10 Best Hermes Agent Skills

If you only install one skill from this list, make it this one. The pack ships 750+ individual skills mapped directly to the MITRE ATT&CK framework, covering tactics from initial access through exfiltration. SOC teams use it for automated threat analysis, incident triage, and detection rule generation. The skills are sourced from a security-verified GitHub repository and undergo community review before each release.

The reason this beats everything else on the list is sheer coverage. You point Hermes at an indicator of compromise and the relevant ATT&CK skills activate automatically, with no manual selection. It turns a general-purpose agent into a security analyst that already knows the playbook.

Still, 750+ skills is a lot of context for the agent to carry. Initial load time is noticeable on slower machines, and you will want to pin the subset you actually use rather than loading the entire pack every session. The pack also assumes you have some SOC familiarity. If you don't know what a T1059 technique is, the skill output won't mean much to you without additional context.

Install: hermes skills install github:anthropics/anthropic-cybersecurity-skills

This is the one that surprised us. It's not a community skill. It's self-generated. Users running Hermes on a VPS with multiple platform gateways (Telegram, Discord, Slack) keep asking the agent the same thing: "what's running, is anything broken?" After five or more of those checks, Hermes creates mission-control on its own. The pattern is so universal that unrelated users end up with nearly identical versions of this skill.

It surfaces active gateways, memory usage, skill execution stats, and error rates in a single dashboard view. Think of it as htop for your Hermes fleet. The fact that it's self-generated means it adapts to your specific instance layout rather than following a one-size-fits-all template.

The downside: since every user's version is slightly different, there's no standardized output format. If you're feeding mission-control data into external monitoring (Grafana, Datadog), you'll need to normalize the output yourself.

This one solves the biggest pain point of using any AI coding agent: it forgets what you were working on. kanban-orchestrator creates persistent kanban boards backed by Hermes memory files, auto-categorizes tasks from conversation context, and tracks them across sessions. Users report 40% faster task completion with structured kanban workflows because the agent stops losing track of where things stand between sessions.

The skill ranks above git-workflow and research-synthesizer because it addresses a more fundamental problem. You can work around bad commit messages; you can't work around an agent that doesn't remember your project has 12 open tasks.

The catch: all the kanban data lives in Hermes memory files on a single machine. There's no cloud sync, no export to Jira or Linear, and if your memory files get corrupted the board is gone. For personal task tracking it's great; for team-wide project management you'll outgrow it quickly.

Automatically reviews and cleans MEMORY.md and USER.md files by removing stale entries, consolidating duplicates, and reorganizing content by topic. It runs as scheduled maintenance at a configurable interval. Without it, memory bloat gradually degrades agent performance as context windows fill with outdated information from weeks-old conversations. Anyone who has watched an agent get slower and less coherent over a month of heavy use knows why this matters.

Fair warning: the skill is aggressive by default. It sometimes removes entries you actually wanted to keep because it scored them as stale. Pin important memory entries before running it, or adjust the staleness threshold if you find yourself re-teaching the agent things it used to know.

A meta-skill: a skill for building better skills. It provides templates, a testing framework, and a quality scoring system that evaluates how well your custom skills follow Hermes conventions. If you're writing skills for a team, this standardizes output quality before distribution.

The quality scoring is opinionated: it sometimes flags perfectly functional skills as low-quality because they don't follow its preferred markdown structure. You'll need to decide whether to conform or ignore the score. Still, the templates alone save enough boilerplate time to justify the install.

Handles git branching, committing, and PR creation. Context-aware: it reads project structure and matches existing naming conventions instead of imposing its own. Commit messages come from diff analysis, so they reflect what actually changed rather than a generic "update files" placeholder. Useful, but not transformative. Most coding agents already handle basic git, and this skill mostly smooths out the rough edges rather than adding new capabilities.

Chains web search, content extraction, synthesis, and citation into a single workflow. Uses Firecrawl (via Nous Portal) or built-in web tools for scraping, then outputs structured research briefs with citation chains. Solid for turning a broad question into a sourced brief without opening dozens of tabs, but the citation accuracy depends entirely on the underlying scraping quality. Without a Firecrawl subscription, the fallback web tools produce noticeably rougher results.

Manages Docker container lifecycle for the Docker terminal backend: creating, monitoring, and cleaning up execution containers with resource limits and networking. If you run Hermes in sandboxed mode (code execution inside disposable containers), this prevents resource exhaustion. If you don't use Docker backend, this skill does nothing for you. It's ranked #8 precisely because its audience is narrow, but for that audience it's essential.

Generates periodic activity reports (daily, weekly, monthly) from Hermes logs, summarizing tasks completed, skills used, and errors hit. Output goes to markdown or through your messaging gateway. Answers the question "what did the agent actually do this week?" without digging through raw logs. Useful for accountability, though the reports tend to be verbose. Expect to skim rather than read them word for word.

Routes messages between Telegram, Discord, and Slack gateways while preserving threading and user attribution. Niche, but if your team is split across messaging platforms and you want Hermes to relay conversations, it works. The forwarding is one-directional per rule, so setting up bidirectional sync requires configuring two separate routing rules. It made the list because cross-platform teams keep asking for it, not because it's polished.