Skills Guide
Hermes Agent + Superpowers: A Development Methodology, Installed
Superpowers is an MIT-licensed skills framework built around one claim: agents fail less when they follow a process. Hermes Agent is one of the fourteen runtimes it officially supports, and the install is a single command.
Install
hermes plugins install obra/superpowers --enableThat is the documented command in the Superpowers repository. The --enable flag turns it on at install time rather than leaving it registered but dormant. If Hermes is not installed yet, start with the installation guide.
Blocked: Security scan blocked plugin install: Blocked (dangerous verdict, 213 findings), and --force does not override a dangerous verdict — by design, as covered in our Skills Hub guide. Superpowers is a large pack of skills that legitimately run shell commands and write files, which is exactly the pattern the scanner is built to flag, so a high finding count is not itself evidence of anything malicious. Until upstream resolves it, treat the install as version-dependent: try it, and if it is blocked, do not go looking for a way to defeat the scanner. Install the individual skills you want from the repository instead, the same way the OpenClaw route does it, so each one is reviewed on its own.What changes after you enable it
Superpowers describes itself as a software-development methodology built on composable skills, and the distinction from a normal skill pack is real. A skill pack gives the agent more it can do. Superpowers changes the order in which it does things.
- Test-driven development as a hard sequence: write the test, watch it fail, then implement. The failing run is treated as evidence, not ceremony — a test that passes before the implementation exists is a broken test.
- Four-phase debugging that requires a root cause to be identified before any fix is attempted. This is the skill that most changes day-to-day behaviour, because the default failure mode of every coding agent is to patch the symptom it can see.
- Socratic brainstorming that refines requirements by questioning before code starts.
- Subagent-driven development with a review step built into the hand-off, so work is checked by something other than the process that produced it.
The hand-offs are the product
It is worth being precise about where the value sits, because it is easy to install this, read the skill list, and conclude it is a collection of prompts. The framework is a workflow system: deciding what to build hands off to implementing it under test, which hands off to proving the result before anything is called done. Each hand-off is a checkpoint where a wrong assumption gets caught while it is still cheap.
When to leave it off
The overhead is real and front-loaded. For a one-line config change, a quick lookup, or the kind of chat task most people actually run an always-on agent for, the planning phase costs more than the work it is protecting. Superpowers earns its keep on multi-step engineering work where a wrong turn is expensive to unwind.
Writing your own skills in the same shape
The repository documents its own authoring conventions at skills/writing-skills/SKILL.md, and skills live under the repo's skills/ directory. If you plan to add house skills alongside the framework, matching its conventions keeps the hand-offs working; a skill that ignores the workflow contract just becomes a prompt the agent sometimes reads.
Prefer not to run the plumbing yourself? OpenClaw Launch deploys a managed Hermes Agent in about thirty seconds, and you install Superpowers into it with the command above — subject to the Skills Guard caveat.
Hermes Agent and Superpowers FAQ
Does Superpowers officially support Hermes Agent?
Yes. Hermes Agent is one of the fourteen agents with a documented install path in the Superpowers repository, alongside Claude Code, Codex CLI, Cursor, Gemini CLI, OpenCode and others. The command is hermes plugins install obra/superpowers --enable, though on current versions Hermes' Skills Guard scanner may block it — see the warning below before you rely on it.
What does Superpowers actually add?
A methodology, not a tool. It installs composable skills that push the agent through structured workflows: red-green-refactor TDD where the test must fail first, a four-phase debugging process that demands a root cause before a fix, Socratic brainstorming before implementation, and subagent-driven development with built-in review.
Is it free, and what is the licence?
It is open source under the MIT licence. There is no paid tier to install.
Will it slow my agent down?
On small tasks, yes — deliberately. The framework inserts planning and verification steps before code gets written, which is overhead on a one-line change and a saving on anything you would otherwise have to debug twice. Match it to the work rather than leaving it on for everything.
How is this different from the Skills Hub?
The Skills Hub is the delivery mechanism for individual skills. Superpowers is one opinionated collection installed as a plugin, where the value is in the hand-offs between skills rather than in any single one.