Superpowers
obra/superpowersA 15-skill development methodology that makes coding agents clarify the goal, write a plan, work through it with subagents and verify before calling it done.
A very large Claude Code harness: 311 skills, 71 agents, commands, hooks, memory, rules and a scanner, with adapters for Codex, Cursor and others.
Active and widely used, but it is a sprawling harness with hooks, overlapping skills and a commercial tier, so choose a minimal install profile.
ECC is a harness-level bundle rather than a single skill. The README counts 71 agents, 311 skills and 97 legacy command shims, plus hooks, selectable rules, memory and continuous-learning features, and AgentShield, a scanner for prompts, hooks, MCP configuration, permissions and secrets. The repository contains over 1,000 skill folders because skills are copied into several harness layouts. It installs as the ecc@ecc plugin in Claude Code, through npx ecc-universal setup, as a native Codex plugin, and with more limited adapters for Cursor, OpenCode, Gemini, GitHub Copilot and others.
The skills span TDD, API design, backend patterns, coding standards, benchmarking methodology, research, content and brand voice, competitive analysis, ClickHouse and LLM cost control. Individual skills are mostly instruction files averaging about 9 KB, for example an agent-introspection-debugging skill that walks a failing agent through capture, diagnosis, contained recovery and a report. The runtime parts, hooks, session summaries and learned instincts, run in your harness once installed, so installation is a bigger commitment than adding a skill. The README warns to install only from official channels and not to stack install methods, since this can duplicate hooks and commands. A separate paid GitHub App serves private repositories. Some optional commands call external providers or the Codex CLI.
About 311 skills averaging 9 KB plus agents, hooks and rules, with CI but only a handful of eval files, and uneven depth across skills.
Counted from the repository file tree. How we score depth.
This skill asks for more access than most, or sends data to another service. That is usually a design choice rather than bad intent, but you should know before you install it.
A very large multi-harness kit (the README counts 71 agents, 311 skills and many commands); only the .agents/skills part (about 97 files) was available to us. Two skills pre-approve unrestricted Bash, and the real install is a pinned npm installer that sets up a plugin with hooks, MCP servers and learned-instinct context, none of which we could read. The README says diagnostics are never uploaded automatically and hooks can be disabled with ECC_DISABLED_HOOKS.
npx ecc-universal@2.2.3 setup (npm/pnpm/yarn/bun variants); the installer package, the ecc CLI and the plugin hooks it enables are not in our files.README.md:196The repository (4649 files) on Oct 29, 2026. Not a full review: 4507 of 4649 files were not downloaded (large repository; we reviewed the skill folders); we read 0 of 2 code files; it relies on code outside the repository: npm package ecc-universal 2.2.3 (guided installer and ecc CLI); ECC plugin hooks, scripts and MCP configs (hooks.json, scripts/, .mcp.json are not among the downloaded files); 4,507 of 4,649 repository files were not downloaded; only the .agents/skills folders were available.
NVIDIA SkillSpector static scan: 55 pattern hits (Agent Snooping 18, Data Exfiltration 9, Rogue Agent 8, MCP Rug Pull 8, Tool Misuse 4, Privilege Escalation 2). The 55 hits are mostly documentation: example DELETE routes, .env and token wording in frontend/security guidance, agent config paths in setup notes, unpinned npx/uvx/docker strings inside the mcp-dependency-review skill's own rules, and an ElevenLabs example. The one subprocess hit is the docs-governance script's git call.
A review, not a certification. Skills change after we check them — read the files before installing.
MIT — commercial use allowed. A separate hosted GitHub App, ECC Pro, is paid for private repos.
Not really. It supplies loops, memory and recovery skills, but its own docs frame it as a harness for supervised sessions rather than a verified unattended runner.
A 15-skill development methodology that makes coding agents clarify the goal, write a plan, work through it with subagents and verify before calling it done.
A multi-harness plugin marketplace with 94 plugins, 202 agents, 185 skills and 105 commands for Claude Code, Codex, Cursor, OpenCode, Copilot, Antigravity and Pi.
A Claude Code plugin with 41 skills and many hooks that stores your corrections in SQLite and adds searchable research wikis, quality gates and cost tracking.
A large catalog and npx installer (aitmpl.com) of agents, commands, hooks, MCP configs and skills for Claude Code, much of it collected from other authors.
ECC is a large open-source harness for Claude Code and other agents. It bundles about 311 skills, 71 agents, commands, hooks, rules, memory features and a scanner called AgentShield.
For Claude Code use the ecc@ecc plugin or run npx ecc-universal@2.2.3 setup, which the README says inventories existing installs before changing anything. Do not stack a manual install on top of the plugin.
We saw no obfuscated code in the sampled files, but it installs hooks and can enable MCP servers and external-provider commands. The README says to install only from the official repository, npm packages and plugin slug, and to pick a hooks profile deliberately.
ECC is a broad harness with hundreds of skills, agents and runtime hooks. Superpowers is a smaller, opinionated process for planning and verification. Use Superpowers for a focused workflow and ECC if you want a full configurable harness.