Skip to content

CLAUDE.md

Latest commit

 

History

History
179 lines (150 loc) · 9.17 KB

CLAUDE.md

File metadata and controls

179 lines (150 loc) · 9.17 KB

10x-Analyst-Loop v2.1.1 — Supercharged Agentic Analysis Swarm

A Claude Code plugin by 10x.in — multi-agent swarm for end-to-end data analysis automation, powered by Skills v2, /loop scheduling, agent swarms with worktree isolation, MCP integration, webhook notifications, cron tasks, dynamic context injection, multi-source research, bundled statusline with session tracking, and voice-friendly commands.

Works on ALL models: Every instruction is explicit step-by-step so Haiku, Sonnet, and Opus all produce full-quality output.

How It Works

  1. Give any command a project name or folder path (from anywhere on your system)
  2. Plugin auto-copies data files into input/<project-name>/ (the project registry)
  3. All results appear in output/<project-name>/
  4. Agent swarms run in parallel for maximum throughput

Plugin Commands (18 total)

Core Analysis Commands (7)

  • /10x-analyst-loop:analyze <project> [--safe] — Full 5-agent swarm pipeline with parallel fork (Stats + Dashboard simultaneously)
  • /10x-analyst-loop:profile <project> — Data profiling and quality assessment
  • /10x-analyst-loop:clean <project> — Data cleaning (swarm mode for 10+ files)
  • /10x-analyst-loop:query <project> <question> — Ask natural language questions about your data
  • /10x-analyst-loop:visualize <project> <description> — Generate charts and visualizations
  • /10x-analyst-loop:report <project> — Generate a comprehensive Markdown analysis report
  • /10x-analyst-loop:dashboard <project> — Build a standalone interactive HTML dashboard

Power Commands (4)

  • /10x-analyst-loop:watch <project> [interval] [--profile|--dashboard|--full] — Live-monitor with /loop + CronCreate
  • /10x-analyst-loop:batch-analyze <p1> <p2> ... | input/* — Swarm: parallel agents per project (worktree isolated)
  • /10x-analyst-loop:compare <project-a> <project-b> — Parallel worktree profiling + diff report
  • /10x-analyst-loop:research <topic> — Multi-source research across Reddit, X, YouTube, HN, web

System & DevOps Commands (5)

  • /10x-analyst-loop:debug <project> [error] — Auto-diagnose pipeline failures, check deps, search fixes
  • /10x-analyst-loop:schedule <project> <command> <when> — Schedule future/recurring tasks via CronCreate
  • /10x-analyst-loop:notify <project> <webhook-url> — Configure webhook notifications
  • /10x-analyst-loop:simplify <project> — 3-agent swarm code review (Reuse + Quality + Efficiency)
  • /10x-analyst-loop:api <project> [--serve] — Export all artifacts as structured JSON

Integration Commands (2)

  • /10x-analyst-loop:connect <project> [mcp-name|list] — Discover & configure MCP data sources and messaging
  • /10x-analyst-loop:live-update <project> [target] [message] — Send results to connected apps (Slack, Gmail, Discord)

Path Convention

Every command accepts EITHER:

  • A project name already in input/ (e.g., my-sales)
  • A full path to any folder on disk (e.g., C:/Users/data/q1-export)

If you pass a path, the plugin auto-copies data files into input/<folder-name>/ and registers it. The input/ directory is your permanent project registry.

Output Structure (per project)

output/<project-name>/
|-- data-profile.md          # Data structure & quality report
|-- data-profile.json        # Machine-readable profile
|-- cleaning-log.md          # Actions taken during cleaning
|-- insights.json            # Structured findings
|-- report.md                # Full Markdown analysis report
|-- dashboard.html           # Interactive HTML dashboard
|-- api-export.json          # API-ready JSON bundle (if :api used)
|-- watch-log.md             # /loop monitoring log (if :watch used)
|-- .webhook-config.json     # Webhook config (if :notify used)
|-- .mcp-config.json         # MCP connections (if :connect used)
|-- cleaned-data/            # Cleaned versions of input files
+-- charts/                  # Generated PNG visualizations

Agent Swarm Architecture

User Request (text or voice)
     |
     v
+-----------------------+
|     ORCHESTRATOR      |  <-- INDEX.md routes everything
|   (Skill Router)      |
+-----------+-----------+
            |
   +--------+--------+----------+-----------+
   v        v        v          v           v
+------+ +------+ +----------+ +--------+ +----------+
| Data | | Stats| |Visualizer| |Reporter| |Strategist|
|Engine| | ician| |          | |        | |          |
+--+---+ +--+---+ +----+-----+ +---+----+ +----+-----+
   |        |          |            |           |
   v        v          v            v           v
 Clean    EDA &     Charts &    Markdown    Business
 Data     Stats     Dashboard    Report     Actions

Swarm Patterns

Pattern Used By Description
Parallel Fork :analyze Stats + Dashboard run simultaneously
Worktree Swarm :batch-analyze One isolated agent per project
Worktree Pair :compare 2 parallel profiling agents
Chunk Swarm :clean (10+ files) Split files, one agent per chunk
Review Swarm :simplify 3 parallel code review agents
Search Swarm :research 6 parallel WebSearch calls
Loop Swarm :watch CronCreate for recurring execution
MCP Discovery :connect 9+ parallel ToolSearch calls

Pipeline by Command

Command Agents / Pattern
:analyze Data Engineer -> [Stats + Dashboard PARALLEL] -> Charts -> Reporter -> Strategist
:profile Data Engineer
:clean Data Engineer (swarm if 10+ files)
:query Data Engineer -> Statistician -> Strategist
:visualize Data Engineer -> Visualizer
:report Data Engineer -> Statistician -> Reporter -> Strategist
:dashboard Data Engineer -> Visualizer
:watch Data Engineer (CronCreate loop)
:batch-analyze N parallel worktree agents
:compare 2 parallel worktree profilers -> Statistician
:research 6 parallel WebSearches -> Synthesis
:debug Self-contained diagnostician
:schedule CronCreate scheduler
:notify Webhook configurator
:simplify 3 parallel review agents
:api JSON exporter
:connect MCP auto-discovery
:live-update MCP message dispatcher

Claude Code v2 Features Used

Feature How This Plugin Uses It
Skills v2 frontmatter All 18 commands use full SKILL.md with model, context, agent, allowed-tools, hooks
/loop scheduling :watch and :schedule use CronCreate for recurring tasks
Worktree isolation :batch-analyze and :compare spawn agents in isolated worktrees
Agent swarms :analyze, :clean, :simplify, :batch-analyze use parallel subagents
Dynamic context injection !command`` injects live file lists, schemas, state into prompts
$ARGUMENTS substitutions $0, $1, $ARGUMENTS[N] for flexible argument parsing
${CLAUDE_SKILL_DIR} All skills reference scripts with portable relative paths
${CLAUDE_SESSION_ID} All outputs tagged with session ID for tracking
context: fork Complex skills run in forked subagent contexts
Pre/post hooks pre-validate.py blocks on bad input, post-session-log.py tracks history
Webhook hooks post-notify.py POSTs results to configured URLs
MCP integration :connect discovers MCPs, :live-update dispatches via MCPs
Voice-friendly Descriptions match natural speech patterns for voice mode
Model-agnostic Explicit numbered steps with exact code — Haiku runs as reliably as Opus

Model Strategy

Model Best For Why
Opus :analyze, complex :query, :compare Maximum reasoning for deep analysis
Sonnet :report, :dashboard, :batch-analyze, :research, :simplify Balanced quality/speed
Haiku :profile, :clean, :visualize, :watch, :schedule, :notify, :api, :connect Fast, token-efficient

Voice Examples

These work with Claude Code voice mode (hold spacebar, speak, release):

  • "analyze my sales project"
  • "watch inventory data every five minutes with dashboard mode"
  • "compare q1 with q2"
  • "research e-commerce trends"
  • "debug my analysis"
  • "schedule a daily report for my-sales"
  • "connect my Shopify store"
  • "send the results to Slack"
  • "analyze all my projects" (batch-analyze with glob)
  • "simplify the generated code"

Bundled Statusline

The plugin includes a production-grade Claude Code statusline at statusline/. Install with bash statusline/install.sh.

  • Session tracking: Real-time token counts (in/out/total), API call count, cost per call
  • Agent monitoring: Active agent count with visual ● indicators
  • Cron monitoring: Active scheduled job count with ◆ indicators
  • 4-level warnings: At 75%, 85%, 90%, 95% context usage (safe stop at 95%)
  • 5 themes: default, nord, tokyo-night, catppuccin, gruvbox
  • 4 layouts: compact (2 rows), standard (4), full (7), 10x-swarm (8)

10x-Analyst-Loop v2.1.1 | Built by 10x.in | GitHub | All rights reserved.