Superpowers is Jesse Vincent's open-source agentic skills framework that turns Claude Code, Codex, Cursor, Gemini CLI, Factory Droid, OpenCode, and GitHub Copilot CLI into disciplined software engineers rather than overconfident autocomplete. With 206,523 stars and 18,407 forks under an MIT license, Superpowers has become the de facto methodology layer that sits between a raw coding agent and the production codebases people actually ship. The framework is now distributed through the official Anthropic plugin marketplace as `superpowers@claude-plugins-official`, which is the clearest signal yet that the broader ecosystem treats it as a baseline rather than a curiosity. ## Methodology First, Not Tools First The core insight behind Superpowers is that coding agents do not fail because they lack tools — they fail because they jump straight to writing code without a spec, a plan, or a test. Superpowers fixes the workflow rather than the model. As soon as it detects you are building something, the agent steps back and asks what you are actually trying to build, then teases out a spec in chunks short enough for a human to read and digest. Only after you sign off on the design does it produce an implementation plan written for an enthusiastic junior engineer with no project context, then launches subagent-driven execution against that plan. ## The Skills Library Superpowers ships a curated library of automatically triggered skills covering testing, debugging, collaboration, and meta-work. Testing skills enforce the RED-GREEN-REFACTOR cycle and explicitly delete code written before tests. Debugging skills include a four-phase root cause process with defense-in-depth and condition-based waiting. Collaboration skills cover Socratic brainstorming, writing detailed plans, executing in batches with human checkpoints, dispatching parallel subagents, and structured code-review request and response flows. The meta layer includes a writing-skills skill so teams can extend Superpowers with their own house rules. ## Subagent-Driven Development The most distinctive runtime behavior is subagent-driven development. Rather than letting one long-running agent drift across an entire feature, Superpowers dispatches a fresh subagent per task with a two-stage review — first for spec compliance, then for code quality. This pattern is what allows Claude and similar harnesses to work autonomously for hours on a single plan without losing the thread or accumulating low-quality code. Subagent dispatch is paired with git worktrees so each task runs in an isolated workspace on its own branch, with a verified clean test baseline before any code is written. ## Universal Harness Support Unlike most agent-productivity tooling, Superpowers is not Claude-only. The repository ships installation paths for the official Claude plugin marketplace, the Codex CLI and Codex App plugin marketplaces, Factory Droid, Gemini CLI extensions, OpenCode, Cursor's plugin marketplace, and the GitHub Copilot CLI plugin marketplace. The skills themselves are written once and trigger automatically across every harness, which means a team that switches from Claude Code to Codex does not lose its methodology investment. ## Why It Matters The broader argument Superpowers makes is that the bottleneck on coding agents is no longer raw capability — it is process discipline. By enforcing test-driven development, YAGNI, DRY, and evidence-over-claims as automatically triggered skills, Superpowers turns the same underlying model into a measurably more reliable engineer. The 200K-plus stars and the official Anthropic marketplace inclusion suggest that the industry now agrees: in 2026, the value of an agent is dominated by the methodology layer wrapped around it, and Superpowers is the canonical open-source implementation of that layer.