code996

The development roadmap for code996 focuses on Work-Life Balance (WLB) and identifying 'work pits' (unhealthy work environments). Planned features include:

• Refined overtime ratio analysis: More granular metrics for measuring overtime.
• Shit-mountain index (Complexity analysis): Analyzing code complexity to identify high-maintenance codebases, which is often a driver of excessive overtime.

The code996 tool analyzes Git repositories to generate reports on team work intensity, overtime patterns, and the '996' index (working 9 AM to 9 PM, 6 days a week). It uses semantic analysis of git log data and can incorporate user-provided team context (like team size or flexible work policies) to produce a narrative report for managers or teams.

Key Capabilities:

• Detect Git repositories in the current or specified directories.
• Analyze work-hour distributions (24-hour and weekday patterns).
• Identify overtime trends (late nights, weekends, and peak overtime days).
• Compare work intensity across multiple repositories.
• Generate structured JSON data for programmatic use.

code996 automatically detects the environment to choose the most appropriate analysis mode:

• Single Repository Mode: Triggered when running code996 inside a Git repository or pointing to a single repository path.
• Multi-Repository Mode: Triggered when the command is run in a directory containing multiple repositories, or when multiple paths are provided as arguments. This mode allows for one-click scanning and comparison of multiple repositories.

Examples:

code996                    # Single repo (current)
code996 /path/to/repo      # Single repo (specified)
code996 /proj1 /proj2      # Multi-repo mode
code996 /workspace         # Scans all sub-repositories in /workspace

When analyzing multiple repositories, it is best practice to use a shared time range to allow for meaningful comparison.

1. Identify the shared end date (<shared-until>): Run the following command for each repository and take the latest date found:

   git -C "<path>" log --no-merges --format="%cd" --date=format:%Y-%m-%d --max-count=1

2. Identify the shared start date (<shared-since>): Calculate this by going back 365 days from the <shared-until> date.
3. Execute analysis: Pass both -s <shared-since> and -u <shared-until> to every code996 call in the set.

When analyzing commit timestamps, you can enable or disable Chinese Holiday Mode to distinguish between workdays and rest days accurately.

• Enabling Mode: Explicitly use the --cn flag. If not explicitly set, the system checks the holidayMode field in the JSON. If the field is missing, it auto-enables if $\ge 50%$ of commits show a +08:00 timezone offset.
• Behavior without Holiday Mode: Monday–Friday are workdays; Saturday–Sunday are rest days.
• Behavior with Holiday Mode: Uses a reliable holiday-calendar to distinguish between public_holiday and transfer_workday.

Note: If a reliable calendar cannot be obtained, the system will fallback to analyzing 23:00–05:59 samples only, skipping weekday evening and weekend daytime attribution, and will note this limitation in the report.

The tool performs local analysis of Git commit data to estimate work patterns and a '996 Index'.

Data Collection Flow: Git Repo $\rightarrow$ git log collection $\rightarrow$ Daily first commit + Hourly distribution $\rightarrow$ Percentile estimation of start/end work $\rightarrow$ 996 Index calculation $\rightarrow$ Result output.

Key Concepts:

• Work Time Identification: Uses the 10%-20% percentile of recent daily-first-commit samples to estimate start times.
• 996 Index: A calculated value based on the overtime ratio.
• Project Type Identification: Automatically identifies if a project is "corporate" or "open source" based on work regularity and weekend/evening activity.
• Privacy: All analysis is performed locally; no data is uploaded unless you explicitly run the benchmark command.

If you wish to help calibrate the code996 algorithms, you can run a local benchmark. This command analyzes the repository locally and generates an unlabeled code996-benchmark-*.json file. You can then review this file before sending it to the maintainer.

To provide more useful context for a labeled sample, use the following flags:

• --reference-hours: The team's actual or typical standard hours.
• --team-size: Used to study algorithm behavior by team-size segment.
• --schedule: Optional schedule context.
• --label-confidence: Optional confidence context.
• -y / --since / --until: Use a fixed range for reproducible datasets.

# Basic benchmark
npx code996 benchmark

# Labeled benchmark with context
npx code996 benchmark \
  --reference-hours 9.5-18.5 \
  --team-size 20

code996 is an analysis tool that statistics Git project commit time distribution to deduce coding work intensity and identify overtime culture. It requires Node.js 18 or newer.

Run without installation

You can run the tool immediately using npx in the current repository or a parent directory of repositories:

npx -y code996

Run with global installation

To avoid re-downloading the tool every time, install it globally:

npm i -g code996
code996

Viewing Reports

By default, the tool prints a text report to the terminal and saves a visual bilingual report in Downloads/code996-report/[timestamp]_[project_name]/.

To automatically open the generated visual report in your browser, use the --open flag:

npx code996 --open

# Run in current repo or parent directory of repos
npx -y code996