Superpowers
Complete agentic workflow — brainstorm → plan → implement → review → merge
/plugin install superpowers@claude-plugins-official
-
Surfaces requirements gaps before any code is written
when Start here for any non-trivial feature — before touching code
-
/using-git-worktrees gitisolation
Sets up an isolated git worktree for the current feature
when Before starting implementation to keep main branch clean
-
/subagent-driven-development agents
Delegates subtasks to focused subagents
when When a task has clearly separable independent pieces
-
/test-driven-development testing
Enforces RED → GREEN → REFACTOR — won't proceed until tests pass
when Any feature where correctness matters; pairs with /executing-plans
-
/systematic-debugging debugging
Structured failure diagnosis — reads the error, checks assumptions
when When a bug is not obvious from the first read
-
/requesting-code-review review
Prepares work for human or agent review
when Before opening a PR or handing off to a reviewer
-
/verification-before-completion quality
Quality gate: checks the work meets the original plan
when Before marking any task done
compound-engineering-plugin
80% planning + review, 20% execution — each cycle improves the next
/plugin marketplace add EveryInc/compound-engineering-plugin
/plugin install compound-engineering
-
/ce:ideate planningideation
Discovers high-impact improvements through divergent ideation
when When you're not sure what to work on next; before /ce:plan
-
Explores and clarifies requirements
when When the task is vague or has hidden complexity
-
Converts a feature request into a technical roadmap
when After requirements are clear; before any implementation
-
/ce:work implementationgit
Executes the plan with worktree isolation and task tracking
when After /ce:plan is approved — the actual implementation step
-
Multi-agent code review before merging
when After implementation is complete, before /ce:compound or PR
-
/ce:compound memorylearning
Extracts learnings and codifies them for future sessions
when After every completed task — this is the compounding step
Everything Claude Code
150+ skills, 70+ commands, hooks, AgentShield — comprehensive harness optimization
curl -fsSL https://ecc.tools/install.sh | bash
-
/harness-audit setupquality
Scores your current Claude Code configuration
when First thing after install; whenever you change settings or hooks
-
Shows which of the 150+ skills are relevant to your current stack
when After /harness-audit — narrows down what to actually use
-
Structured multi-step planning with explicit deliverables
when For any task that spans more than one file or concept
-
Spawns and coordinates multiple subagents
when When a task has 3+ independent parallel workstreams
-
Runs multiple independent tasks in parallel
when Batch work — e.g. update tests across many files simultaneously
-
Captures a pattern or lesson from the current session
when After solving a non-obvious problem worth remembering
-
Evaluates and promotes a pending learning to permanent memory
when After /learn — before closing a session
-
/quality-gate qualitytesting
Checks build, tests, lint, and types before considering a task done
when Before any commit or PR — replaces manual pre-commit checklist
-
Routes the current task to the appropriate model tier
when For expensive tasks — routes heavy reasoning to Opus, fast tasks to Haiku
-
Reloads context from the previous session summary
when At the start of any session continuing prior work
-
Enters a continuous agent loop with backpressure
when For long autonomous tasks — replaces manual re-prompting
last30days-skill
Research any topic across social, prediction markets, and web — past 30 days
/plugin marketplace add mvanhorn/last30days-skill
/plugin install last30days@last30days-skill
-
/last30days <topic> research
Full research run across Reddit, HN, Polymarket, X, YouTube, Bluesky
when When you need a current pulse on a topic — takes 2–8 min, saves to ~/Documents/Last30Days/
-
/last30days <topic> --quick research
Faster research with less depth — Reddit, HN, Polymarket only
when Quick gut-check on a topic without waiting for the full run
-
/last30days <topic> --days=90 research
Extends the research window to 90 days
when Slower-moving topics where 30 days misses the full picture
-
/last30days A vs B research
Comparative mode — side-by-side table + verdict
when Evaluating two tools, libraries, or approaches with real-world signal
pi-self-learning
Persistent git-backed memory — distilled lessons that travel across machines
npm install -g @mariozechner/pi-coding-agent
pi install npm:pi-self-learning
-
/learning-now memorylearning
Captures a learning from the current session into the daily memory file
when After solving something non-obvious or discovering a project-specific pattern
-
/learning-month memorylearning
Synthesizes the month's daily learnings into a durable CORE.md summary
when End of month — or before a long break — to consolidate and commit
Agent-Reach
Zero-config access to 17 platforms — web, social, YouTube, GitHub
# Tell Claude:
# Help me install Agent Reach:
# https://raw.githubusercontent.com/Panniantong/agent-reach/main/docs/install.md
-
agent-reach doctor setupdebug
Shows which platforms are live and which need configuration
when After install, and whenever a channel starts failing
-
Reads any web page as clean markdown
when When the agent needs to read a URL without a browser MCP
-
yt-dlp --write-subs <url> webresearch
Downloads YouTube subtitles for agent consumption
when Research from video content without transcription APIs
RTK — Rust Token Killer
60–90% token savings by compressing shell output before it enters context
# macOS
brew install rtk
# Linux
curl -fsSL https://raw.githubusercontent.com/rtk-ai/rtk/refs/heads/master/install.sh | sh
# Hook into Claude Code
rtk init -g
-
Installs the global Claude Code hook that pipes all shell output through RTK
when Once after install — sets up automatic compression for every session
-
rtk init -g --gemini setupcost
Also hooks RTK into Gemini CLI
when If you use Gemini CLI alongside Claude Code
-
Confirms RTK is installed and shows the current version
when Verifying the install or checking for updates
jai — Stanford Linux Sandbox
Copy-on-write filesystem overlay — containment for AI agents with no Docker
# Arch
yay -S jai
# From source
git clone https://github.com/stanford-scs/jai.git
cd jai && ./autogen.sh && ./configure && make && sudo make install
sudo systemd-sysusers && jai --init
-
jai claude isolationsetup
Runs Claude Code inside a copy-on-write sandbox (casual mode)
when Day-to-day — any session where you want to contain file writes
-
jai --mode strict claude isolation
Sandbox with separate UID isolation — stronger protection
when Before running an untrusted skill, hook, or long autonomous task
-
jai --mode bare claude isolationdebug
Hidden home directory, medium protection
when Testing new hooks or skills that write to ~/.claude
-
jai --mode casual bash isolationdebug
Drops into a sandboxed shell to inspect what would have changed
when Reviewing side effects after a sandboxed Claude session
No commands match this filter.