Second Brain note — not a Claude Code project file

This file is a personal knowledge base note about the Claude AI ecosystem. It is excluded from Claude Code’s auto-loading via claudeMdExcludes in ~/.claude/settings.json. Claude is Anthropic’s AI assistant family, available as a web/desktop chat app (Claude Desktop) and a terminal-based agentic coding tool (Claude Code). Models range from the fast Haiku to the balanced Sonnet and the most capable Opus, each optimized for different cost/performance trade-offs.

Claude Code

Ecosystem

Plugins & Extensions

Tool	Description
ccusage	CLI tool for analyzing Claude Code usage from local JSONL files.
claude-mem	Plugin that automatically captures everything Claude does during coding sessions and compresses it with AI.
claude-hud	Plugin that shows in the terminal statusline: context usage, rate limits, tool activity, task progress, git status, and active agents. Updates every ~300ms.
ralph	Autonomous AI agent loop that runs repeatedly until all PRD items are complete.
Understand-Anything	Plugin that turns any codebase into an interactive, queryable knowledge graph.
HolyClaude	Containerized AI development workstation: Claude Code + web UI (CloudCLI) + headless Chromium + 5 AI CLIs + 50+ dev tools in a single Docker image. Supports AMD64/ARM64.
everything-claude-code	Anthropic hackathon winner. Performance optimization system: 28 agents, 50+ skills, 25+ slash commands, hooks, language rules, and MCP configs. Built from 10+ months of daily use.
claudian	Obsidian plugin that integrates Claude Code directly into your vault. Supports agentic file ops, bash commands, inline editing with diff previews, slash commands, skills/subagents, vision, and MCP servers.
notebooklm-ai-plugin	Claude Code skill that brings Google NotebookLM into the terminal: chat with notebook AI, generate 9 artifact types (slides, audio, video, mind maps, quizzes, flashcards, infographics, reports, data tables), manage sources (URLs, YouTube, files, text), and run fast/deep web research.

Skills & Collections

Resource	Description
anthropics/skills	Official Anthropic repository of Agent Skills — self-contained skill folders Claude loads dynamically for specialized tasks.
obsidian-skills	Agent skills (by Obsidian’s CEO) for Claude Code to read, write, and navigate Obsidian vaults.
impeccable	Design skill system for AI coding assistants; teaches typography, color, and UX anti-patterns via 20+ slash commands.
awesome-claude-code-subagents	A collection of 100+ specialized Claude Code subagents covering a wide range of development use cases.
awesome-claude-agents	An orchestrated sub-agent dev team powered by Claude Code.
claude-code-best-practice	Comprehensive guide to Claude Code best practices: subagents, commands, skills, memory systems, and MCP servers.
marketingskills	40+ marketing skills for Claude Code and other AI coding agents: conversion optimization, copywriting, SEO, paid ads, analytics, A/B testing, retention, and growth engineering.

Commands

Slash Commands

Command	Description
`/clear`	Clears the conversation history and resets the context window, starting fresh without losing your work. Aliases: `/reset`, `/new`.
`/compact`	Summarizes the current conversation into a compressed context to free up token space while preserving key information.
`/context`	Displays the current context window usage as a colored grid. Shows optimization suggestions for context-heavy tools, memory bloat, and capacity warnings.
`/reload-plugins`	Reloads all installed plugins without restarting Claude Code.
`/plugin`	Lists installed plugins and manages plugin lifecycle (install, remove, enable, disable).
`/feedback`	Sends the full conversation history to Anthropic (encrypted). Optionally creates a public GitHub issue.
`/rename [name]`	Rename the current session and show the name on the prompt bar. Without a name, auto-generates one from conversation history. Pairs with `claude -n` / `claude --resume <name>`.
`/resume [session]`	Resume a conversation by ID or name, or open an interactive session picker. Alias: `/continue`.
`/rewind`	Rewind the conversation and/or code to a previous point, or summarize from a selected message. Alias: `/checkpoint`.
`/branch [name]`	Create a named branch of the current conversation at this point, to explore a different path. Alias: `/fork`.
`/export [filename]`	Export the current conversation as plain text. Without a filename, opens a dialog to copy to clipboard or save to a file.
`/btw <question>`	Ask a quick side question without adding to the conversation history. Runs even while Claude is working; ephemeral and answer never enters the context.
`/diff`	Open an interactive diff viewer showing uncommitted changes and per-turn diffs. Use arrow keys to switch between git diff and individual Claude turns.
`/cost`	Show token usage statistics for the current session.
`/stats`	Visualize daily usage, session history, streaks, and model preferences.
`/effort [low\|medium\|high\|max\|auto]`	Set model effort level. `low`/`medium`/`high` persist across sessions; `max` is session-only (Opus 4.6). `auto` resets to model default.
`/history`	Search your global conversation history from `~/.claude/history.jsonl`. Shows recent conversations in an easy-to-scan format with date, project, topic, and session ID.

# Start a named session (resumable by name)
claude -n "my-feature-work"
 
# Resume a session by name or ID
claude --resume my-feature-work
 
# Fork a session on resume (new ID, preserves original)
claude --resume my-feature-work --fork-session

Code Review Commands

Three specialized review commands for analyzing code changes:

Command	Usage	Output
`/review`	Posts review to GitHub PR. Optional context: `/review Check the database migration compatibility.`	Posted to PR as comment
`/code-review [effort]`	Reviews local uncommitted changes. Efforts: `low`, `medium`, `high`, `max`, `ultra`. Options: `--comment` (post to PR), `--fix` (apply corrections).	Terminal output or PR comments
`/security-review`	Scans for vulnerabilities, injection attacks, credential leaks. Optional focus: `/security-review Focus on the JWT validation logic.`	Terminal output with severity levels

Quick examples:

# Review changes locally before pushing
/code-review high
 
# Find security issues
/security-review
 
# Auto-fix issues and post to PR
/code-review high --fix --comment
 
# Post final review to GitHub PR
/review Ready for merge — all checks passed.

Statusline

Mix /statusline with ccusage to get real context usage visible in your statusline at all times.

Context thresholds — personal rules:

Usage	Action
> 60%	Start thinking about summarization
> 80%	Compact manually using the Context Management technique below

Context Management

Instead of using /compact (which lets Claude decide what to keep), take manual control of the context by externalizing it to a Markdown file.

Step 1 — Summarize into a file. When the context is getting full, ask:

Let's summarize where we are. What have we accomplished, what's the current state, and what are our next steps? Put this in progress.md

Step 2 — Clear the context.

/clear

Step 3 — Resume with full context.

Let us continue our work @progress.md

You now have full control over the context: it lives in a versioned .md file, survives /clear, and you decide when to roll it over.

Guided Compaction

/compact accepts a free-text instruction telling it what to keep in detail and what to summarize briefly. This beats blind auto-compact, which treats everything equally.

/compact Focus on preserving our authentication refactoring discussion in detail.
The database work is complete — summarize it briefly.

What matters most is preserved verbatim; finished work shrinks to a few lines.

Context References

Use @ to inject file or folder context explicitly — this avoids broad repo searches and keeps token usage low.

Syntax	Result
`@path/to/file.md`	Injects the full content of that specific file
`@path/to/folder/`	Injects a directory listing (filenames only, not content)

Prefer @folder/ over @folder when you only need to know what files exist. Reserve full file injection for when you actually need the content.

# Example: reference a specific file
Fix the bug described in @docs/known-issues.md

# Example: list available files in a folder, not their content
Pick the right migration script from @db/migrations/

Subagent Context Isolation

Each subagent gets its own fresh context window — it does the deep work, returns a summary, and the main context stays clean.

Use an Explore subagent for research-heavy phases so the main agent never accumulates the cost of reading many files:

Use the Explore subagent to analyze the Authentication module in depth.
Return a summary of how it works.

Parallelize across multiple files or topics by spawning one subagent per item:

Let's review our roadmap items.
This task will require reading many files. Use a subagent for the research phase —
one for each roadmap idea in @roadmap_ideas/

When to reach for this pattern:

Exploring a large module or unfamiliar codebase section
Processing many files in parallel (one subagent per file/topic)
Any research phase before writing code — offload it, get the summary, then act

Customs Commands (history)

Description: Search your global conversation history from ~/.claude/history.jsonl in an easy-to-scan format.

Path: ~/.claude/skills/history.md

Code:

Please read my global conversation history from ~/.claude/history.jsonl and present it in an easy-to-scan format.
 
For each conversation, show:
- Entry number
- Date/time (human readable format: "Nov 10, 2025 15:48")
- Project name (just the folder name, not full path)
- First 60-80 characters of the conversation topic
- Session ID (if available)
 
IMPORTANT: Format as a plain text table with properly padded columns (NOT markdown tables).
 
Focus on the most recent 10 conversations in the first table. If there are more, show another 5-7 in an "Additional Recent Conversations" table.
 
At the end, include:
---
Tip: Resume any conversation by running:
- claude --resume <session-id>
- claude --resume (to see an interactive list of recent sessions)

What it does: Displays your recent conversations (up to 10 most recent) with date, project name, conversation topic, and session ID — allowing you to quickly find and resume old sessions.

Plugins

Install anthropics/skills

# Add the official Anthropic skills marketplace
/plugin marketplace add anthropics/skills

Install Understand-Anything

# Install the plugin
/plugin marketplace add Lum1104/Understand-Anything
/plugin install understand-anything

Command	Description
`/understand`	Analyze the current codebase and build a knowledge graph.
`/understand-dashboard`	Open the interactive visualization dashboard.
`/understand-chat [QUERY]`	Ask questions about the codebase.
`/understand-diff`	Analyze the impact of changes (git diffs / PRs).
`/understand-explain [FILE]`	Deep-dive explanation of a specific file or function.
`/understand-onboard`	Generate an onboarding guide for new team members.

Install context-mode

# Install the plugin
/plugin marketplace add mksglu/context-mode
/plugin install context-mode@context-mode

Then run /reload-plugins and verify with /context-mode:ctx-doctor.

Configuration

settings.json Template

{
  "env": {
    "DISABLE_TELEMETRY": "1",
    "DISABLE_ERROR_REPORTING": "1",
    "DISABLE_FEEDBACK_COMMAND": "1",
    "DISABLE_EXTRA_USAGE_COMMAND": "1",
    "CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY": "1",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1",
    "ENABLE_TOOL_SEARCH": "true",
    "CLAUDE_CODE_DISABLE_1M_CONTEXT": "1",
    "CLAUDE_CODE_ATTRIBUTION_HEADER": "1",
    "CLAUDE_CODE_DISABLE_TERMINAL_TITLE": "1",
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1" 
  },
  "feedbackSurveyRate": 0
}

Environment Variables

Telemetry & Privacy

Variable	Documented	Description
`DISABLE_TELEMETRY`	✅	Disables all telemetry collection (metrics, usage data). Setting this also suppresses the feedback survey. The official opt-in counterpart is `CLAUDE_CODE_ENABLE_TELEMETRY=1`.
`DISABLE_ERROR_REPORTING`	✅	Disables automatic error/crash reporting to Anthropic. Already off by default on Bedrock, Vertex, and Foundry — only active on the `claude.ai` API.
`CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC`	✅	Master kill-switch for all non-essential outbound traffic: surveys, telemetry, error reporting, bug reports. Covers the two variables above and more. docs

Feedback & Surveys

Variable	Documented	Description
`CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY`	✅	Disables the in-session “How is Claude doing?” pop-up. Only a numeric rating is collected (no transcripts), but this suppresses it entirely. docs
`feedbackSurveyRate`	✅	Top-level `settings.json` field, not an env var. Float between `0`–`1` (probability of showing the survey). Setting to `0` is equivalent to `CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1`.
`DISABLE_FEEDBACK_COMMAND`	✅	Disables the `/feedback` slash command, which otherwise lets users send their full conversation history to Anthropic (encrypted, optionally creates a public GitHub issue). docs
`DISABLE_EXTRA_USAGE_COMMAND`	⚠️	Disables extra usage/stats commands exposed in the CLI. Community-discovered — not in official docs.

Features & Behavior

Variable	Documented	Description
`ENABLE_TOOL_SEARCH`	✅	Controls deferred tool search. Values: `true` (always on), `false` (off), `auto` (auto-detect by tool count), `auto:N` (custom % threshold). Requires Sonnet 4+ or Opus 4+. docs
`CLAUDE_CODE_DISABLE_1M_CONTEXT`	⚠️	Disables the 1-million-token context window when available. Useful to reduce costs or avoid memory pressure on large sessions. Community-discovered.
`CLAUDE_CODE_ATTRIBUTION_HEADER`	⚠️	Injects an attribution/user-agent header into API requests (useful for identifying Claude Code traffic in API logs). Community-discovered.
`CLAUDE_CODE_DISABLE_TERMINAL_TITLE`	⚠️	Prevents Claude Code from dynamically updating the terminal window/tab title with the current task. Useful when it conflicts with shell title management (tmux, iTerm, etc.). Community-discovered.

MCP Servers

Add an MCP via Streamable HTTP

Use mcp-remote to proxy any Streamable HTTP MCP endpoint into Claude Code’s mcpServers config. This lets you connect to local or remote MCP servers that speak the HTTP transport without needing a native stdio adapter.

Add the following to ~/.claude/settings.json (or the project-level .claude/settings.json):

"mcpServers": {
  "mcp-local": {
    "command": "npx",
    "args": [
      "-y",
      "mcp-remote@latest",
      "http://localhost:8889/mcp"
    ]
  }
}

mcp-remote is a zero-install bridge — npx -y fetches the latest version on first run and caches it. Replace http://localhost:8889/mcp with the actual endpoint URL of your MCP server.

Claude Desktop

Troubleshooting

Reset MCP Cache

Kills lingering mcp-remote processes, wipes the auth cache, clears temp files, and relaunches the app. Run this when OAuth authorization gets stuck or MCP servers fail to connect.

echo "Clearing MCP cache..."
pkill -f "mcp-remote" 2>/dev/null
pkill -f "Claude" 2>/dev/null
rm -rf ~/.mcp-auth
find /tmp -name "*mcp*" -delete 2>/dev/null
sleep 2
echo "Restarting Claude Desktop..."
open -a "Claude"
echo "Done — authorize the OAuth flow when the browser opens"

Personal Second Brain

Most recent

Explorer

Claude