Clarity and Style

Overview

Write documentation that readers can immediately act on. Core principle: Every abstract concept needs a concrete, runnable example. Every audience needs information in their language.

Key insight: Good writing = easy scanning + clear actions + adapted to reader's context.

When to Use

Load this skill when:

• Writing documentation (README, API docs, runbooks, ADRs)
• Reviewing documentation for clarity
• Explaining technical concepts to different audiences
• Creating user guides or tutorials

Symptoms you need this:

• Documentation says "configure appropriately" without showing how
• Passive voice everywhere ("tests should be run" vs "run tests")
• Same explanation for developers and executives
• Wall-of-text paragraphs without headings
• Jargon without definitions

Don't use for:

• Code comments (use standard practices)
• Commit messages (use conventional commits)
• Chat/email (conversational style different)

Core Patterns

Pattern 1: Active Voice (Who Does What)

Rule: Subject performs action directly. "X does Y", not "Y is done by X".

Passive (❌)	Active (✅)
"The token is validated by the system"	"The system validates the token"
"Tests should be run with pytest"	"Run tests with pytest"
"The configuration file is read at startup"	"The application reads the config file at startup"
"Rate limiting can be configured"	"Configure rate limiting with environment variables"
"Errors are logged to CloudWatch"	"The service logs errors to CloudWatch"

Why: Active voice shows WHO/WHAT does the action, making responsibilities clear.

Common passive constructions to avoid:

• "should be done" → "do X"
• "is processed by" → "X processes"
• "can be configured" → "configure X with Y"
• "is validated" → "the service validates"

When passive is okay: When actor is unknown or irrelevant:

• "The server was compromised" (attacker unknown)
• "The file was deleted" (focus on state change, not actor)

---

Pattern 2: Concrete Examples (Show, Don't Tell)

Rule: Every instruction needs a runnable example. Never say "configure" without showing exact config.

Abstract (❌)	Concrete (✅)
"Set the timeout appropriately"	"Set API_TIMEOUT=30 in .env for 30-second timeout"
"Configure the database connection"	"Set DATABASE_URL=postgresql://user:pass@localhost:5432/dbname"
"Run the tests"	"Run pytest tests/ -v from project root"
"Increase the rate limit"	"Set RATE_LIMIT=1000 (requests per hour) in config.yml"
"Handle errors properly"	"Wrap API calls in try/except and log to CloudWatch:\npython\ntry:\n response = api.call()\nexcept APIError as e:\n logger.error(f\"API failed: {e}\")\n"

Pattern: [Abstract concept] + [Concrete example] + [Expected outcome]

Example:

Configure rate limiting (concept) by setting `RATE_LIMIT=1000` in `.env` (example).
The API will reject requests after 1000/hour per client (outcome).

When to provide examples:

• Commands to run
• Config values to set
• API calls to make
• Error messages you'll see
• File paths to check

Example formats:

• Code blocks for commands: `pytest tests/`
• File snippets for config
• API request/response pairs
• Before/after comparisons

---

Pattern 3: Progressive Disclosure (Essentials First, Details On-Demand)

Rule: Start with minimum viable information. Provide detail progressively, not all at once.

Structure:

1. One-sentence summary (what it is)
2. Minimal quick start (get started in <5 min)
3. Common use cases (cover 80% of users)
4. Advanced topics (expandable sections or separate pages)
5. Complete reference (link to API docs/spec)

Example: Rate Limiting Documentation

❌ Bad (Everything Upfront):

# Rate Limiting

Rate limiting is implemented using a token bucket algorithm with distributed
state management via Redis Cluster. The system tracks requests per client using
API keys extracted from the Authorization header or IP addresses for unauthenticated
requests. Limits are enforced using sliding windows with configurable window sizes
(1 hour, 1 day, 1 month) and multiple tiers (Free: 100/hour, Pro: 10k/hour,
Enterprise: custom). When limits are exceeded, the API returns 429 with Retry-After
header calculated based on the token bucket refill rate. The system supports
distributed deployments with eventual consistency guarantees and graceful degradation
when Redis is unavailable by falling back to in-memory rate limiting...

✅ Good (Progressive Disclosure):

# Rate Limiting

The API limits requests to prevent abuse. Free tier: 100 requests/hour.

## Quick Start

Check your remaining quota:
\`\`\`bash
curl -i https://api.example.com/status
# See X-RateLimit-Remaining header
\`\`\`

## What Happens When Limited

API returns 429 status. Wait time shown in `Retry-After` header (seconds).

## Rate Limit Tiers

| Tier | Limit | Use Case |
|------|-------|----------|
| Free | 100/hour | Development |
| Pro | 10k/hour | Production |
| Enterprise | Custom | High volume |

<details>
<summary>Advanced: How It Works</summary>

Uses token bucket algorithm with Redis. Sliding windows, distributed state...
[Technical details here]
</details>

<details>
<summary>Advanced: Custom Limits</summary>

Contact sales@example.com for Enterprise tier with negotiated limits...
</details>

Benefits:

• New users get started in 30 seconds (check header)
• 80% of users find answer in main sections (tiers, what happens when limited)
• 20% power users access advanced details (expandable)
• Nobody is overwhelmed with token buckets upfront

---

Pattern 4: Audience Adaptation (Write for Your Reader)

Rule: Same information, different framing for different audiences.

Three primary audiences:

Developer Audience

What they need: HOW it works (architecture, APIs, code examples, data flows)

Style:

• Technical precision
• Code examples first
• Architecture diagrams
• API reference details
• Error codes and debugging

Example:

## Authentication (For Developers)

API uses JWT with RS256 signing.

**Request**:
\`\`\`bash
curl -H "Authorization: Bearer eyJhbG..." https://api.example.com/users
\`\`\`

**Token structure**:
\`\`\`json
{
  "sub": "user_12345",
  "scope": ["read:users", "write:posts"],
  "exp": 1730145600
}
\`\`\`

**Validation process**:
1. Extract token from Authorization header
2. Verify signature using public key (fetch from `/keys`)
3. Check expiration (`exp` claim > current time)
4. Verify scopes match endpoint requirements

**Errors**:
- 401: Invalid signature or expired token
- 403: Valid token but insufficient scopes

---

Operator Audience

What they need: HOW to run it (deployment, configuration, monitoring, troubleshooting)

Style:

• Step-by-step procedures
• Config file examples
• Monitoring queries
• Troubleshooting checklists
• Runbooks for incidents

Example:

## Authentication (For Operators)

### Deployment

1. Generate RSA keypair:
   \`\`\`bash
   ssh-keygen -t rsa -b 4096 -f jwt-key
   \`\`\`

2. Set environment variables:
   \`\`\`bash
   JWT_PUBLIC_KEY_PATH=/etc/app/jwt-key.pub
   JWT_ALGORITHM=RS256
   TOKEN_EXPIRY_SECONDS=3600
   \`\`\`

3. Restart service:
   \`\`\`bash
   systemctl restart api-service
   \`\`\`

### Monitoring

**Alert on high auth failures**:
\`\`\`promql
rate(auth_failures_total[5m]) > 10
\`\`\`

### Troubleshooting

**Symptom**: All requests returning 401

**Check**:
1. Public key readable? `ls -la /etc/app/jwt-key.pub`
2. Service logs: `journalctl -u api-service | grep JWT`
3. Key format correct? Should be PEM format, not binary

---

Executive Audience

What they need: WHY it matters (business value, risks, costs, timelines)

Style:

• High-level summaries (no technical jargon)
• Business impact (revenue, risk, customer satisfaction)
• Costs and ROI
• Timeline and milestones
• No implementation details

Example:

## Authentication (For Executives)

### Business Impact

**Security**: JWT authentication prevents unauthorized access to customer data,
reducing breach risk and regulatory liability.

**Cost**: Industry-standard implementation, no licensing fees. Scales to millions
of users with existing infrastructure.

**Customer Experience**: Users stay logged in for 1 hour without re-authentication,
reducing friction while maintaining security.

### Risk Mitigation

- **Before**: API keys in URLs, logged in plaintext, exposed in browser history
- **After**: Short-lived tokens, signed cryptographically, revocable

### Timeline

- Implementation: 2 weeks
- Migration: 1 week (parallel run with old system)
- Full rollout: 1 week

**Investment**: 4 engineering weeks ($40k)
**Risk reduction**: Avoid potential $2M+ breach costs (industry average)

---

Pattern 5: Precision Without Jargon

Rule: Be technically accurate using accessible language. Define acronyms on first use.

Jargon-Heavy (❌)	Precise & Clear (✅)
"Utilize the ingress controller to facilitate external traffic ingress"	"Use the ingress controller to route external traffic into the cluster"
"Implement idempotency semantics"	"Make requests safe to retry - calling twice produces same result"
"Leverage the ORM abstraction layer"	"Use the ORM (Object-Relational Mapping) to query the database with Python code instead of SQL"
"Instantiate a singleton factory pattern"	"Create one shared instance that all code uses (singleton pattern)"

Pattern for acronyms: Full term (Acronym) on first use, acronym thereafter.

Examples:

• First use: "JWT (JSON Web Token)"
• Later: "JWT"
• First use: "SLA (Service Level Agreement)"
• Later: "SLA"

Simplification checklist:

• Replace "utilize" → "use"
• Replace "facilitate" → "help" or "enable"
• Replace "instantiate" → "create"
• Replace "leverage" → "use"
• Define domain terms: "idempotency means safe to retry"

When jargon is okay: When writing for technical audience and term is standard.

• "JWT" in developer docs (industry standard)
• "Kubernetes Pod" in operator docs (specific technical concept)
• "SQL injection" in security docs (precise attack name)

---

Pattern 6: Scannable Structure

Rule: Use headings, bullets, tables, code blocks. Make key information findable in <10 seconds.

Scannable elements:

• ✅ Headings (H2, H3) for sections
• ✅ Bullet points for lists
• ✅ Tables for comparisons
• ✅ Code blocks for commands/examples
• ✅ Bold for key terms (use sparingly)
• ✅ Short paragraphs (3-5 sentences max)

Anti-patterns:

• ❌ Wall-of-text paragraphs (>10 lines)
• ❌ Everything in bold (loses emphasis)
• ❌ No headings (can't scan)
• ❌ Inline code for long examples (use blocks)

Example: Scannable vs Not

❌ Not Scannable:

When you need to deploy the application you should first make sure that Docker
is installed on your system and then you need to clone the repository from GitHub
and after that you should copy the .env.example file to .env and edit it to set
your database credentials and API keys and then you can run docker-compose up -d
to start the containers in detached mode and then wait for the database to
initialize which usually takes about 30 seconds and then you can run the migrations
with docker-compose exec app python manage.py migrate...

✅ Scannable:

## Deployment Steps

### Prerequisites
- Docker installed
- GitHub access

### Setup

1. Clone repository:
   \`\`\`bash
   git clone https://github.com/org/app.git
   \`\`\`

2. Configure environment:
   \`\`\`bash
   cp .env.example .env
   # Edit .env: Set DATABASE_URL and API_KEY
   \`\`\`

3. Start services:
   \`\`\`bash
   docker-compose up -d
   # Wait 30 seconds for database initialization
   \`\`\`

4. Run migrations:
   \`\`\`bash
   docker-compose exec app python manage.py migrate
   \`\`\`

### Verification

Check services are running:
\`\`\`bash
docker-compose ps
# All services should show "Up"
\`\`\`

Why scannable works:

• Find information in <10 seconds (headings)
• Copy commands directly (code blocks)
• See prerequisites before starting (avoids mid-process failures)
• Numbered steps (clear sequence)
• Verification step (know when done)

---

Quick Reference: Clarity Checklist

Use this checklist when writing or reviewing documentation:

Check	Pattern	Example
✅ Active voice	"X does Y" not "Y is done"	"Run tests" not "tests should be run"
✅ Concrete examples	Every instruction has runnable example	"Set API_KEY=abc123" not "configure API key"
✅ Progressive disclosure	Essentials first, details expandable	Quick start → Use cases → Advanced (collapsed)
✅ Audience adapted	Sections for Dev/Ops/Exec as needed	"For Developers: API details" / "For Execs: Business impact"
✅ Acronyms defined	Full term (Acronym) on first use	"JWT (JSON Web Token)" then "JWT"
✅ Scannable structure	Headings, bullets, tables, code blocks	H2 sections, bullet lists, comparison tables
✅ Short paragraphs	3-5 sentences max	Break up walls of text
✅ Bold for emphasis	Key terms only, not whole paragraphs	Important: not everything

---

Common Mistakes

❌ Passive Voice Throughout

Wrong:

The database should be configured before the application is started.
Tests can be run after deployment is completed.

Right:

Configure the database before starting the application.
Run tests after completing deployment.

Why: Active voice shows WHO does WHAT, making actions clear.

---

❌ Abstract Instructions Without Examples

Wrong:

Configure rate limiting appropriately for your use case.

Right:

Configure rate limiting by setting `RATE_LIMIT=1000` (requests/hour) in `config.yml`:
\`\`\`yaml
rate_limiting:
  limit: 1000  # requests per hour
  window: 3600  # seconds
\`\`\`

Why: Readers need concrete examples to act on.

---

❌ Same Content for All Audiences

Wrong:

# Authentication

Uses JWT with RS256 signature algorithm, 3600-second expiry, and refresh token rotation.

Right:

# Authentication

**For Developers**: Uses JWT with RS256. Token expires in 1 hour. Refresh tokens rotate on use.

**For Operators**: Requires RSA keypair. Set `JWT_PUBLIC_KEY_PATH` in config.

**For Executives**: Industry-standard auth. Users stay logged in for 1 hour. Low cost, high security.

Why: Developers need technical details, operators need config, executives need business impact.

---

❌ Jargon Without Definition

Wrong:

Implement idempotency semantics using distributed locks with TTL.

Right:

Make requests safe to retry (idempotency) using distributed locks that expire after a set time (TTL).

**Idempotency**: Calling an API twice produces the same result.
**TTL**: Time To Live - how long a lock lasts before expiring.

Why: Not everyone knows domain-specific terms.

---

❌ Wall-of-Text Paragraphs

Wrong:

When you encounter rate limit errors you should first check the X-RateLimit-Remaining
header to see how many requests you have left and then wait for the time specified
in the Retry-After header before making another request and if you continue to hit
rate limits you should consider upgrading to a higher tier or implementing exponential
backoff in your client code to reduce the request rate automatically...

Right:

## Handling Rate Limits

1. Check `X-RateLimit-Remaining` header (requests left)
2. Wait time shown in `Retry-After` header (seconds)
3. If limits persist:
   - Upgrade to higher tier, OR
   - Implement exponential backoff in client

Why: Scannable structure with numbered steps is easier to follow.

---

Cross-References

Use BEFORE this skill:

• muna/technical-writer/using-technical-writer - Determine document type first

Use WITH this skill:

• muna/technical-writer/documentation-structure - Structure AND clarity

Use AFTER this skill:

• muna/technical-writer/documentation-testing - Verify documentation is clear

Real-World Impact

Well-written documentation using these patterns:

• 38% reduction in support tickets after rewriting config docs with concrete examples (removed "configure appropriately", added exact .env values)
• Developer onboarding time from 2 days → 4 hours after adding progressive disclosure quick start (get running in <10 min, details later)
• Executive buy-in achieved in single meeting after splitting technical docs into "For Executives" section showing business impact, not implementation details

Key lesson: Clarity = concrete examples + active voice + adapted to audience + scannable structure.