Yeachan-Heo/oh-my-claudecode v3.3.0

v3.3.0 - Persistent Python REPL & Research Orchestration

on GitHub

Highlights

Persistent Python REPL - Variables now survive across python_repl tool calls. No more reloading data between invocations.

Scientist Agent - New specialized agents for data analysis with structured output markers.

/research Command - Orchestrate parallel scientist agents for comprehensive research workflows with optional AUTO mode.

What's New

Persistent Python REPL Tool (python_repl)

• Unix socket-based Python bridge server
• Variables persist across tool invocations
• Actions: execute, reset, get_state, interrupt
• Memory tracking (RSS/VMS) in output
• Session locking with PID verification

Scientist Agent Tier

• scientist (Sonnet) - Standard data analysis
• scientist-low (Haiku) - Quick lookups
• scientist-high (Opus) - Complex analysis
• Structured markers: [FINDING], [STAT:*], [DATA], [LIMITATION]

/research Command

• Multi-stage decomposition (3-7 independent stages)
• Smart model routing (LOW/MEDIUM/HIGH tier)
• Parallel execution with 5 agent concurrency limit
• Cross-validation and verification loop
• AUTO mode for fully autonomous execution
• Session persistence and resume support

Usage

# Variables persist across calls!
python_repl(action="execute", researchSessionID="analysis",
            code="import pandas as pd; df = pd.read_csv('data.csv')")

# df still exists - no need to reload
python_repl(action="execute", researchSessionID="analysis",
            code="print(df.describe())")

/research <goal>           # Standard research with checkpoints
/research AUTO: <goal>     # Fully autonomous until complete

Stats

• 28 Specialized Agents
• 30 Skills
• 6,851 lines added

Full changelog: https://github.com/Yeachan-Heo/oh-my-claudecode/blob/main/CHANGELOG.md