Superpowers for Hermes Agent and ChatGPT

An unofficial, GPT/Hermes-compatible adaptation of the Superpowers AI coding workflow.

Superpowers is a workflow skill pack for making AI coding agents work with more discipline: clarify before building, plan before executing, test behavior changes, debug from evidence, request review, and verify before claiming success.

This repository is an adaptation of the Claude Superpowers skill concepts for Hermes Agent and ChatGPT-style workflows. The goal is behavioral parity, not a verbatim port: the skills are rewritten in original wording and adjusted for explicit routing, Hermes-style skill loading, ChatGPT conversations, and tool environments where automation may vary by session.

Quick Start

1. Start with superpowers-using-superpowers.
2. Ask the agent to use it as the router for coding, planning, debugging, review, git, and skill-writing work.
3. Let the router hand off to the relevant workflow skill, such as superpowers-brainstorming, superpowers-writing-plans, or superpowers-systematic-debugging.
4. Follow each selected skill’s hard gates, protocols, handoffs, and verification checklist.
5. Before claiming work is done, run superpowers-verification-before-completion.

Recommended install path for Hermes:

~/.hermes/skills/superpowers-gpt//SKILL.md

For a first smoke test, ask:

I have a vague app idea; help me shape it using the Superpowers workflow.

Expected route: superpowers-using-superpowers -> superpowers-brainstorming.

For new product, app, or tool ideas, use Idea Workflow first:

https://github.com/AkoliteZA/hermes-agent-idea-workflow

Idea Workflow is designed to produce a concise design doc, implementation spec, and 03-agent-build-handoff.md file before Superpowers begins implementation planning. That gives Superpowers cleaner source material, fewer ambiguous product decisions, and a more efficient path into superpowers-writing-plans.

What This Pack Does

The pack gives an agent a set of reusable operating procedures for common software-development phases:

• Route a request to the right workflow before acting.
• Turn vague ideas into approved design direction.
• Convert approved direction into executable plans.
• Execute plans step by step with verification.
• Use test-driven development for behavior changes.
• Debug from reproduction and root cause instead of guessing.
• Coordinate subagents and parallel work safely.
• Request and receive code review with evidence-based fixes.
• Use git worktrees and finish branches intentionally.
• Verify claims before saying work is done.
• Author new skills with durable structure and pressure tests.

The skills are designed to reduce the agent habits that most often create bad work: premature implementation, hidden assumptions, stale verification, vague plans, unreviewed changes, and broad completion claims that are not supported by fresh evidence.

Contents

This folder contains 14 Superpowers-prefixed skills:

Supporting files:

• Superpowers GPT Skills Index.md is an Obsidian-friendly index.
• MANIFEST - Superpowers GPT Skill Pack.json records the pack metadata and name map.
• Parity Matrix - Superpowers GPT Skill Pack.md maps the original workflow concepts to this GPT/Hermes rewrite.
• README - Superpowers GPT Skill Pack.md is an earlier draft/reference note.

How the Workflow Fits Together

Start with superpowers-using-superpowers. It acts as the router and chooses the right skill for the current phase.

Typical flow:

using-superpowers
  -> brainstorming
  -> writing-plans
  -> executing-plans or subagent-driven-development
  -> requesting-code-review
  -> receiving-code-review, if needed
  -> verification-before-completion
  -> finishing-a-development-branch, if branch cleanup or PR work is needed

Debugging can enter from any point:

failure, bug, broken test, or regression
  -> systematic-debugging
  -> test-driven-development, when a code fix is needed
  -> verification-before-completion

Parallel work is deliberately gated. superpowers-dispatching-parallel-agents is only for independent workstreams with non-overlapping responsibilities and a synthesis step.

Upstream Idea Workflow

Superpowers works best when it receives a clear, build-ready Markdown handoff instead of a vague idea. For the pre-build product/spec phase, use the companion Idea Workflow repository:

https://github.com/AkoliteZA/hermes-agent-idea-workflow

Idea Workflow covers the idea-before-building phase:

rough idea
  -> idea-workflow
  -> design doc
  -> implementation spec
  -> agent build handoff
  -> spec review
  -> Superpowers for GPT
  -> implementation planning
  -> coding
  -> review
  -> verification

Use Idea Workflow when the task still needs product thinking, UX intent, scope boundaries, or a build-ready spec. It should produce 03-agent-build-handoff.md, a single Markdown source of truth that can be fed into Superpowers.

Then use Superpowers to inspect that handoff, write an implementation plan, split work if useful, code, test, review, and verify.

Hermes Installation Model

This folder is currently a reference copy for Obsidian and repository publishing. Each superpowers-*.md file contains the content intended for a Hermes SKILL.md.

Recommended install layout:

~/.hermes/skills/superpowers-gpt//SKILL.md

For example:

~/.hermes/skills/superpowers-gpt/superpowers-using-superpowers/SKILL.md
~/.hermes/skills/superpowers-gpt/superpowers-brainstorming/SKILL.md
~/.hermes/skills/superpowers-gpt/superpowers-writing-plans/SKILL.md

The superpowers- prefix is intentional. It avoids collisions with other Hermes skills that may already be named writing-plans, test-driven-development, systematic-debugging, and so on.

Runtime Differences From Claude-Oriented Skills

Claude-oriented skill packs may rely on runtime hooks, automatic skill invocation, or specific plugin behavior. Hermes Agent and ChatGPT-style environments often require more explicit routing.

This adaptation handles that by:

• Making superpowers-using-superpowers the bootstrap router.
• Defining clear triggers and non-triggers in every skill.
• Translating automatic hooks into explicit gates and handoffs.
• Describing subagent workflows in runtime-neutral terms.
• Requiring fresh verification before completion claims.
• Preserving workflow intent while avoiding runtime-specific assumptions.

Publishing Notes

Before publishing this as a standalone GitHub repository, consider adding:

• A repository-level LICENSE file. The skill frontmatter currently declares MIT.
• An install script or documented copy command if you want one-command Hermes installation.
• A release note explaining that this is an original GPT/Hermes-compatible rewrite inspired by the Claude Superpowers workflow concepts.

Status

Status: prefixed install draft.

The pack has been structured for review and future installation, but it should be smoke-tested in a fresh Hermes session before being described as installed, active, or production-ready.

Suggested smoke tests:

1. Ask for help shaping a vague app idea. Expected route: superpowers-using-superpowers -> superpowers-brainstorming.
2. Ask to implement an approved plan. Expected route: superpowers-executing-plans or superpowers-subagent-driven-development.
3. Present a failing test or bug. Expected route: superpowers-systematic-debugging.
4. Ask whether a task is done. Expected route: superpowers-verification-before-completion.
5. Ask what to do with a completed branch. Expected route: superpowers-finishing-a-development-branch.

Attribution

This pack adapts the Superpowers workflow concepts for Hermes Agent and ChatGPT. It is written as an original implementation of those concepts, with wording and runtime behavior adjusted for this environment. It is not an official Claude, Anthropic, Hermes, or OpenAI project.