GitHub Repos intermediate 4 min read Mar 16, 2026 · Updated Mar 19, 2026
Public Preview Sign in free for the full digest →

The Claude Code plugin with 41k stars that keeps your AI coding for 2+ hours straight

“Claude Code can work autonomously for 2+ hours straight without drifting — if you install this 41k-star plugin first.”

The Claude Code plugin with 41k stars that keeps your AI coding for 2+ hours straight
9 Views
2 Likes
2 Bookmarks
Source · github.com

“"I can't recommend this post strongly enough. The way Jesse is using these tools is wildly more ambitious than most other people." — Simon Willison (@simonw), Hacker News, October 2025”

You know that feeling when you ask Claude Code or Cursor to build something non-trivial, it produces a blob of code with no tests, skips the edge cases you specifically mentioned, and then asks you 'what should I do next?' after 10 minutes of work — instead of just continuing? Before Superpowers, getting an AI to behave like a senior engineer required you to manually specify the entire process in your prompt every single time: plan first, write tests, use TDD, spawn subagents for parallel tasks, do code review. The AI ignored half of it anyway because the instructions were just suggestions, and a clever enough AI will always find a rationalization for skipping the uncomfortable parts. Now: you install Superpowers once, and the agent reads mandatory SKILL.md files before every task — it can't skip them, it can't rationalize around them, and it keeps building for hours without deviating.

claude-codeagentic-codingaidevtoolsopen-sourcetddsoftware-engineering

Superpowers installs into Claude Code (or Cursor, Codex, OpenCode) as a plugin and injects a session-start hook that tells the agent: 'You have Superpowers — RIGHT NOW, read this SKILL.md file.' Each SKILL.md is a structured markdown document with YAML frontmatter that defines when a skill applies, what the agent must do step-by-step, and explicit counter-arguments to the rationalizations the AI would normally use to skip the process. Before coding any feature, the agent reads the relevant skill (TDD, systematic debugging, subagent delegation, etc.) and follows it as a mandatory workflow rather than a suggestion. The clever design detail: each skill was tested by running AI subagents through pressure scenarios without the skill, documenting exactly which rationalizations they used to skip steps, and then writing the skill to explicitly forbid those specific rationalizations — so the instructions are hardened against known failure modes, not generic best-practice advice. The result is a 7-phase pipeline: Socratic brainstorming → spec writing → task breakdown → parallel subagent development → spec compliance review → code quality review → integration — all automated, all mandatory.

01
Mandatory skill loading before every task — the agent reads the relevant SKILL.md before starting work, not as an optional reference but as a required pre-task step enforced by the session hook, which means planning and TDD happen even whe...
02
Subagent-driven parallel development — complex tasks get broken into parallel workstreams dispatched to specialist subagents, with two-stage review (spec compliance first, then code quality), enabling Claude to work autonomously for 2+ hou...
03
TDD enforcement with automatic code deletion — if Claude writes implementation code before writing failing tests, Superpowers' TDD skill tells it to delete the code and start over; the skill explicitly forbids the rationalization 'I alread...
04
Systematic 4-phase debugging — instead of the AI guessing at the most obvious fix, the systematic-debugging skill enforces root-cause tracing, defense-in-depth analysis, and condition-based waiting before any fix is proposed, eliminating t...
05
Skills are self-improving and writable — you can ask Claude to add a new skill to Superpowers by describing the workflow you want, and it will write the SKILL.md, test it against pressure scenarios, and add it to your personal skills libra...
06
Cross-platform with one install — works natively in Claude Code (official Anthropic marketplace), Cursor, Codex, and OpenCode; platform-specific tool name mappings are handled automatically (a Claude Code 'Read' tool maps to 'read_file' in...
07
Verification-before-completion skill — before marking any task done, the agent runs a structured check that the fix actually works under the conditions described, not just that the obvious test passes; this prevents the 'it worked in my te...
Who it’s for

If you use Claude Code, Cursor, or any CLI-based AI coding agent for anything beyond simple scripts — and you've hit the wall where the agent produces half-baked code, skips tests, or loses direction after 20 minutes — Superpowers is the direct fix. It's particularly valuable for engineers building features that require 30+ minutes of sustained autonomous work, or for teams trying to enforce consistent engineering practices across AI-assisted development. Not worth setting up if you're only using AI for autocomplete suggestions or single-shot code generation where you review every line anyway.

Worth exploring

Yes, and this is production-level useful right now, not weekend-project territory — it's in the official Anthropic plugin marketplace, has 41k stars, and the creator (Jesse Vincent) has been refining the skill-writing methodology for months with real-world testing against agent failure modes. Simon Willison calling it 'wildly more ambitious than most other people' is a strong signal from someone who tracks the entire AI coding space. The one caveat worth knowing: subagent-heavy runs consume significantly more tokens, which matters if you're on Claude's free or standard tier; the payoff is proportional to the size and complexity of what you're building.

Developer playbook
Tech stack, code snippet, sentiment, alternatives.
PM playbook
Adoption angles, user fit, positioning.
CEO playbook
Traction signals, ROI, build vs buy.
Deep-dive insight
Full long-form analysis, no fluff.
Easy mode
Core idea, fast — when you need the gist.
Pro mode
Technical nuance, edge cases, tradeoffs.
Read the full digest
Go beyond the preview

Deep-dive insight, Easy and Pro modes, plus action playbooks — the full breakdown is one tap away.

Underrated tools. Unfiltered takes.

Read the full digest in the Snaplyze app for deep-dive insight, Easy and Pro modes, and the playbooks you can actually use.

Install Snaplyze →