/leaderboard — ELO Ratings

Show the persistent ELO ratings leaderboard. Ratings are updated after each /forge run based on who wins and loses.

How to invoke

/leaderboard              # show all task classes
/leaderboard algorithm    # show only algorithm class
/leaderboard bugfix       # show only bugfix class

Step-by-step workflow

1. Parse optional task class from the user's message.
2. Run the leaderboard formatter:

# All classes
bash "${CLAUDE_PLUGIN_ROOT}/scripts/elo-show.sh"

# Specific class
bash "${CLAUDE_PLUGIN_ROOT}/scripts/elo-show.sh" --task-class "algorithm"

3. Present the output to the user. If no data exists yet, explain that ratings start building after the first /forge run.

Task classes

Ratings are tracked per task class (auto-detected from the forge task description):

Class	Keywords
algorithm	algorithm, sort, search, scoring, math, compute
refactor	refactor, rename, extract, simplify, reorganize
bugfix	fix, bug, error, crash, broken, regression
feature	add, implement, create, build, feature, new
test	test, spec, coverage, assert
docs	doc, readme, comment, changelog
other	(default)

How ELO works

• All buddies start at 1200
• K-factor: 32 (configurable)
• Winner gains points, loser loses points (zero-sum)
• Ratings below 100 are floored
• Provisional status for < 10 games

Configuration

Key	Default	Description
elo_enabled	true	Enable/disable ELO tracking
elo_k_factor	32	ELO K-factor (higher = more volatile)

Rules

• Read-only. This skill only displays ratings, never modifies them.
• Suggest /forge if no data exists yet.
• Keep it brief. Just show the table, no analysis unless asked.