain text portable files under assistant management

Best for: Academic literature search, preprint tracking, and topic discovery across scientific domains

arxiv connects your personal assistant to the public arXiv repository, enabling fast paper searches by title, author, category, or keyword. It extracts abstract summaries and publication dates to keep you informed on relevant scientific literature.

arXiv queries retrieve paper preprints and metadata over the public network; this retrieval capability is separate from claim validation or citation extraction. Pair it with structured note-taking or citation ledgers when synthesizing multi-source findings.

• Direct arXiv API queries for preprints and abstracts
• Author, category, and subject keyword filtering
• Fast research paper discovery and summary extraction
• Retrieval focus distinct from citation ledger verification

Best for: Validating selected citations and evidence against source excerpts during research synthesis

grounded-citations provides an explicit verification ledger to validate selected citations and evidence against referenced source documents during research synthesis. It verifies cited claims against specific source excerpts, though it does not guarantee general hallucination prevention across unrestricted assistant generation.

This verification ledger is functionally distinct from arXiv literature retrieval. Source documents reference bundled helper scripts and supporting files; check your installed copy for their availability—do not assume packaging or installation behavior—and inspect script code before execution on your host.

• Structured claim-to-source verification ledger
• Distinct from raw arXiv literature search
• Validates selected citations and evidence against source excerpts
• Source documents reference bundled helper scripts; check installed copy for availability

Best for: Managing email volume with read-and-draft prompt instructions before approving external actions

email-inbox-triage is conditional because it requires a configured mail backend, such as the Google Workspace integration or the local Himalaya CLI. It helps personal assistants summarize incoming threads, identify action items, and draft suggested replies.

Read, draft, and approval workflows are documented prompt instructions in SKILL.md rather than technical runtime enforcement. The instructions direct the assistant to review threads and draft replies without sending messages or modifying mailbox state (such as archiving or deleting) until explicit human approval is received.

• Conditional on Google Workspace (`gws`) or Himalaya CLI
• Read, draft, and approval workflows are SKILL.md instructions, not runtime enforcement
• Instructs assistant to request approval before send, archive, or delete
• Thread summarization and priority categorization

Best for: Automating Google productivity services when OAuth credentials and project tokens are configured

google-workspace provides broad access to the Google productivity ecosystem, including Gmail, Calendar, Drive, Docs, and Sheets. Source documents reference bundled helpers and supporting Python scripts (or the gws utility) to manage calendar appointments, drafts, and documents; check your installed copy for their availability rather than assuming packaging behavior.

This setup is consequential and requires deliberate configuration. You must set up a Google Cloud Console project, enable respective APIs, download google_client_secret.json, and complete an OAuth2 consent flow to generate google_token.json. Verify account permissions and scopes before enabling access.

• Gmail, Google Calendar, Drive, Docs, and Sheets integration
• Source documents reference bundled helpers and supporting scripts; check installed copy
• Requires Google Cloud Console OAuth2 project setup
• Manages `google_client_secret.json` and `google_token.json`

Best for: Users who organize personal knowledge in Obsidian and want the assistant to manage vault files

obsidian integrates the assistant directly into your existing Obsidian note vault. It reads markdown documents, updates daily notes, creates research entries, and respects Obsidian-style frontmatter and wiki-links (`[[Note]]`).

This is a conditional alternative note store, contrasting with macOS-specific Apple Notes and the standalone `llm-wiki` method. The vault location is configurable via `OBSIDIAN_VAULT_PATH`; if unset, the documented fallback defaults to `~/Documents/Obsidian Vault`. Review note edit boundaries to prevent unwanted overwrites to existing vault files.

• Reads and writes directly inside local Obsidian vaults
• Configurable vault path with `~/Documents/Obsidian Vault` fallback
• Conditional alternative to Apple Notes and standalone `llm-wiki`
• Markdown frontmatter and wikilink compatibility

Best for: Apple ecosystem users who rely on Apple Notes as their primary personal note repository

apple-notes connects your assistant to Apple's native Notes application on macOS. Through the external `memo` CLI, it can search note titles and bodies, retrieve folder structures, and create new formatted notes.

This integration is strictly macOS-only and relies on the documented `memo` CLI alongside macOS Automation permission; verify local system prompts during initial setup. Confirm target folders and account selection before creating or modifying personal notes.

• Strictly macOS-only via the documented `memo` CLI
• Search, read, and create native Apple Notes
• Requires documented `memo` CLI and macOS Automation permission; verify local system prompts
• Conditional alternative store to Obsidian vaults

Best for: Personal assistant communication workflows on macOS requiring message lookups and drafted replies

imessage gives your assistant access to search chat histories and draft messages on macOS. By inspecting the local Messages database and utilizing the external `imsg` CLI with a signed-in macOS `Messages.app`, it enables fast lookups of recent conversations.

Accessing personal message history is highly sensitive. This skill requires four components: the external `imsg` CLI, a signed-in macOS `Messages.app`, Full Disk Access to query `~/Library/Messages/chat.db`, and Messages Automation permission. For outbound messages, SKILL.md instructions direct the assistant to present the recipient and message content for explicit confirmation before dispatch rather than sending automatically.

• Strictly macOS-only; requires external `imsg` CLI and signed-in `Messages.app`
• Requires Full Disk Access for `chat.db` and Messages Automation permission
• Highly consequential access to personal chat history
• SKILL.md instructions direct explicit confirmation of recipient and content before dispatch

Best for: Bounded, user-approved desktop GUI actions when programmatic APIs or command-line tools are absent

computer-use equips the assistant with desktop automation capabilities through Hermes `computer_use` backed by `cua-driver`. While designed to operate background-first, automation actions may escalate and raise application focus when interactive desktop targets require it—never assume focus will not be raised.

Desktop automation presents operational risks and should be invoked only when necessary and for bounded, user-approved actions. Always inspect planned GUI actions beforehand and never permit unsupervised autonomous execution over sensitive applications or financial accounts.

• Hermes `computer_use` backed by `cua-driver`
• Background-first operation, but may escalate and raise window focus when required
• High-impact capability; strictly bounded and user-approved actions
• Fallback option when dedicated APIs or CLI tools do not exist