---

name: doc-audit description: Expert in auditing Sphinx/RST documentation for Drift project. Use when validating code examples and removing subjective language before release.

Documentation Audit Skill

Learn how to audit documentation for Drift project - validate code examples and ensure objective language.

When to Use This Skill

• Reviewing documentation before release
• Validating code examples against implementation
• Ensuring objective, factual documentation language
• Checking for outdated or incorrect examples
• Maintaining documentation quality standards

How to Audit Documentation

Overview of Audit Process

A complete documentation audit includes three phases:

1. Code Example Validation - Verify all code examples match actual implementation
2. Subjective Language Detection - Flag subjective or marketing language
3. Cross-Reference Implementation - Check against actual code using serena MCP

Work through each phase systematically for each documentation file.

How to Validate Code Examples

Step 1: Extract Code Examples

Look through documentation files for code blocks:

RST files:

Copy.. code-block:: python

   from drift.cli import cli
   drift.analyze("log.json")

Markdown files:

Copy```python
from drift.cli import cli
drift.analyze("log.json")

Copy
Categorize as:
- Complete examples (should run as-is)
- Snippets (partial code for illustration)

### Step 2: Validate Imports with Serena MCP

Use `mcp__serena__find_symbol` to verify imports:

**Example validation:**
```python
# Documentation says:
from drift.cli import cli

# Verify with serena:
mcp__serena__find_symbol(
    name_path_pattern="cli",
    relative_path="src/drift/cli.py"
)

Common import patterns to check:

• from drift.cli import ...
• from drift.core.analyzer import ...
• from drift.config.loader import ...
• from drift.providers import ...
• from drift.validation import ...

If symbol not found, import path is incorrect.

Step 3: Verify CLI Commands

Check all CLI examples in documentation:

Commands to verify:

Copydrift                    # Basic usage
drift --no-llm           # Programmatic validation only
drift --days 7           # Analyze last 7 days
drift --format json      # JSON output
drift --cwd PATH         # Working directory
drift --rules NAMES      # Specific rules

How to verify:

Copy# Run command to check it exists and works
drift --help

# Check for specific flags
drift --help | grep -- --no-llm
drift --help | grep -- --format

Step 4: Validate Configuration Examples

Check YAML configuration examples:

Example documentation config:

Copy# From docs/configuration.rst
drift_rules:
  provider: anthropic
  model: claude-3-sonnet-20240229

Validate with serena:

Copy# Check config structure
mcp__serena__find_symbol(
    name_path_pattern="Config",
    relative_path="src/drift/config"
)

# Read actual config loading code
mcp__serena__find_symbol(
    name_path_pattern="load_config",
    relative_path="src/drift/config",
    include_body=True
)

Look for:

• Correct parameter names
• Valid provider values (anthropic, bedrock, claude-code)
• Correct structure and nesting

Step 5: Verify API Methods

Use serena to check that documented APIs exist:

Documentation example:

Copyanalyzer = Analyzer(config_path=".drift.yaml")
results = analyzer.analyze()

Verify with serena:

Copy# Find Analyzer class
mcp__serena__find_symbol(
    name_path_pattern="Analyzer",
    relative_path="src/drift",
    depth=1  # Include methods
)

# Check analyze method specifically
mcp__serena__find_symbol(
    name_path_pattern="Analyzer/analyze",
    relative_path="src/drift",
    include_body=True
)

Compare:

• Method exists
• Parameters match
• Return type matches
• No deprecated methods

Step 6: Check Example Completeness

For each code example, verify:

Complete example checklist:

• All imports included
• No undefined variables
• No missing context
• Could theoretically run (or marked as snippet)
• Error handling shown (where relevant)

Example of incomplete documentation:

Copy# BAD - missing imports and context
result = analyzer.analyze()
print(result)

Should be:

Copy# GOOD - complete example
from drift.core.analyzer import Analyzer

analyzer = Analyzer(config_path=".drift.yaml")
result = analyzer.analyze()
print(result)

How to Detect Subjective Language

Step 1: Scan for Subjective Adjectives/Adverbs

Search documentation for these patterns:

Common subjective words:

• "easily", "simply", "just"
• "obviously", "clearly"
• "powerful", "elegant", "beautiful"
• "amazing", "best", "better"
• "great", "excellent"
• "effortlessly", "seamlessly"
• "straightforward"

How to search:

Copygrep -n "easily\|simply\|just\|powerful" docs/*.rst

Or use serena:

Copymcp__serena__search_for_pattern(
    substring_pattern="easily|simply|just|powerful",
    relative_path="docs/"
)

Step 2: Flag Marketing Language

Bad - marketing language:

CopyDrift provides a powerful validation system that makes
it easy to ensure code quality.

Good - objective technical description:

CopyDrift validates project structure using rule-based checks.
Validation runs programmatically without LLM calls.

What to look for:

• Superlatives without data ("best", "most powerful")
• Emotional appeals
• Vague benefit claims
• Comparisons without specifics

Step 3: Identify False User Behavior Claims

CRITICAL - never claim user behavior without data:

Bad examples:

CopyMost users run drift --no-llm in CI/CD
Many teams use this for code review
Typically developers check this daily
Users commonly integrate with GitHub Actions

Good alternatives:

Copydrift --no-llm runs without API calls, suitable for CI/CD
Integrates with GitHub Actions via workflow files
Can be scheduled to run periodically

Prohibited phrases:

• "most users", "most teams", "most workflows"
• "many users", "many developers"
• "typically", "commonly", "usually"
• Any claim about what users do without hard data

Step 4: Suggest Replacements

Pattern: Subjective claim → Objective fact

Example 1:

CopyBad:  "You can easily analyze AI conversations..."
Good: "You can analyze AI conversations..."

Example 2:

CopyBad:  "Drift's powerful validation system..."
Good: "Drift's validation system..."

Example 3:

CopyBad:  "Most users run validation before commits"
Good: "Validation can run before commits via pre-commit hooks"

How to Use Serena MCP for Validation

Find Symbols

Copy# Find a class
mcp__serena__find_symbol(
    name_path_pattern="Analyzer",
    relative_path="src/drift",
    depth=1  # Include methods
)

# Find specific method
mcp__serena__find_symbol(
    name_path_pattern="Analyzer/analyze",
    relative_path="src/drift",
    include_body=True  # See implementation
)

Check Module Structure

Copy# Get overview of a module
mcp__serena__get_symbols_overview(
    relative_path="src/drift/cli.py",
    depth=1
)

Shows all functions, classes, methods in the file.

Search for Usage Patterns

Copy# Find how something is used
mcp__serena__search_for_pattern(
    substring_pattern="Analyzer\\(",
    relative_path="tests/",
    restrict_search_to_code_files=True
)

Look at test files to see correct usage patterns.

Verify Import Paths

Copy# Check if import exists
mcp__serena__find_symbol(
    name_path_pattern="cli",
    relative_path="src/drift/cli.py"
)

# If not found, import path in docs is wrong

How to Generate Audit Reports

See Report Structure for:

• Complete report template
• Issue severity guidelines
• Example issue documentation

Drift-Specific Validation Areas

Validation Rules Documentation

Check that documented rules exist:

Copy# Find validation rules
mcp__serena__list_dir(
    relative_path="src/drift/validators",
    recursive=True
)

# Verify rule names match documentation
mcp__serena__search_for_pattern(
    substring_pattern="class.*Validator",
    relative_path="src/drift/validators"
)

Provider Configuration Examples

Validate provider examples in docs:

Anthropic configuration:

Copyprovider: anthropic
model: claude-3-sonnet-20240229

Bedrock configuration:

Copyprovider: bedrock
model: anthropic.claude-v2
region: us-east-1

Verify:

Copy# Check provider loading code
mcp__serena__find_symbol(
    name_path_pattern="load_provider",
    relative_path="src/drift"
)

CLI Documentation

Validate all CLI examples:

Copy# Documented commands to check:
drift                    # Basic usage
drift --no-llm           # No LLM mode
drift --days 7           # Time range
drift --format json      # Output format
drift --rules NAMES      # Specific rules
drift --cwd PATH         # Working directory

Verify each flag exists:

Copydrift --help | grep "flag-name"

Example Workflows

See Workflows for detailed audit workflows:

• Auditing a single documentation file
• Validating all CLI examples
• Complete audit workflow (3 phases)

Resources

📖 Code Validation

Detailed guide for validating code examples in documentation.

Use when: Checking code examples, imports, and API usage in docs.

📖 Workflows

Step-by-step workflows for different types of documentation audits.

Use when: Performing a documentation audit before release.

📖 Report Structure

Documentation audit report template with issue severity guidelines.

Use when: Generating an audit report with findings and recommendations.