Gamification Guide

[English] · Gamification-Guide-zh

Gamification Guide

cc-proficiency is a progression system for Claude Code. Sessions earn points, good habits unlock achievements, and your badge shows how much of the tool you actually use.

This guide walks through going from a fresh install to a maxed-out badge.

---

Table of contents

• How the progression system works
• Your first day
• Week one: building momentum
• Power user: intermediate level
• Expert: advanced techniques
• Domain walkthrough
• Achievement unlock guide
• Feature mini-bars: 0 to 100
• The streak system
• Leaderboard
• Depth scoring: the math
• Badge embedding
• Anti-gaming transparency

---

How the progression system works

Three phases

Your badge evolves as you accumulate sessions:

Phase	Sessions	What happens
Calibrating	0--2	Config signals boosted 2x. Your setup efforts count double. Badge shows a setup checklist.
Early	3--9	Config boost drops to 1.5x. Behavior signals start counting at full weight. Domain bars and feature mini-bars appear.
Full	10+	Behavior boosted to 1.15x, config at 1x. This is the steady state -- your real usage patterns drive the score.

In practice: Set up your CLAUDE.md, hooks, and plugins early. During calibration, config alone can push a domain to ~50. After 10 sessions, your actual behavior (investigation chains, prompt structure, agent usage) determines the score.

Two evidence types

Every rule in the engine checks one of two things:

Type	Source	Cap Per Domain
Config	Files on disk: CLAUDE.md, hooks, plugins, .claude/rules/, etc.	25 points max
Behavior	Transcript evidence: tool sequences, prompt patterns, agent usage	75 points max

Config gets you started. Behavior gets you to 100.

Five domains, equal weight

Domain	What it rewards
CC Mastery	Knowing your tool: CLAUDE.md, hooks, plugins, skills, plan mode, rules
Tool & MCP	Smart tool usage: investigation chains, Read-before-Edit, MCP servers
Agentic	Orchestration: subagents, parallel dispatch, worktrees, background tasks
Prompt Craft	Communication: structured requests, code blocks, error traces, file references
Context Mgmt	Memory: CLAUDE.md updates, memory files, sustained multi-session projects

Each domain is 20% of your overall score. You cannot compensate for a weak domain by over-investing in another.

Note: Scores are displayed as percentages of each domain's theoretical maximum, not raw points out of 100. Different domains have different achievable ceilings based on their rule mix and phase scaling, so percentage normalization ensures fair comparison across domains.

---

Your first day

Goal: Install, configure, get into the Calibrating phase.

Step 1: Install and initialize

npm install -g cc-proficiency
cc-proficiency init

This installs the Stop hook, runs an initial analysis, and creates your Gist badge (if gh is authenticated).

Points from init alone: 0. You need config or sessions first.

Step 2: Create your CLAUDE.md

If you do not have a global CLAUDE.md yet:

mkdir -p ~/.claude
cat > ~/.claude/CLAUDE.md << 'EOF'
# My Preferences

## Code Style
- Use TypeScript strict mode
- Prefer functional patterns
- Always add error handling

## Project Conventions
- Tests go in `tests/` mirroring `src/`
- Use vitest for testing
EOF

Rules triggered:

Rule	Points	Domain
ccm-claudemd-exists	+5	CC Mastery

Step 3: Add structure to CLAUDE.md

Use @import directives or ## Section Headers in your CLAUDE.md:

@import .claude/rules/code-style.md
@import .claude/rules/architecture.md

# My Preferences
## Code Style
...
## Architecture
...

Bonus rule triggered:

Rule	Points	Domain
ccm-claudemd-structured	+10	CC Mastery

Step 4: Install a plugin

# In Claude Code settings, add a plugin:
claude plugins add @anthropic/github

Rule	Points	Domain
ccm-plugins-installed	+5	CC Mastery

Step 5: Run your first real session

Just use Claude Code normally. Work on a real task. When the session ends, the hook fires automatically.

cc-proficiency analyze

First session unlocks:

Achievement	Description
First Steps	Complete your first session

After your first day, expect:

• CC Mastery: 15--25 (config signals boosted during calibration)
• Other domains: 0--15 depending on session behavior
• 1 achievement unlocked

---

Week one: building momentum

Goal: Hit 3 sessions (Early phase), fill out config signals, start accumulating behavior evidence.

Day 2--3: Complete 3 sessions

Work normally. At 3 sessions, the phase shifts from Calibrating to Early. Config boost drops from 2x to 1.5x, but behavior signals now count at full weight.

Set up hooks

Add a custom hook to your Claude Code settings. For example, a pre-commit lint check:

{
  "hooks": {
    "PreToolUse:Edit": [
      { "command": "echo 'Editing file: $CC_FILE_PATH'" }
    ],
    "PostToolUse:Bash": [
      { "command": "echo 'Bash command completed'" }
    ]
  }
}

Rule	Points	Domain
ccm-hooks-exist	+10	CC Mastery
ccm-hooks-with-matchers	+15	CC Mastery (if 2+ hooks with tool matchers)

Create a project CLAUDE.md

In your project root:

cat > CLAUDE.md << 'EOF'
# Project: my-app

## Architecture
- src/ -- source code
- tests/ -- vitest tests

## Conventions
- Use kebab-case for filenames
- All exports must have JSDoc
EOF

Rule	Points	Domain
ccm-project-claudemd	+10	CC Mastery

Practice good tool habits

During sessions, let Claude follow this pattern:

1. Search first (Grep or Glob) to find the relevant code
2. Read the file to understand it before changing anything
3. Edit with precision, targeted edits not rewrites

This triggers tool-investigation-chain (+15 per occurrence, up to 3x per session = +45 max).

Week one checklist

Task	Points Impact	Achievement Progress
3 sessions completed	Phase -> Early	Getting Started: 3/10
Global CLAUDE.md with structure	+15 CC Mastery	--
Project CLAUDE.md	+10 CC Mastery	--
Hooks configured with matchers	+25 CC Mastery	--
1 plugin installed	+5 CC Mastery	--
Investigation chains in sessions	+15--45 Tool & MCP	--

Expected scores after Week One:

• CC Mastery: 40--65
• Tool & MCP: 20--40
• Other domains: 5--20

---

Power user: intermediate level

Goal: All domains above 50. Unlock Well-Rounded achievement.

MCP servers

Configure and use MCP servers. These are some of the highest-value signals.

{
  "mcpServers": {
    "github": { "command": "mcp-github", "args": [] },
    "linear": { "command": "mcp-linear", "args": [] }
  }
}

Using MCP tools in sessions:

Rule	Points	Domain
tool-mcp-configured	+10	Tool & MCP
tool-mcp-used	+15	Tool & MCP
tool-mcp-multi-server	+20	Tool & MCP (2+ servers used)

Subagent usage

Start delegating tasks to subagents:

Use an agent to refactor the authentication module while I work on the API routes.

Rule	Points	Domain
agent-used	+5	Agentic
agent-detailed-prompt	+10	Agentic (prompt >100 chars)
agent-multi-types	+15	Agentic (2+ subagent types)
agent-background	+10	Agentic (background agent)

Structured prompts

Instead of:

fix the login bug

Write:

The login endpoint returns 403 for valid tokens. Here is the error:

Error: JWT validation failed at src/auth/validate.ts:42

The relevant file is src/auth/validate.ts. Please:

• Read the file first
• Check the token expiry logic
• Fix the validation without changing the public API

This single prompt can trigger multiple rules:

Rule	Points	Domain
prompt-structured-request	+15	Prompt Craft
prompt-code-block	+15	Prompt Craft
prompt-error-trace	+15	Prompt Craft
prompt-file-reference	+10	Prompt Craft
prompt-context-first	+15	Prompt Craft (code + file + >100 chars)

That is +70 Prompt Craft points from a single prompt.

Memory files

Create memory files so context persists between sessions:

mkdir -p ~/.claude/projects/-your-project/memory
cat > ~/.claude/projects/-your-project/memory/MEMORY.md << 'EOF'
# Project Memory

## Key Decisions
- Migrated from Express to Fastify (2026-02)
- Using Drizzle ORM for database layer

## Current Sprint
- Implementing OAuth2 flow
EOF

Rule	Points	Domain
ctx-memory-exists	+10	Context Mgmt
ctx-memory-active	+15	Context Mgmt (2+ active files)
ctx-memory-diverse	+15	Context Mgmt (3+ memory files)

Power user checklist

Milestone	What to Do
5+ tool types per session	Use Grep, Glob, Read, Edit, Bash in natural workflow
Plan mode	Start a session with claude --plan for deliberate design
3+ skills used	Use slash commands like /commit, /review-pr, /explain
Deep sessions	Have 10+ back-and-forth exchanges (triggers ctx-session-depth)
Iterative refinement	Correct and refine Claude's output ("actually, change X to Y")
Multiple projects	Work across 5+ repos for the Explorer achievement

Expected scores at Power User level:

• All domains: 50--75
• Achievement: Well-Rounded unlocked (all 5 domains at 50+)

---

Expert: advanced techniques

Goal: All domains above 80. Get a Perfect Score (100 in any domain).

Parallel agent dispatch

Send multiple agents to work at the same time:

I need three things done in parallel:
1. Agent: Write unit tests for the payment module
2. Agent: Refactor the email templates to use the new design system
3. Agent: Update the API documentation for the v2 endpoints

Rule	Points	Domain
agent-parallel	+20	Agentic

Worktree isolation

Use worktrees so agents can work on separate branches without conflicts:

Use a worktree to implement the new caching layer on a separate branch.

Rule	Points	Domain
agent-worktree	+15	Agentic

LSP integration

Use language server protocol tools before editing:

Use LSP to find all references to the `AuthService` class, then read the main implementation, then update the constructor.

Rule	Points	Domain
tool-lsp-before-edit	+15	Tool & MCP

Custom skills, agents, and rules

Create project-level definitions:

.claude/
  skills/
    deploy/SKILL.md        # Custom deployment skill
    review/SKILL.md        # Custom code review skill
  agents/
    refactor.md            # Refactoring specialist agent
    test-writer.md         # Test generation agent
  rules/
    architecture.md        # Architecture constraints
    code-style.md          # Style guide
    error-handling.md       # Error handling patterns
    naming.md              # Naming conventions
    testing.md             # Testing standards
    security.md            # Security rules

Rule	Points	Domain
ccm-custom-skills	+10	CC Mastery
ccm-custom-agents	+10	CC Mastery
ccm-rules-files	+10	CC Mastery

The Rules feature bar saturates at 6 rules files, so the above gets you 100% on that mini-bar.

Update your CLAUDE.md

Update your CLAUDE.md during sessions as the project evolves:

Update the CLAUDE.md to add our new database migration conventions.

Rule	Points	Domain
ctx-claudemd-updated	+20	Context Mgmt

Expert checklist

Technique	Rules Triggered	Points
Parallel agents	agent-parallel	+20
Worktree isolation	agent-worktree	+15
Agent resume	agent-resume	+15
LSP before edit	tool-lsp-before-edit	+15
3+ slash commands	prompt-slash-commands	+15
Update CLAUDE.md	ctx-claudemd-updated	+20
Write memory files	ctx-memory-write	+15
5+ plugins	ccm-plugins-diverse	+10

Expected scores at Expert level:

• All domains: 80--100
• Achievements: Master Class unlocked (all 5 at 80+)

---

Domain walkthrough

CC Mastery (max 100)

Config actions (up to 25 pts after cap):

Action	Points	Difficulty
Create ~/.claude/CLAUDE.md	5	Easy
Add @import or structured sections to CLAUDE.md	10	Easy
Create project-level CLAUDE.md	10	Easy
Configure custom hooks	10	Medium
Add 2+ hooks with tool matchers (PreToolUse:Edit)	15	Medium
Install 1 plugin	5	Easy
Install 5+ plugins	10	Medium
Create .claude/rules/ files	10	Medium
Create .claude/agents/ definitions	10	Medium
Create .claude/skills/ definitions	10	Medium
Set effort level to high	5	Easy

Behavior actions (up to 75 pts after cap):

Action	Points	How
Use plan mode	10	Start with claude --plan or switch to plan mode
Invoke a skill	10	Use /commit, /review-pr, or custom skills
Use 3+ different skills	15	Use diverse slash commands in one session
Hooks actively firing	10	Configure hooks that trigger on tool use

Total possible raw points: ~155. After the 25 config cap + 75 behavior cap = 100 max.

Tool & MCP (max 100)

Action	Points	How
Use 3+ tool types	5	Natural workflow: Grep + Read + Edit
Use Grep or Glob	5	Search before editing
Investigation chain (Grep/Glob -> Read -> Edit)	15 x3	The gold standard workflow
Read before Edit	10	Always read a file before modifying
5+ tool types in session	10	Add Bash and Glob to your repertoire
Selective Edit	10	Use targeted old_string replacements
MCP servers configured	10	Add MCP config
Use MCP tools	15	Actually call MCP tools in sessions
2+ MCP servers used	20	Use tools from different MCP providers
LSP before edit	15	Navigate with LSP, then edit

Anti-pattern to avoid: Firing 5+ tools in parallel and having >50% fail. That costs -10 points.

Agentic (max 100)

Action	Points	How
Use Agent tool	5	Delegate any subtask to an agent
2+ subagent types	15	Use different agent specializations
Detailed agent prompt (>100 chars)	10	Give agents clear, specific instructions
Resume an agent	15	Continue work from a prior agent run
Parallel agents (same timestamp)	20	Dispatch multiple agents simultaneously
Background agent	10	Use run_in_background for async work
Worktree isolation	15	Isolate agent work in git worktrees
Task management	10	Use TaskCreate/TaskUpdate for coordination
Parallel tool execution	10	Any 2+ tools dispatched simultaneously
3+ sessions in project	10	Sustained multi-session work

Prompt Craft (max 100)

Action	Points	How
Meaningful prompts (avg >20 chars)	5	Write real sentences, not single words
Structured request (markdown list)	15	Use - item or 1. step formatting
Include code block	15	Paste relevant code in triple backticks
Include error trace	15	Paste the actual error message
Reference file path	10	Mention src/auth/validate.ts explicitly
Iterative refinement	10 x3	Follow up with "actually, change X"
3+ slash commands	15	Use /commit, /review-pr, /explain, etc.
Context-first prompt	15	Code block + file path + detailed request (>100 chars)

Anti-pattern to avoid: Walls of text over 2000 characters with no code, lists, or structure. That costs -5 points.

Context Mgmt (max 100)

Action	Points	How
Memory files configured	10	Create files in .claude/projects/.../memory/
2+ active memory files	15	Keep memory files recently updated
3+ memory files total	15	Maintain separate memory for different concerns
Work on 2+ projects	5	Use Claude Code across different repositories
3+ sessions in a project	15	Sustained engagement, not one-off usage
Read CLAUDE.md in session	10	Consult guidelines during work
Update CLAUDE.md in session	20	Evolve project knowledge actively
Write to memory files	15	Persist decisions and context for future sessions
3+ hook events in session	10	Active hook integration
10+ prompts in session	10	Deep, engaged sessions

---

Achievement unlock guide

Session milestones

Achievement	Unlock	Strategy
First Steps	1 session	Just use Claude Code once
Getting Started	10 sessions	Use Claude Code daily for 2 weeks
Century Club	100 sessions	Sustained daily usage over 3--4 months

Hours milestones

Achievement	Unlock	Strategy
10 Hour Mark	10 hours total	~1 hour/day for 10 days
Centurion	100 hours total	~1 hour/day for 3+ months

Score milestones

Achievement	Unlock	Strategy
Perfect Score	Any domain = 100	Focus on one domain -- Prompt Craft is often easiest
Well-Rounded	All 5 domains >= 50	Follow the Power User checklist above
Master Class	All 5 domains >= 80	Follow the Expert checklist; requires breadth

Fastest path to Perfect Score: Prompt Craft. One good session with code blocks, error traces, file references, iterative refinement, and slash commands can max this domain.

Fastest path to Well-Rounded: Fix your weakest domain first. Run cc-proficiency analyze to see which one needs work.

Streak milestones

Achievement	Unlock	Strategy
Week Warrior	7-day streak	Use Claude Code every day for a week -- even a short session counts
Monthly Grinder	30-day streak	Build the daily habit. Weekends included.

Feature milestones

Achievement	Unlock	Strategy
Explorer	5+ projects	Use Claude Code across different repositories
Agent Master	Agents feature score > 0	Use subagents for any delegated task
MCP Explorer	2+ MCP servers used	Configure and use multiple MCP integrations
Night Owl	Have at least one active session day	Unlocks automatically after your first session

Community

Achievement	Unlock	Strategy
Community	Join leaderboard	Run cc-proficiency share

---

Feature mini-bars: 0 to 100

Below your domain bars, the badge shows 8 feature mini-bars. These use depth-based scoring, not simple on/off checks. Here is how to progress each one.

Hooks

Score Range	What Gets You There
0	No hooks configured
~15	Custom hooks configured (binary(hasCustomHooks, 15))
~30	Hooks + 2 tool matchers (binary(matcherCount >= 2, 15))
~50	Config maxed + ~10 total hook fires across sessions
~75	~40 hook fires + 4 unique hooks
100	~100 hook fires + 8 unique hooks

Formula: config(30 max) + logCurve(totalFires, 20, 25) + cappedRatio(uniqueHooks, 8) * 0.3

Plugins

Score Range	What Gets You There
0	No plugins
~25	3 plugins installed
~50	6 plugins installed
~75	9 plugins + some actual usage
100	12+ plugins installed + 5 actively used

Formula: cappedRatio(pluginCount, 12) + cappedRatio(pluginsUsed, 5) * 0.3

Plugins are unique: install count IS depth. Using them adds a bonus.

Skills

Score Range	What Gets You There
0	No custom skills, no skill invocations
~15	Custom skill definitions in .claude/skills/
~30	Config + ~5 skill invocations
~60	~20 skill calls + 4 unique skills
100	~80 skill calls + 8 unique skills

Formula: binary(hasCustomSkills, 15) + logCurve(totalCalls, 15, 25) + cappedRatio(uniqueSkills, 8) * 0.35

MCP

Score Range	What Gets You There
0	No MCP config
~15	MCP servers configured
~30	Config + ~5 MCP tool calls
~60	~30 MCP calls + 2 servers used
100	~100 MCP calls + 4 servers used

Formula: binary(hasMcpServers, 15) + logCurve(mcpCalls, 20, 25) + cappedRatio(serverCount, 4) * 0.35

Agents

Score Range	What Gets You There
0	Never used agents
~15	Custom agent definitions in .claude/agents/
~30	Config + ~3 agent calls
~60	~15 agent calls + 3 agent types
100	~50 agent calls + 5 agent types

Formula: binary(hasCustomAgents, 15) + logCurve(agentCalls, 10, 25) + cappedRatio(agentTypes, 5) * 0.35

Plan

Score Range	What Gets You There
0	Never used plan mode
~25	~1 plan-mode prompt
~50	~4 plan-mode prompts
~75	~8 plan-mode prompts
100	~15+ plan-mode prompts

Formula: logCurve(planModePrompts, 2, 40)

Plan mode has no config component -- it is purely behavioral.

Memory

Score Range	What Gets You There
0	No memory files
~20	Memory files exist (binary(hasMemoryFiles, 20))
~50	Memory exists + 2--3 files
~70	3 files + 2 actively maintained
100	5+ memory files + 2+ actively maintained (recent mtime)

Formula: binary(hasMemoryFiles, 20) + cappedRatio(fileCount, 5) * 0.5 + binary(activeCount >= 2, 30)

Rules

Score Range	What Gets You There
0	No .claude/rules/ files
~17	1 rules file
~33	2 rules files
~50	3 rules files
~83	5 rules files
100	6+ rules files

Formula: cappedRatio(rulesFileCount, 6)

Rules is the simplest feature bar -- purely a count of rule files. Create 6 well-organized rules files and you hit 100%.

---

The streak system

How streaks work

A streak is a run of consecutive days with at least one Claude Code session. The system tracks:

• Current streak -- your active consecutive-day count
• Longest streak -- your all-time best (never decreases)
• Last active date -- the most recent day you used Claude Code

Streaks are computed from unique UTC dates in your session history (last 90 days). A day counts if you had at least one session that day.

Maintaining your streak

Scenario	Effect
Use Claude Code today	Current streak += 1 (if yesterday was also active)
Skip one day	Current streak resets to 0 (checked on next session)
Skip a day, then resume	New streak starts at 1. Longest streak unchanged

Note: Streaks are recalculated each time scores are updated. The current streak reflects the trailing consecutive run in your session history.

Tips:

• Even a 30-second session counts as an active day
• The streak uses UTC dates -- know your timezone offset
• Your longest streak can never decrease, so a broken streak does not erase your record
• Streak data syncs to your Gist, so it persists across machines

Streak achievements

Streak Length	Achievement	Icon
7 days	Week Warrior	Fire
30 days	Monthly Grinder	Muscle

---

Leaderboard

How to join

cc-proficiency share

This command:

1. Shows a privacy summary of exactly what will be shared
2. Creates a public Gist with your profile (scores, streak, achievement count)
3. Opens a GitHub Issue to register you in the leaderboard
4. Unlocks the Community achievement

What others can see

Shared	NOT Shared
GitHub username	Session details
5 domain scores (0--100 each)	Project names
Current and longest streak	File paths or code
Achievement count	Prompt content
Total sessions and hours	Which specific tools you used
Member since date	Private Gist content

Viewing rankings

# Default: top 20 by average score
cc-proficiency leaderboard

# Sort by specific domain
cc-proficiency leaderboard --sort=agentic

# Sort by hours
cc-proficiency leaderboard --sort=hours --limit=10

Keeping your profile current

Your profile updates automatically after each session (via the Stop hook). You can also force an update:

cc-proficiency push

Leaving

cc-proficiency share --remove

This removes your entry from the registry and deletes the public Gist. Your private data and badge are untouched.

---

Depth scoring: the math

The feature mini-bars use three mathematical curves. Here is what each one does.

Logarithmic curve (logCurve)

score = min(100, k * log(1 + value / scale))

The first few uses earn the most points, and each subsequent use earns less. Going from 0 to 10 MCP calls might gain you 40 points; going from 90 to 100 calls gains about 2 more. This prevents gaming by repetition.

Capped ratio (cappedRatio)

score = min(100, (value / cap) * 100)

Linear progression up to a cap, then flat. If the cap is 6 rules files, then 3 files = 50%, 6 files = 100%, 12 files = still 100%. Some features have a natural ceiling where more is not meaningfully better.

Binary (binary)

score = present ? value : 0

You either have it or you don't. Custom hooks configured = 15 points. Configuring them twice is still 15.

How they combine

Most features use a pattern like:

total = config_component + depth_component

Where:

• config_component = binary checks for whether files/settings exist (maxes around 15--30)
• depth_component = logCurve on usage counts + cappedRatio on variety (drives the remaining 70--85)

Installing something gets you started. Actually using it is what drives the score up.

---

Badge embedding

In your GitHub README

After running cc-proficiency init with gh authenticated, add to your README:

<!-- Static badge -->
![CC Proficiency](https://gist.githubusercontent.com/<username>/<gist-id>/raw/cc-proficiency.svg)

<!-- Animated badge (bars fill on load, numbers fade in) -->
![CC Proficiency](https://gist.githubusercontent.com/<username>/<gist-id>/raw/cc-proficiency-animated.svg)

The init command prints the exact URLs to copy. Both SVGs update in place after every session; the URLs never change.

Local badge

If you prefer not to use Gist:

cc-proficiency badge --output ./my-badge.svg
cc-proficiency badge --animated --output ./my-badge-animated.svg

What the badge shows

Element	Description
Username	Your GitHub username
5 domain bars	Color-coded progress bars (0--100) with confidence dots
8 feature mini-bars	Heatmap row showing depth of each feature area
Achievement count	How many of the 15 achievements you have unlocked
Total hours	Cumulative time in Claude Code
Streak	Current consecutive-day streak (if on leaderboard)

---

Anti-gaming transparency

The scoring engine is designed so that gaming does not work. Here is why.

Per-session caps

Every rule has a maxPerSession limit. Most rules cap at 1 fire per session. The investigation chain caps at 3. This means:

• Running Grep -> Read -> Edit 100 times in one session scores the same as doing it 3 times
• Invoking the same skill 50 times counts as 1 skill invocation
• A single "mega session" cannot substitute for regular usage over time

Bucket caps

Each domain has hard limits:

Bucket	Cap
Config evidence	25 points max per domain
Behavior evidence	75 points max per domain
Anti-pattern penalty	-15 points max per domain

You cannot max a domain purely through configuration (the ceiling is 25 during full phase, or ~50 during calibration with the 2x boost). You need real behavioral evidence.

Logarithmic saturation

Feature depth scores use log curves with diminishing returns:

MCP Calls	Approximate Score Gain
0 -> 5	+0 -> ~20
5 -> 20	~20 -> ~40
20 -> 50	~40 -> ~55
50 -> 100	~55 -> ~65

Doubling your usage does not double your score. Varied usage beats repetitive grinding.

Anti-pattern deductions

Bad habits cost points:

Pattern	Penalty	What It Detects
Shotgun parallel calls	-10	5+ tools fired simultaneously with >50% error rate
Wall of text	-5	Prompt >2000 chars with no code, lists, or structure

These deductions are capped at -15 per domain, so a bad session cannot destroy your score. But they do prevent inflated scores from low-quality usage.

Recency weighting

Sessions are weighted by age:

Age	Weight
Last 7 days	1.0x
8--30 days	0.7x
31--90 days	0.4x
90+ days	0.2x

This means your score reflects your current habits, not historical peaks. You cannot coast on a great month from three months ago.

The short version

The most effective way to improve your score is to use Claude Code well: search before editing, structure your prompts, use agents for parallel work, keep your project context up to date, and configure your environment. No shortcut.

---

Quick reference

Phases:         Calibrating (0-2) → Early (3-9) → Full (10+)
Domains:        CC Mastery · Tool & MCP · Agentic · Prompt Craft · Context Mgmt
Features:       Hooks · Plugins · Skills · MCP · Agents · Plan · Memory · Rules
Achievements:   15 total across sessions, hours, scores, streaks, features, community
Caps:           Config 25 | Behavior 75 | Penalty -15 (per domain)

Commands:
  cc-proficiency init          # Setup hook + first analysis
  cc-proficiency analyze       # Incremental analysis
  cc-proficiency analyze --full # Full re-analysis
  cc-proficiency achievements  # View achievement progress
  cc-proficiency leaderboard   # View community rankings
  cc-proficiency share         # Join the leaderboard
  cc-proficiency explain       # Explain your current scores
  cc-proficiency push          # Force-push badge to Gist