Command Format by Tool

Your command format depends on your AI tool:

Tool Type	Separator	Example
CLI agents (Claude Code, Gemini, Qwen, Crush, LLXPRT, Augment)	Colon (:)	/clavix:improve
IDE extensions (Cursor, Windsurf, Cline, Kilocode, Roocode, Droid, etc.)	Hyphen (-)	/clavix-improve

Rule of thumb: CLI tools use colon, IDE extensions use hyphen.

---

Command Overview

CLI Commands (4 total)

These run TypeScript code to set up your environment:

Command	Purpose
clavix init	Initialize Clavix in a project
clavix update	Regenerate templates after package update
clavix diagnose	Check installation health
clavix version	Show version

Slash Commands (10 total)

These are markdown templates that AI agents read and execute:

Command	Stage	Purpose
/clavix:improve	Optimize	Smart prompt optimization with auto-depth
/clavix:prd	Document	Guided PRD generation
/clavix:plan	Plan	Task breakdown from PRD
/clavix:implement	Implement	Execute tasks or prompts
/clavix:start	Explore	Conversational requirements gathering
/clavix:summarize	Document	Extract requirements from conversation
/clavix:refine	Refine	Update existing PRD or prompt
/clavix:verify	Verify	Check implementation against checklist
/clavix:review	Review	Criteria-driven PR review for teammates
/clavix:archive	Manage	Archive completed projects

---

CLI Commands

clavix init

Initialize Clavix in the current project.

clavix init

What it does:

• Creates .clavix/ directory with configuration
• Generates provider-specific slash command templates
• Injects documentation blocks into agent files (AGENTS.md, CLAUDE.md, etc.)

Outputs:

• .clavix/config.json - Selected providers and preferences
• .clavix/INSTRUCTIONS.md - Local quick-start guide
• Provider command files (e.g., .claude/commands/clavix/*.md)

Common messages:

• Clavix is already initialized. Reinitialize? - Choose Yes to rebuild
• You must select at least one provider. - Select at least one integration

---

clavix update

Refresh templates and documentation blocks.

clavix update [options]

Flags:

• --docs-only - Update documentation blocks only
• --commands-only - Regenerate command files only
• -f, --force - Overwrite even if no changes detected

Common messages:

• No .clavix directory found. - Run clavix init first
• Unknown provider: <name>, skipping... - Provider not in adapter registry

---

clavix diagnose

Check installation health.

clavix diagnose

Checks performed:

1. Version check - Package version readable
2. Directory check - .clavix/ exists with required subdirectories
3. Configuration check - config.json valid
4. Integration check - Each configured integration working
5. Template check - Package templates intact

Sample output:

Results:
  ✓ Version: v5.10.2
  ✓ .clavix directory OK
  ✓ config.json OK (2 integration(s))
  ✓ Claude Code: 8 command(s)
  ✓ AGENTS.md: Generated
  ✓ Package templates OK

Summary: 6 passed, 0 warnings, 0 failed

---

clavix version

Show installed version.

clavix version

Output: Clavix vX.Y.Z

---

Slash Commands

/clavix:improve

Smart prompt optimization with auto-depth selection.

/clavix:improve [prompt]

What it does:

1. Analyzes prompt quality across 6 dimensions (clarity, efficiency, structure, completeness, actionability, specificity)
2. Auto-selects depth: standard for good prompts, comprehensive for complex ones
3. Applies optimization patterns
4. Saves result to .clavix/outputs/prompts/

Mode: Planning (no code allowed)

When to use: Any prompt that needs improvement - feature requests, bug reports, refactoring tasks.

---

/clavix:prd

Generate PRD through guided Socratic questions.

/clavix:prd [options]

Flags:

• -q, --quick - Shorter question flow
• -p, --project <name> - Name the project directory
• --skip-validation - Skip CLEAR analysis

What it does:

1. Asks strategic questions about your project
2. Detects tech stack from project files
3. Generates comprehensive PRD (full-prd.md)
4. Creates AI-optimized version (quick-prd.md)

Output files:

• .clavix/outputs/<project>/full-prd.md - Detailed PRD
• .clavix/outputs/<project>/quick-prd.md - Condensed version

When to use:

• Building something completely new
• Complex features with multiple requirements
• Greenfield development

When NOT to use: Modifications to existing code - use /clavix:improve instead.

---

/clavix:plan

Generate task breakdown from PRD.

/clavix:plan [options]

Flags:

• -p, --project <name> - Target specific project
• --source <auto|full|quick|mini|prompt> - Choose artifact to use
• --max-tasks <number> - Limit tasks per phase (default: 20)
• -o, --overwrite - Replace existing tasks.md

What it does:

1. Reads PRD artifacts
2. Extracts top-level features (not nested details)
3. Groups into phases
4. Generates tasks.md checklist

Task generation algorithm:

• Top-level numbered lists = Features (tasks)
• Nested bullets = Implementation notes (NOT separate tasks)
• "Technical Stack" sections = Dependencies (NOT tasks)

Output:

• .clavix/outputs/<project>/tasks.md - Phased checklist

---

/clavix:implement

Execute tasks or prompts.

/clavix:implement [options]

Auto-detection:

Without flags, implement auto-detects the source:

1. If tasks.md exists → Task mode (sequential execution)
2. Else if prompts/*.md exists → Prompt mode (most recent)
3. Else → Asks user what to build

Use --tasks or --latest to override auto-detection.

Task mode flags:

• -p, --project <name> - Select PRD project
• --tasks - Force task mode
• --no-git - Skip git auto-commits
• --commit-strategy <per-task|per-5-tasks|per-phase|none> - Auto-commit strategy

Prompt mode flags:

• --latest - Execute most recent saved prompt
• --prompt <id> - Execute specific prompt by ID

Git commit strategies:

Strategy	Behavior
none (default)	No automatic commits
per-task	Commit after every task
per-5-tasks	Commit every 5 completed tasks
per-phase	Commit when phase completes

---

/clavix:start

Start conversational mode for requirements exploration.

/clavix:start

What it does:

• Enters exploration mode
• Asks clarifying questions instead of jumping to implementation
• Tracks conversation topics
• Suggests /clavix:summarize when ready

Mode boundaries:

• Asking clarifying questions
• Helping think through ideas
• Identifying edge cases
• Tracking requirements
• NO code writing

Next steps:

• /clavix:summarize - Extract requirements
• /clavix:prd - Switch to guided PRD

---

/clavix:summarize

Extract structured requirements from conversation.

/clavix:summarize

What it does:

1. Pre-validates conversation coverage
2. Extracts features, constraints, success criteria
3. Creates documentation files
4. Applies optimization patterns
5. Flags unclear areas

Output files:

• .clavix/outputs/<project>/mini-prd.md - Comprehensive requirements
• .clavix/outputs/<project>/original-prompt.md - Raw extraction
• .clavix/outputs/<project>/optimized-prompt.md - Enhanced version

Confidence indicators:

• [HIGH] - Explicitly stated multiple times
• [MEDIUM] - Mentioned once or inferred
• [LOW] - Assumed based on limited info

---

/clavix:refine

Refine existing PRD or prompt through continued discussion.

/clavix:refine

What it does:

1. Detects available PRDs and saved prompts
2. Asks what you want to refine (PRD or prompt)
3. Loads and displays current content
4. Enters conversational refinement mode
5. Updates files with tracked changes

Refinement targets:

• PRD projects in .clavix/outputs/<project>/
• Saved prompts in .clavix/outputs/prompts/

Change tracking:

• [ADDED] - New requirement or feature
• [MODIFIED] - Changed from original
• [REMOVED] - Explicitly removed
• [UNCHANGED] - Kept as-is

PRD refinement options:

• Add new features
• Modify existing features
• Change technical constraints
• Adjust scope (add/remove items)
• Update success criteria

Prompt refinement options:

• Clarify the objective
• Add more context or constraints
• Make it more specific
• Change the approach

After refinement:

• PRDs: Run /clavix:plan to regenerate tasks
• Prompts: Run /clavix:implement --latest to execute

---

/clavix:verify

Perform a spec-driven technical audit of the implementation.

/clavix:verify

What it does:

1. Loads the Spec: Reads tasks.md (Technical Plan) and full-prd.md (Requirements).
2. Reads the Code: Inspects actual source files associated with completed tasks.
3. Performs Gap Analysis: Compares implementation against the plan and requirements.
4. Generates Review: Outputs a structured "Review Board" with categorized issues.

Review Output Categories:

• 🔴 CRITICAL: Architectural violations, broken features, or missing core requirements.
• 🟠 MAJOR: Logic errors, missing edge case handling, or deviation from PRD.
• 🟡 MINOR: Code style, naming, comments, or optimizations.
• ⚪ OUTDATED: The code is correct, but the plan/PRD was wrong.

Example Output:

# Verification Report

## 🔍 Review Comments
| ID | Severity | Location | Issue |
|:--:|:--------:|:---------|:------|
| #1 | 🔴 CRIT | src/auth.ts | Used fetch instead of configured ApiClient |
| #2 | 🟠 MAJOR | src/login.tsx | Missing "Forgot Password" link (Req 3.1) |

Recommended: "Fix #1 and #2"

Note: This is a code audit, not just a test runner. It verifies design compliance.

---

/clavix:review

Perform criteria-driven PR review with structured actionable feedback.

/clavix:review [options]

Flags:

• -b, --branch <name> - Target branch to diff against (default: main/master)
• -c, --criteria <preset> - Use predefined criteria preset
• --quick - Skip questions, use sensible defaults

What it does:

1. Asks for PR context - Which branch or PR to review
2. Gathers review criteria - Security, architecture, standards, or custom focus
3. Collects additional context - Team conventions, specific concerns
4. Analyzes the diff - Reads changed files and surrounding context
5. Generates review report - Structured findings with severity levels
6. Saves report - To .clavix/outputs/reviews/

Review Criteria Presets:

• 🔒 Security - Auth, validation, secrets, XSS/CSRF, injection
• 🏗️ Architecture - Design patterns, SOLID, separation of concerns
• 📏 Standards - Code style, naming, documentation, testing
• ⚡ Performance - Efficiency, caching, query optimization
• 🔄 All-Around - Balanced review across all dimensions

Output Categories:

• 🔴 CRITICAL: Security vulnerabilities, broken functionality
• 🟠 MAJOR: Architecture violations, missing tests for critical paths
• 🟡 MINOR: Code style, naming, minor improvements
• ⚪ SUGGESTION: Nice-to-have improvements

Example Output:

# PR Review Report

**Branch:** `feature/user-auth` → `main`
**Review Criteria:** Security

## 📊 Executive Summary

| Dimension | Rating | Key Finding |
|-----------|--------|-------------|
| Security | 🔴 NEEDS WORK | SQL injection in user search |

## 🔍 Detailed Findings

### 🔴 Critical (Must Fix)

| ID | File | Line | Issue |
|:--:|:-----|:----:|:------|
| C1 | `src/api/users.ts` | 45 | SQL injection: uses string concatenation |

When to use:

• Reviewing a teammate's PR before manual review
• Ensuring PR meets team standards
• Getting structured feedback on code changes

When NOT to use:

• Checking your own implementation against your PRD (use /clavix:verify)
• Code is not yet in a branch/PR

Difference from /clavix:verify:

• verify = checks YOUR implementation against YOUR PRD/tasks
• review = reviews SOMEONE ELSE's PR against configurable criteria

---

/clavix:archive

Archive completed projects.

/clavix:archive [project] [options]

Flags:

• -l, --list - Display archived projects
• -f, --force - Archive even if tasks incomplete
• -r, --restore <project> - Restore from archive

What it does:

• Moves project from .clavix/outputs/ to .clavix/outputs/archive/
• Checks task completion status
• Supports restore for later reference

---

Recommended Workflow

/clavix:start          # Explore requirements
     ↓
/clavix:summarize      # Extract documentation
     ↓
/clavix:prd            # (Alternative) Guided PRD
     ↓
/clavix:refine         # (Optional) Update requirements
     ↓
/clavix:plan           # Generate task breakdown
     ↓
/clavix:implement      # Execute tasks
     ↓
/clavix:verify         # Check implementation
     ↓
/clavix:archive        # Archive when complete

For quick improvements:

/clavix:improve "add user authentication"
     ↓
/clavix:implement --latest
     ↓
/clavix:verify

For iterative refinement:

/clavix:refine         # Update existing PRD/prompt
     ↓
/clavix:plan           # Regenerate tasks (if PRD)
     ↓
/clavix:implement      # Execute updates

For PR reviews (teammate's code):

[Teammate creates PR]
     ↓
/clavix:review         # Analyze PR with criteria
     ↓
[Review report generated]
     ↓
[Manual PR review with guidance]