Superpowers as inspiration

Research Question

What ideas, patterns, and workflow practices from davidamitchell/superpowers (a fork of obra/superpowers) can be used as inspiration for improving agent tooling in davidamitchell/Latest-developments-, davidamitchell/Agent-Evaluation, davidamitchell/Memory-System, and davidamitchell/Research? Specifically: which superpowers concepts fill genuine gaps in the current davidamitchell/Skills library, and how could those ideas be adapted to the owner's actual workflow?

Findings

Executive Summary

davidamitchell/superpowers is a clean upstream fork of obra/superpowers — a plugin system for interactive IDE agents. The value is not in installing it (incompatible with the GitHub Copilot Agent workflow) but in using it as inspiration: superpowers embeds mature, battle-tested software development disciplines that are absent from davidamitchell/Skills. Three concepts deserve adaptation: test-driven-development (prescriptive TDD with Iron Law and rationalisation countermeasures), systematic-debugging (4-phase root-cause process), and subagent-driven-development (multi-agent coordination pattern). Adapting these into davidamitchell/Skills via PRs to that repo would immediately improve agent quality across all three active target repos. Memory-System does not exist on GitHub and cannot be assessed.

Key Findings

1. Superpowers is a plugin for interactive session agents (Claude Code, Cursor, Codex). Using it as a plugin is not the right path — using its ideas as inspiration is.
2. All three active target repos (Latest-developments-, Agent-Evaluation, Research) already share the davidamitchell/Skills submodule at .github/skills/.
3. Three superpowers concepts fill genuine gaps in davidamitchell/Skills and are worth adapting: test-driven-development (prescriptive TDD enforcement with Iron Law), systematic-debugging (4-phase root-cause process), subagent-driven-development (multi-agent task coordination with context isolation).
4. Adapting these three concepts into davidamitchell/Skills format — opening PRs to that repo — is the recommended path. It requires zero per-repo changes to the target repos.
5. Memory-System does not exist (GitHub 404 as of 2026-03-12). Cannot be assessed.
6. The superpowers philosophy — encode missing disciplines as non-negotiable skills — is itself worth borrowing as a principle when identifying future gaps in davidamitchell/Skills.

Assumptions

• Assumption: Owner's primary agent is GitHub Copilot Agent, not Claude Code. Justification: copilot-instructions.md states "GitHub website or iOS GitHub app only; no local IDE, no Codespaces."
• Assumption: Adapted skills in davidamitchell/Skills will be loadable by GitHub Copilot Agent for coding tasks. Justification: Skills submodule is already in place across all repos and used for coding-adjacent skills (swe, code-review).

Analysis

Superpowers and davidamitchell/Skills serve different delivery models but are compatible at the idea level. Superpowers auto-triggers skills in interactive IDE sessions; Skills are named and invoked in GitHub Copilot Agent sessions. The workflow disciplines in superpowers — TDD enforcement, root-cause debugging, multi-agent coordination — translate naturally into the Skills format and will be just as valuable when invoked explicitly.

The three highest-priority adaptations are chosen on the basis of:

• Coverage gap: not currently in davidamitchell/Skills
• Applicability: all three target repos have Python code under test; multi-agent patterns are directly relevant to Agent-Evaluation
• Quality of the source material: superpowers TDD and debugging skills are comprehensive, include explicit anti-patterns and rationalisation-busting checklists, and have been refined in production sessions

The remaining superpowers skills (brainstorming, writing-plans, executing-plans, using-git-worktrees, finishing-a-development-branch) are lower priority: git-worktree commands don't apply to Copilot Agent; writing-plans and brainstorming are partially covered by swe and strategy-author; executing-plans is redundant given the agent autonomy model. They remain available in the fork as inspiration for future work.

Risks, Gaps, and Uncertainties

• Content drift: Skills adapted from superpowers will diverge from upstream over time. Keeping davidamitchell/superpowers in sync with upstream obra/superpowers (e.g. via a scheduled GitHub Action) ensures the reference stays current for future inspiration.
• Sub-agent capability: If Copilot Agent gains explicit sub-agent dispatch, subagent-driven-development becomes more directly executable. The adapted skill should be written to be useful as guidance today and executable as a workflow tomorrow.
• Memory-System: If created in the future, the same path applies — add davidamitchell/Skills as a .github/skills/ submodule, inheriting any adapted skills automatically.

Open Questions

• Does GitHub Copilot Agent support explicit sub-agent dispatch? This affects how subagent-driven-development is framed in the adaptation.
• Should the superpowers fork be kept in sync with upstream via a scheduled workflow, or is periodic manual update sufficient?
• Of the secondary superpowers skills, is brainstorming worth adapting to complement swe and strategy-author for software design conversations?

sources

• [x] davidamitchell/superpowers — README, plugin.json, skill catalogue, commit history
• [x] davidamitchell/Latest-developments- — repo root, .github/
• [x] davidamitchell/Agent-Evaluation — repo root, .github/
• [x] davidamitchell/Memory-System — 404 Not Found
• [x] davidamitchell/Research — local clone, .github/skills/, copilot-instructions.md
• [x] davidamitchell/Skills — local submodule (via Research repo clone)