0xNyk/awesome-hermes-agent

nyk.dev/oss · Hermes Agent · Brand kit

Awesome Hermes Agent

Curated skills, tools, integrations, and resources for Hermes Agent, the self-improving agent from Nous Research.

Hermes Agent is the open-source agent maintained by Nous Research. Awesome Hermes Agent is 0xNyk's independent open-source directory for the optional skills, plugins, memory providers, surfaces, and bridges around it. This directory is not an official Nous Research project.

Ecosystem status (last reviewed: 2026-07-16)

• Hermes Agent: v0.18.2 (v2026.7.7.2)
• Core repo: NousResearch/hermes-agent
• Releases: Hermes releases

Where Do I Start?

Three steps from zero to productive:

1. Run Hermes: Official docs quickstart (install, CLI, first chat).
2. Add skills: wondelai/skills (broad library) or litprog-skill (literate programming).
3. Add a surface: hermes-workspace (Hermes GUI) or mission-control (fleet / dispatch dashboard).

Then use the full list. Every entry has a maturity tag:

Tags are editorial snapshots based on documentation, setup evidence, maintenance, and adoption signals. Re-check the linked project before depending on it.

Check the trust boundary

An ecosystem listing is a discovery aid, not a security endorsement. Before enabling a community skill, plugin, MCP server, or unattended job, verify who can trigger it, which tools it receives, where commands execute, what credentials it can read, and how you can stop it. Prefer the smallest toolset and an isolated backend for untrusted work.

Use the official security guide, tools reference, and cron documentation for the current controls.

For the operator routine around /new, /resume, /compress, /branch, export, and pruning, use the Hermes Agent session management guide.

Contents

• Where Do I Start?
• Check the Trust Boundary
• Official Resources
• Skills & Plugins
  • Community Skills
  • Plugins
  • agentskills.io Ecosystem
  • Skill Registries & Discovery
• Memory Providers
• Tools & Utilities
  • Deployment
• Integrations & Bridges
• Detection & Media Forensics
• Multi-Agent & Swarms
• Domain Applications
• Forks & Derivatives
• Guides & Documentation
• Operational Playbooks
• Level-Up Blueprints
• Contributing
• License

Official Resources

Core repositories and resources maintained by Nous Research.

• Hermes Agent by Nous Research - The core project. Self-improving agent with a closed learning loop, skill creation and curation, persistent memory, 20+ messaging platforms, six terminal backends (local, Docker, SSH, Singularity, Modal, and Daytona), cron scheduling, MCP integration, isolated profiles, provider routing, and automatic migration from OpenClaw. 215k+ stars.
• autonovel by Nous Research - Autonomous novel-writing pipeline built on Hermes. Generates long-form manuscripts (100k+ words) end-to-end using the agent loop.
• hermes-paperclip-adapter by Nous Research - Run Hermes as a managed employee in Paperclip companies. Connects the agent to Paperclip's task management and governance system.
• hermes-agent-self-evolution by Nous Research - Evolutionary self-improvement using DSPy and GEPA (Genetic Evolution of Prompt Architectures). The research pipeline for optimizing Hermes's own prompts and behaviors.
• Official Documentation - Comprehensive docs covering quickstart, CLI, configuration, messaging gateway, security, tools, skills, memory, MCP, cron, ACP, API server, and architecture.
• Release Notes - Official changelog with feature highlights, migration notes, and reliability fixes for each Hermes version.
• tinker-atropos by Nous Research - Standalone Atropos integration with Thinking Machines Tinker API. RL training infrastructure for fine-tuning tool-calling models on real agent trajectories.
• Skills Hub - The open standard for agent skills. Compatible across Hermes, Claude Code, Cursor, Codex, and other agents.
• Discord - The Nous Research community. Bug reports, feature requests, and general discussion.

Skills & Plugins

Skills are procedural memory — reusable capabilities that Hermes creates from experience and improves during use. Plugins extend core functionality.

Community Skills

• [beta] hermes-plugins by 42-evey - Plugin suite: Discord voice bridge with Gemini Live, WhatsApp bridge, goal management, inter-agent bridge, model selection, and cost control. Maintained by a frequent Hermes core contributor.
• [beta] hermes-skill-factory by Romanescu11 - Meta-skill that auto-generates reusable skills from your workflows. Point it at a task you repeat and it creates a skill for it.
• [beta] litprog-skill by tlehman - Literate programming skill that works across Claude Code, OpenCode, and Hermes. Weaves code and prose into documented, executable notebooks.
• [experimental] Wizards-of-the-Ghosts by Hmbown - Fantasy spell-themed skill pack. Wraps real development operations (refactoring, linting, testing) in a tabletop RPG interface.
• [experimental] super-hermes by Cranot - Teaches Hermes to write its own analytical prompts. Adds a meta-reasoning layer where the agent generates better prompts for itself before executing tasks.
• [experimental] hermes-life-os by Lethe044 - Personal OS agent that detects daily patterns and learns your routines over time. Uses Hermes's memory system for lifestyle tracking, not just code.
• [beta] hermes-incident-commander by Lethe044 - Autonomous SRE agent for production incident detection and self-healing. Monitors services, diagnoses failures, and applies fixes. Works well with Hermes's cron scheduling.
• [beta] hermes-dojo by Yonkoo11 - Self-improvement system that monitors agent performance, identifies weak skills, and iterates on them automatically.
• [beta] hermes-spotify-skill by Alexeyisme - Spotify playback control for headless Linux and Raspberry Pi 4/5. Search, play, pause, skip, set volume, transfer between Spotify Connect devices. No daemon — Hermes writes spotipy snippets and runs them via execute_code. Works with raspotify for Pi-as-speaker. Tested on Raspberry Pi OS Lite Bookworm 64-bit. The only Linux-native Spotify skill in the ecosystem.
• [experimental] hermes-skill-marketplace by Lethe044 - Agent that writes, tests, and publishes new skills autonomously. Automates the skill creation and distribution lifecycle.
• [experimental] personal-api by beiyuii - Turn your Obsidian vault into an identity layer any AI agent can read in under 30 seconds
• [beta] hermes-nextcloud by adnw-vinc - Self-hosted Nextcloud bridge — manage files (WebDAV), notes (Nextcloud Notes API), calendar/tasks (CalDAV), and contacts (CardDAV) from Hermes. App Password auth, configurable timezone, guided setup. Fills the self-hosted-cloud gap for Hermes users running their own infrastructure.
• [beta] oh-my-hermes by witt3rd - Multi-agent orchestration skills for Hermes inspired by oh-my-claudecode and rebuilt on Hermes primitives. Suite covers deep-research, deep-interview, ralplan (Planner → Architect → Critic consensus), ralph (verified execute → verify → iterate), triage, and autopilot, plus driver skills encoding the dispatcher's playbook. Composes end-to-end: research → interview → consensus plan → verified execution.
• [beta] blacktea by nmrtn - Spending controls for agents that pay online via x402. Holds over-limit payments for human approval in the chat, enforces a JSON policy, and audits every spend. Ships as an MCP server (pay, approve_payment, reject_payment, audit_query) plus an SDK and CLI.
• [beta] hermes-skill-kit by duruonanni - Three community-vetted skills: feishu-document-api (programmatic Feishu/Lark document creation), feishu-response-format (Feishu response formatting with 3 table schemes, Card JSON 2.0, Markdown compatibility, 53 tests), and hermes-memory-maintenance (MEMORY.md/USER.md audit and drift recovery). Codex-audited.
• [beta] PolyBrain by mosesman831 - Multi-agent, multi-model orchestration skill that decomposes an objective into a JSON task plan, runs research and build subtasks in parallel, then synthesizes and verifies a final answer. Each role can use a different LLM/provider. Enforces strict citations and drops uncited claims.
• [experimental] tech-debt-management-skill by Vincent-crypto-coder - Systematic technical debt management for Python projects. Automates scanning (pylint, radon, safety, bandit, flake8), prioritizes debt with a 4-dimension matrix, and generates executable refactoring plans with verification scripts.
• [beta] pingpong by 0xAaronx0 - Spontaneous local leisure meetups: agent posts a pseudonymous, coarse-geo offer to a shared board; nearby agents match locally and negotiate place/time with double-opt-in before contact.
• [beta] hermes-skills by RobinBeraud - Growing skills library covering finance (FMP stock/crypto data), web (Firecrawl scrape/search), WhatsApp (WBizTool contacts), and marketing/SEO. Each skill follows standard SKILL.md format.
• [beta] hermes-skills (liuyunnb666) by liuyunnb666 - Seven curated skills for Hermes: debugging, parallel dev, TDD, and UI workflows.
• [production] hermes-skills (winterliu6) by winterliu6 - Chinese-focused skills pack: content-studio (小红书/抖音/公众号), security-newsletter (weekly cybersecurity awareness emails), and tech-ops-bot (企业微信/飞书 IT ops with fault classification and web dashboard). Production use at Yaolai Cinema IT.
• [beta] hurmoz by Moshe-ship - Arabic-first skills pack — 63 skills across Islamic tools (prayer times, zakat, Quran), dialect-aware NLP, content, and travel. Install: hermes skills tap add Moshe-ship/hurmoz.
• [experimental] humanizer-ru by ilyautov - Removes AI-writing tells from Russian text (bureaucratese, calques, ChatGPT/Claude fingerprints). 52 patterns, deterministic offline scanner. SKILL.md-compatible. MIT.
• [experimental] hinge-projects by tsol - Starts Vite projects, a Hinge editing overlay, Cloudflare tunnels, and Telegram control for phone-based development with Hermes. Early project; review the install script before running it.
• [beta] stock-analysis by AdvancingTitans - Python CLI and portable skill for A-share, Hong Kong, US stock, fund, and portfolio research. Produces Markdown reports and machine-readable evidence packs; research only, with no trading execution.
• [experimental] nika-agents by SuperNovae Studio - Delegates repeatable work from Hermes to Nika workflows with static checks, cost caps, permits, and hash-chained execution traces. The integration skill is MIT; the Nika repository is AGPL-3.0.
• [beta] LinkedIn Skills by Serge Bulaev - Ten portable skills for drafting, reviewing, and planning LinkedIn content. MIT licensed and compatible with agentskills.io hosts, including Hermes.
• [experimental] routerbase-agent-skills by zenlee123 - Skills for RouterBase model routing and media generation. Requires RouterBase credentials; MIT licensed and still at its initial public revision.
• [experimental] skill-evolution by okdk7788 - GEPA-style measured evolution for learned agent skills, adapted from Hermes self-improvement ideas for Claude Code. MIT licensed, but still at its initial public revision.
• [beta] emulo by ohad6k - Mines local Claude Code, Codex, and Copilot session logs into a portable user profile that Hermes can load as a skill. Runs locally and is MIT licensed.
• [experimental] code-assembly-skill by baoyu0 - Search-first software assembly method for finding and composing existing tools before writing new code. Portable across Hermes, Claude Code, and Codex.
• [experimental] hermes-do-task-skill by baoyu0 - One-command workflow that plans a task, delegates execution, verifies the result, and writes a handoff. Early project with no published release.
• [beta] hermes-telegram-checklist by johnsje183 - Native Telegram To-Do lists (real interactive checkboxes with a progress counter) in chats and forum topics — create, read, append, toggle with a write allowlist and offline plan validation. Uses a Telethon user session (MTProto) because the Bot API can't post checklists into groups; treat the session account accordingly. MIT, CI across Python 3.9–3.13.
• [beta] hermes-tour-assistant by MaikD77 - Three-in-one outdoor skill suite: Outdoor Tour Assistant (GPX route monitoring with weather, deviation alerts, and silent-by-default cron), City Walk Guide (personal OSM/Wikipedia walking tour with source-backed stories and optional voice), and Live Location Nearby (water, food, repair, and shelter via OSM/Overpass). MIT, Python 3.11+, 85%+ test coverage.
• [beta] weread-notes-export-skill by dongwei6688 - WeChat Reading (微信读书) notes export: chapter-tree Markdown for highlights and comments, incremental sync for daily cron, safe filenames, same-title/different-author separation. Pure Python stdlib; needs a WeRead API key.
• [beta] policy-search-china-skill by dongwei6688 - Search Chinese government policy documents across 16 ministries and extract verbatim citations with source links for reports and planning. Ships a large local policy cache plus scripted search chains; dual-space layout keeps user data separate from system updates. MIT.
• [beta] true402-token-safety by true402 - Pre-trade Base (8453) ERC-20 safety gate: gas-free on-chain buy→sell simulation that proves whether a token can be sold, plus liquidity and ownership/mint/proxy checks, returning avoid / caution / ok. Free daily checks need no wallet; heavier use pays per call over x402 with a hard $0.10 client-side cap. MIT-0 skill wrapping @true402.dev/rugcheck.
• [beta] twitr-skills by lnvestor - Five X/Twitter agent skills: live reads and search, bulk dataset exports, real-time account/keyword monitors (HMAC-signed webhooks), publishing via a connected account, and an unattended presence routine for Hermes cron. Pay-per-call in USDC (x402 on Base/Solana, MPP on Tempo) — no API key or signup; treat publish/presence skills as high-privilege. MIT.
• [experimental] loremaster by loremaster-ai - AI scrum-master and PM skill pack extracted from a production multi-project deployment. Sprint ceremonies with planning poker, backlog proposals with verifiable done criteria, plan-conflict detection, and a per-project wiki vault with a knowledge graph. Every external write (Jira, wiki commits) goes through per-item human approval. MIT.
• [experimental] hermes-backup-recovery by JimmyHuang2002 - Keeps a long-running Hermes recoverable: age-encrypted backups of everything under $HERMES_HOME (memory, skills, config, chat history) with verification, restore drills, and a health check. Four short bash scripts plus a SKILL.md — no lock-in, archives are plain tar.gz. MIT.
• [experimental] obsidian-auto-context by SkillForge Labs - Captures each session's findings, decisions, and action items into an Obsidian vault as tagged, wiki-linked Markdown with daily-note rollups. Secrets are redacted before anything touches disk. MIT.
• [experimental] appnz-agent-skills by lee101 - Portable deploy skill for app.nz: static sites, server apps, scale-to-zero Cog models, and bounded coding-agent jobs, with target/cost checks, health checks, and rollback evidence built into the workflow. Skill is MIT; deployments require an app.nz account and may consume paid credits.
• [experimental] url-manager by Piccolo123 - Agent-first URL collection and knowledge management: agents save, categorize, tag, search, and share web links, with collaborative shared categories and magic-link delivery back to the user. MIT client fronting a hosted backend (ai.ocean94.com) — an account is auto-created on first use and collected links are stored there; SKILL.md requires informing the user before first use.
• [beta] OMH (oh-my-hermes by rlaope) by rlaope - Wrapper orchestration layer installed next to Hermes — one omh setup installs 80+ generated workflow skills plus a deterministic chat router that turns free-form Korean/English chat into the right prepared workflow, clarifying instead of guessing when confidence is low. Coding work is delegated to an explicitly selected executor (Codex, Claude Code, or the Hermes runtime) with a strict prepared-vs-observed evidence boundary. A 156-case bilingual routing corpus and ~1,500 unit tests gate every change in CI. Pure Python, zero runtime dependencies, never patches Hermes; docs in four languages. Distinct project from witt3rd's oh-my-hermes above. MIT.
• [beta] roadmap-tutor by instax-dutta - Turns any of the 91 official roadmap.sh roadmaps into a resumable, one-topic-at-a-time curriculum with progress tracking across sessions. Topology-sorted learning path from reactflow graphs, stdlib-only Python, runs on Raspberry Pi 4 with 2GB RAM. Commands: start, next, done, skip, status, outline, search, goto, switch, prefetch, reset. Weekly CI verifies lossless parse across all 91 live roadmaps. MIT.
• [experimental] hermes-startup by 33hodl - First-dollar onboarding skill: walks a new operator from a working Hermes install to one genuine paid outcome, rather than to a demo.
• [beta] alter by alter-persona - Builds a digital persona from your own writing, so an agent can draft in your voice instead of a generic one. Apache-2.0.
• [beta] shopify-app-review-brief by alfredtech2026 - Privacy-first Shopify app review triage: source-checked worksheets an operator can audit rather than a black-box score.
• [experimental] schedulala-agent by schedulala - Schedules and publishes social posts to 12 platforms from an agent. Holds credentials for every platform it posts to, so scope its tokens deliberately.

agentskills.io Ecosystem

Skills built on the agentskills.io open standard — compatible across Hermes and other agent platforms.

• [production] wondelai/skills by wondelai - Cross-platform agent skills for Claude Code and agentskills.io-compatible platforms.
• [production] youtube-skills by therohitdas - Adds YouTube search, channel browsing, playlist extraction, and reliable transcripts on top of Hermes's built-in youtube-content. Built-in transcript fetch silently fails on VPS because YouTube blocks cloud IP ranges; this routes through TranscriptAPI's backend (15M+ transcripts/month) to stay unblocked. 12 sub-skills covering search → fetch → bulk extraction. Auto-invokes for creator lookups, topic research, and tutorials. The fix for "my Hermes can't watch YouTube on a $5 VPS."
• [production] Anthropic-Cybersecurity-Skills by mukul975 - 753+ structured cybersecurity skills mapped to MITRE ATT&CK. 25k+ stars.
• [production] chainlink-agent-skills by Chainlink - Official Chainlink agent skills on the agentskills.io spec. Oracle network data, CCIP, and smart contract interaction.
• [production] black-forest-labs/skills by Black Forest Labs - Official FLUX model skills for image generation. First-party skills from the FLUX creators.
• [production] pydantic-ai-skills by DougTrajano - Pydantic AI with agentskills.io support. Adds type-safe schema validation to agent skill inputs and outputs.
• [beta] execplan-skill by tiann - Manages complex, long-running task execution with proper lifecycle handling — progress tracking, checkpoints, and failure recovery.
• [beta] maestro by ReinaMacCredy - Skill orchestration with Conductor planning and Beads tracking. Structures multi-step skill execution into observable pipelines.
• [beta] bmad-module-skill-forge by armelhbobdad - Transforms repos and docs into agentskills.io-compliant skills. Input a codebase, output installable skills.
• [beta] Agentic-MCP-Skill by cablate - MCP client with agentskills.io validation. Bridges MCP tool servers with the skills standard.
• [experimental] ripley-xmr-gateway by KYC-rip - Monero (XMR) blockchain gateway for agents. Enables private cryptocurrency transactions from agent workflows.
• [beta] skillsdotnet by PederHP - C# implementation of agentskills.io with MCP integration. .NET alternative to the Python/TypeScript SDKs.
• [beta] colony-skill by TheColonyCC - Collaborative intelligence platform where AI agents and humans post findings, discuss ideas, complete tasks, earn karma, and build reputation. Community hub at thecolony.cc.
• [beta] AgentCash by Merit-Systems - Skill giving agents access to 300+ premium APIs and a wallet balance to pay for them through x402 or MPP. A fresh Hermes install with only AgentCash is actually powerful — from web scraping to image generation to email sending, all through one skill with free USDC for trying out.
• [beta] x-twitter-scraper by Xquik-dev - Typed X (Twitter) access via 43 narrow SKILL.md folders covering reads (search, timelines, mentions, trends, articles, bookmarks, for-you), writes (post, DM, follow, profile updates), bulk extraction (followers, communities, lists, spaces), AI composition (write-tweets, write-threads, optimize, going-viral), giveaways, monitors, and webhooks. No browser automation, no scraping — typed JSON in/out through the Xquik API. Fills the X gap for Hermes social workflows.
• [production] drawio-skill by Agents365-ai - Generates draw.io diagrams from natural language and exports to PNG/SVG/PDF. SKILL.md format, works across Claude Code, OpenClaw, Hermes, Codex. Useful when an agent needs to communicate architecture or process visually without a separate design step. 5.8k+ stars.
• [production] open-design by nexu-io - Local-first, open-source alternative to Anthropic's Claude Design. 31 composable skills (web/mobile/decks/dashboards/documents) over 129 design systems (Linear, Stripe, Vercel, Notion, Apple, …) with image (gpt-image-2), video (Seedance 2.0, HyperFrames), and audio generation. Auto-detects 15 coding-agent CLIs from PATH and integrates with Hermes via ACP/JSON-RPC. BYOK proxy, sandboxed previews, can import Claude Design exports. 78k+ stars.
• [beta] authsome by agentrhq - Local OAuth2 and API credential broker for agents. Log in once via browser PKCE or device code, credentials live in a local encrypted vault, and a local proxy injects them at request time so raw keys never enter the agent's environment. 45 providers preconfigured (14 OAuth2, 31 API credentials) including GitHub, Google, OpenAI, Linear, Slack, Notion, Resend, and Stripe. Skill installs into Hermes with hermes skills install agentrhq/authsome/skills/authsome. MIT.
• [production] outreachmagic by Outreach Magic - GTM pipeline agent that syncs Smartlead, Instantly, HeyReach, PlusVibe, EmailBison, Prosp, and Calendly into a local SQLite database your agent can query.
• [production] abvx-agent-skills by markoblogo - Auditable cross-platform coding-agent skillpack for smaller diffs, evidence-led debugging, token control, browser verification, and reviewable SKILL.md workflows with static security audit.
• [production] dev-gtm-claude-skills by Infrasity-Labs - Cross-platform agent skills for SEO, GEO (Generative Engine Optimization), AI discoverability, and developer marketing.
• [production] remoet by remoet-labs - Job-search and career-discovery skill. Searches tech companies by actual tech stack, stars shortlisted companies, pulls jobs from that curated list, and reads/writes the developer profile. Backed by remote MCP with free tier.
• [beta] zooidfund by Ales375 - Lets an agent discover, evaluate, and donate USDC on Base to humanitarian crowdfunding campaigns at zooid.fund. Public campaign browsing, peer-signal review, and direct wallet-to-creator transfers.
• [production] longbridge by Longbridge Securities - Live financial market skill for US, HK, A-share, and Singapore markets. Real-time quotes, fundamentals, options chain, portfolio positions, institutional data, and news/SEC filings. Requires Longbridge account. 125+ companion skills.
• [production] hermes-mine-bean by MineBean - Autonomous $BEAN mining skill for the MineBean 5×5 grid protocol on Base. Supports zero-cost --no-agent cron mode for unattended automation.
• [beta] invinoveritas by babyblueviper1 - Portable agentskills.io skill + MCP recipe for pay-per-call tools over Bitcoin Lightning / USDC (x402). Governance review gate before irreversible actions, markets intelligence, sandboxed code execution, and agent-to-agent payments.
• [beta] arelay-skills by mmmikael - Deliver reports, Markdown, HTML, images, and PDFs via Agent Relay with optional end-to-end encryption. Works with hosted or self-hosted instances. MIT.
• [beta] tiny.place by TinyHumans - Agent-to-agent social network shipped as a portable skill: claim an @handle identity, get discovered in an open directory, message peers over Signal-protocol end-to-end encryption, and transact in USDC/SOL via x402. One SKILL.md runs on Hermes, OpenClaw, and any agentskills.io-compatible harness. GPL-3.0.
• [beta] plasma-wiki by plasma-ai - Portable skill and Python CLI for hierarchical Markdown knowledge bases — create, index, search, read, and lint via wiki map, wiki search, and wiki read. Keeps project knowledge in Git-tracked plain Markdown with deterministic index maintenance, no hosted memory service or vector DB required. Apache-2.0.

Plugins

• [beta] plur by plur-ai - Shared memory layer for AI agents with open engram format (YAML). Useful for persistent learning patterns in Hermes workflows.
• [experimental] hermes-payguard by nativ3ai - Safe USDC and x402 payment plugin. Lets Hermes send and receive payments with configurable spending limits and approval flows.
• [beta] hermes-web-search-plus by robbyczgw-cla - Multi-provider web search with intelligent routing across Serper, Tavily, Exa, and more. Replaces the built-in search with better result quality and source diversity.
• [beta] hermes-weather-plugin by FahrenheitResearch - Professional-grade weather plugin with NWS model imagery, NEXRAD radar, and meteorological calculations.
• [experimental] hermes-wxtrain-plugin by FahrenheitResearch - ML pipeline plugin for building training datasets from HRRR/GFS/ERA5 weather models. Companion to the weather plugin for building weather ML pipelines.
• [experimental] hermes-plugin-chrome-profiles by anpicasso - Switch browser tools between Chrome profiles via CDP. Useful for multi-account testing or browsing with different saved sessions.
• [experimental] hermes-cloudflare by raulvidis - Cloudflare browser rendering plugin. Headless browsing through Cloudflare's infrastructure instead of local browser automation.
• [beta] evey-bridge-plugin by 42-evey - Claude Code plugin for bridging with Evey (hermes-agent). Lets Claude Code and Hermes share context and hand off tasks between each other.
• [beta] agent-analytics-hermes-plugin by Agent-Analytics - Native Signals dashboard tab for Hermes with read-only multi-project analytics, explicit timeframe, and theme-aware UI.
• [beta] hermes-curator-evolver by pingchesu - Evidence-driven companion to v0.12's built-in Curator. Observes tool/skill/session events into local SQLite, backfills existing ~/.hermes/sessions/*.json history, generates reports and dry-run proposals, supports optional Qwen embedding + bge reranking for candidate ordering, and runs a guarded daily autorun loop that only appends low-risk evidence-backed notes (with backups and rollback manifests). Inspired by SkillClaw but adapted to Hermes-native plugin hooks with explicit model opt-ins and pinned-skill safety.
• [beta] rtk-hermes by ogallotti - Plugin that intercepts shell commands via pre_tool_call and rewrites output through RTK, compressing terminal output before it reaches the LLM context window. 60-90% token reduction on shell commands, 96.6% efficiency across 11M+ tokens processed. Zero config — auto-loads on gateway boot. Real benchmarks: cargo test 90-99%, git log --stat 87%, ls -la 78%.
• [beta] eagle-eye by willingning-coder - 5-layer intelligent skill routing engine. Narrows 50+ installed skills to top-5 candidates before each API call via Hard Triggers → FTS5 BM25 → Synonyms → Embeddings → RRF Fusion. Uses pre_llm_call hook; confidence gate allows "no match" as a valid result.
• [beta] cronalytics by 8bit64k - Cron cost and usage analytics. Hooks on_session_end, stores analytics in local SQLite, surfaces via dashboard tab, CLI, and agent skill for cron health diagnostics. Multi-language (en, zh, zh-TW, es).
• [experimental] hermes-drive-index by gregoryhorn - Private local Google Drive full-text search index. Crawls selected folders into SQLite FTS5 and exposes drive_index_search, drive_index_status, and drive_index_update tools. Apache-2.0.
• [beta] hermes-feishu-zh by OLDBAI213 - Community extension for Chinese Feishu display — replaces English UI labels, stabilizes Markdown via post format, and optionally adds lark-cli-toolbox with 12 Feishu tools. One-click install with backup/rollback.
• [beta] hermes-snow-search by mlinquan - Global memory retrieval layer — cross-session SQLite FTS5 search (~0.1s) across full message history, memory, and facts, with CJK/English auto-routing (trigram vs unicode61 tokenizers) and cross-device queries against a shared state.db. Also on PyPI.
• [beta] hermes-bus-plugin by mlinquan - Message bus integration plugin. Auto-starts bus daemon, registers session as endpoint, injects external messages via pre_llm_call. Provides bus_send, bus_status, and bus_info tools for cross-process agent coordination.
• [beta] hermes-motif by Saurav0989 - Post-tool-call plugin that mines live execution traces for repeated sub-sequences and proposes them as evidence-backed micro-skills. n-gram extraction, arg abstraction, deduplication against existing skills.
• [beta] hermes-eval by Saurav0989 - Skill regression testing and trajectory quality scoring. Catch skill drift before it propagates. Exports quality-filtered trajectories in Atropos RL format.
• [beta] hermes-live-discord-agent-plugin by Capslockb - Full-duplex Discord voice bridge to Gemini Multimodal Live — 40+ function tools, vision, per-user memory, mid-conversation delegation, and transcripts. Pure NumPy codec plumbing with hard-clear barge-in.
• [beta] hermes-dynamic-workflows by lingjiuu - Claude Code-style dynamic workflows for Hermes. Model writes sandboxed workflow scripts orchestrating subagents with agent, parallel, and pipeline — up to 1000 subagents, persisted/resumable runs, live dashboard.
• [beta] custom-dangerous-patterns by scross01 - Add custom terminal command patterns requiring session approval before execution. Extends the standard approve-once/session/always/deny flow with hermes custom-dangerous-patterns CLI management.
• [beta] hermes-ops-kit by redoracle - Operational and security plugin: credential management, provider health checks, route diagnostics, MCP security auditing, key rotation, and usage/cost observability.
• [beta] agentplane-hermes-plugin by Basilisk Labs - External worker-lane plugin letting Hermes Kanban assignees matching agentplane-* spawn AgentPlane supervision instead of a Hermes profile.
• [beta] khwarizmi-hermes-plugin by Moshe-ship - Pre-LLM reliability preflight via pre_llm_call/pre_tool_call: restores implicit references, fails loud on contradictions, and gates tool calls against a constraint-driven allowlist. Part of the Mizan stack.
• [beta] paperclip-hermes-file-access-manager by marospekarik / Ordillect - Kernel-enforced filesystem sandbox for Hermes profiles via Docker bind mounts (not app-level checks): denied paths are unmounted, read-only mounts refuse writes even as root, with inheritance, sensitive-path masking, and Paperclip UI apply. MIT.
• [beta] hermes-usage-hook by chenwei791129 - Appends the provider's own rate-limit windows under every reply (Codex 5h/weekly, MiniMax). Reads Codex OAuth credentials read-only; optional Codex weekly auto-reset is off by default (consumes non-refundable reset credits when enabled) with a privacy-minimised JSONL audit. Prefer hermes plugins install chenwei791129/hermes-usage-hook/plugin over the remote uv run installer. MIT.
• [beta] hermes-yandex-mail by akinfold - Yandex Mail over IMAP (no third-party proxy): list/search/read/flag/move/delete with app-password auth. Read-and-organise only (no SMTP send); BODY.PEEK so reads do not mark seen; YANDEX_MAIL_ACTIONS / YANDEX_MAIL_FOLDERS allowlists drop disallowed tools entirely. High-privilege mailbox access — scope the allowlist tightly. MIT, on PyPI.
• [beta] hermes-yandex-disk by akinfold - Yandex Disk REST API v1 plugin: browse, read, download, write/upload, copy/move, share, and trash. Direct to cloud-api.yandex.net; action and path allowlists control what the agent may touch. High-privilege cloud storage — start read-only. MIT, on PyPI.
• [beta] hermes-yandex-calendar by akinfold - Yandex Calendar over CalDAV: list calendars/events, create/update, RSVP, move, delete. Preserves unmodeled iCalendar properties; YANDEX_CALENDAR_ACTIONS / YANDEX_CALENDAR_CALENDARS allowlists; untrusted server XML via defusedxml. High-privilege calendar write — prefer read-first. MIT, on PyPI.
• [beta] hermes-yandex-search-api by akinfold - Yandex Search API plugin: routes Hermes web_search to Yandex and adds a citation-backed generative search tool. Requires a paid Yandex Cloud API key + folder id. MIT, on PyPI.
• [beta] hermes-plugin-slash-prompts by buihongduc132 - Loads existing CLI-agent prompt files (OpenCode, Claude Code, Codex, Gemini CLI) as Hermes slash commands. Auto-discovers six standard prompt directories, reads YAML-frontmatter/TOML/raw formats, and unifies the different argument-interpolation syntaxes ($ARGUMENTS, $1–$9, {{args}}) in one handler. MIT.
• [experimental] cron-status-alert by Paulo Nascimento - Discord alert plugin for overdue Hermes cron jobs and failed deliveries, with recovery notifications and a /cron-status command. New and unadopted; MIT licensed.
• [experimental] hermes-gate by Observer Protocol - Checks agent spending against signed delegation mandates before wallet actions. The current Hermes binding is bypassable by an agent that avoids the gate, and the repository does not declare a license.
• [experimental] hermes-delegation-gate by Ken Pritchard - Adds human approval before delegate_task can spawn subagents, with a block mode for unattended runs. Requires a recent Hermes plugin-hook implementation; MIT licensed.
• [experimental] Extracto by Rutger Clarke - Self-hosted web extraction plugin using local Trafilatura first and Jina Reader as a fallback for JavaScript-heavy pages. No API key is required; early first release.
• [beta] RoBrain by adelinamart - Self-hosted decision memory plugin that records rejected alternatives and resurfaces their reasons when an agent revisits them. Uses Postgres and can run with a local model; Apache-2.0.
• [experimental] hermes-agent-abacus-ai by ZoniBoy00 - Image-generation plugin for Abacus.AI models. Requires an Abacus.AI API key; MIT licensed and still at its initial public revision.
• [experimental] agentiker plugin suite by agentiker - Three Hermes plugins for AST-aware code intelligence, repository scouting, and plan-following. The linked code-intel plugin is MIT licensed; each plugin installs separately.
• [experimental] hermes-doppler by inLeague - Doppler Secrets Manager plugin with per-profile secret isolation for multi-profile gateway deployments. Root config injects into os.environ; per-profile overlays stay scoped to the profile mechanism, with merge/overwrite modes. Fills the gap for teams on Doppler rather than the natively supported Bitwarden/1Password. MIT.
• [beta] ClawRouter-Hermes by BlockRunAI - Adds ClawRouter as a native model provider: 55+ LLMs from 9 providers through one local proxy, paid per-call in USDC via x402 instead of per-provider API keys, with a model picker, routing profiles, and usage stats. Thin Python adapter over the canonical TypeScript proxy; on PyPI. It manages a real wallet (BIP-39, Base + Solana) and recommends a pipe-to-she

README truncated. Continue reading on GitHub