### Few-shot Example for Output Formatting

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Provide examples to guide the model in generating structured output, such as JSON objects with specific fields.

```json
{
  "location": "src/auth/login.ts:42",
  "issue": "SQL injection in the username parameter",
  "severity": "critical",
  "suggested_fix": "Use a parameterized query"
}
```

--------------------------------

### Few-shot Examples for Ambiguous Scenarios

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Use these examples to demonstrate how to handle ambiguous user requests by showing the desired action and rationale.

```text
Request: "My order is broken"
Action: Call get_customer -> lookup_order -> check status.
Rationale: “broken” may mean a damaged item; you need order details.

Request: "Get me a manager"
Action: Immediately call escalate_to_human.
Rationale: The customer explicitly requests a human. Do not attempt to solve autonomously.
```

--------------------------------

### Few-shot Examples for Informal Measurements

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Use these examples to guide the model in converting informal measurements into standardized formats, particularly useful for diverse units.

```text
Text: "about two handfuls of rice"
-> {"amount": "~100g", "original_text": "two handfuls", "precision": "approximate"}
```

```text
Text: "a pinch of salt"
-> {"amount": "~1g", "original_text": "a pinch", "precision": "approximate"}
```

--------------------------------

### Write Tool Usage Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Demonstrates the use of the Write tool for creating new files from scratch.

```text
Create a file from scratch
```

--------------------------------

### Initial Application Setup

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/practical_test_en.html

Initializes the application by building the sidebar, rendering the first question, and updating the sidebar. This runs when the page loads.

```javascript
buildSidebar(); renderQuestion(0); updateSidebar();
```

--------------------------------

### Read Tool Usage Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Shows the basic usage of the Read tool to load the entire content of a file for analysis.

```text
Load a file for analysis
```

--------------------------------

### Few-shot Examples for Data Extraction

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Demonstrate how to extract specific information, like measurements and citations, from different document formats using few-shot examples.

```text
Document with inline citations:
"As shown in the study (Smith, 2023), the rate is 42%."
-> {"value": "42%", "source": "Smith, 2023", "type": "inline_citation"}
```

```text
Document with bibliography references:
"The rate is 42%. [1]"
-> {"value": "42%", "source": "reference_1", "type": "bibliography"}
```

--------------------------------

### Glob Tool Usage Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Provides examples of using the Glob tool to find files by name or pattern, useful for code navigation and analysis.

```text
**/*.test.tsx`, `src/components/**/*.ts`
```

--------------------------------

### Bash Tool Usage Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Provides examples of using the Bash tool to execute shell commands, such as git operations, npm commands, or running tests.

```text
git, npm, run tests, build
```

--------------------------------

### Fixed Pipelines (Prompt Chaining) Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Illustrates a fixed pipeline for task decomposition, where each step is defined in advance for predictable workflows.

```text
Document -> Metadata extraction -> Data extraction -> Validation -> Enrichment -> Final output
```

--------------------------------

### Subagent Delegation Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Demonstrates how a main agent can delegate tasks to subagents and how context is managed. The main agent receives a summary instead of raw file data.

```text
Main agent: "Investigate dependencies of the payments module"
  -> Subagent (Explore): reads 15 files, traces imports
  -> Returns: "Payments depends on AuthService, OrderModel, and the external PaymentGateway API"

Main agent: keeps one line in context instead of 15 files
```

--------------------------------

### Directory Structure for .claude/rules/

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Provides an example directory structure for organizing Claude Code rules by topic within the .claude/rules/ directory.

```tree
.claude/rules/
  testing.md          -- testing conventions
  api-conventions.md  -- API conventions
  deployment.md       -- deployment rules
  react-patterns.md   -- React patterns
```

--------------------------------

### Few-shot Examples for Code Quality

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Illustrate the difference between acceptable and problematic code patterns to help the model identify and flag code quality issues.

```javascript
// Acceptable (do not flag):
const items = data.filter(x => x.active);
```

```javascript
// Problem (flag):
const items = data.filter(x => x.active == true); // Use strict equality ===

```

--------------------------------

### Dynamic Adaptive Decomposition Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Illustrates a step-by-step process for breaking down an open-ended task based on intermediate results. Useful when the full scope is unknown upfront and each step depends on the previous one.

```text
1. "Add tests for a legacy codebase"
2. -> First: map the structure (Glob, Grep)
3. -> Found: 3 modules with no tests, 2 with partial coverage
4. -> Prioritize: start with the payments module (high risk)
5. -> During work: discovered a dependency on an external API
6. -> Adapt: add a mock for the external API before writing tests
```

--------------------------------

### Message Batches API Request Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

An example of a request payload for the Message Batches API, demonstrating the use of `custom_id` for linking and managing requests.

```json
{
  "custom_id": "doc-invoice-2024-001",
  "params": {
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Extract data from: ..."}]
  }
}
```

--------------------------------

### Agent State Persistence Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Shows the JSON structure for exporting an agent's state, including execution details and findings. This is used for crash recovery.

```json
// agent-state/web-search-agent.json
{
  "status": "completed",
  "queries_executed": ["AI music 2024", "AI music composition"],
  "results_count": 12,
  "key_findings": [...],
  "coverage": ["music composition", "music production"],
  "gaps": ["music distribution", "music licensing"]
}
```

--------------------------------

### AgentDefinition Configuration in Python

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Example of configuring an agent using the AgentDefinition class in the Claude Agent SDK. Key parameters include name, description, system prompt, and allowed tools.

```python
agent = AgentDefinition(
    name="customer_support",
    description="Handles customer requests for returns and order issues",
    system_prompt="You are a customer support agent...",
    allowed_tools=["get_customer", "lookup_order", "process_refund", "escalate_to_human"],
    # For a coordinator:
    # allowed_tools=["Task", "get_customer", ...]
)
```

--------------------------------

### Grep Tool Usage Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Illustrates using the Grep tool to search for specific text within files, such as function names, error messages, or import statements.

```text
Function name, error message, import
```

--------------------------------

### Claude Interview Pattern Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Claude asks clarifying questions before implementing a solution, useful for unfamiliar domains or tasks with non-obvious implications.

```text
Claude: "Before implementing caching for the API, a few questions:
1. Which cache invalidation strategy do you prefer—TTL or event-based?
2. Is stale data acceptable when the cache is unavailable?
3. Should caching be per-user or global?
4. What is the expected data volume to cache?"
```

--------------------------------

### Incremental Investigation Strategy

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Outlines a step-by-step strategy for investigating codebases incrementally using tools like Grep and Read, starting with entry points and progressively exploring usages.

```text
1. Grep: find entry points (function definition, export)
2. Read: read the found files
3. Grep: find usages (import, calls)
4. Read: read consumer files
5. Repeat until you have a complete picture
```

--------------------------------

### Trimming Tool Results with PostToolUse Hook

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

A Python example using a hook to filter and retain only essential fields from a tool's result, optimizing context usage by reducing noise.

```python
# PostToolUse hook: keep only relevant fields
@hook("PostToolUse", tool="lookup_order")
def trim_order_fields(result):
    return {
        "order_id": result["order_id"],
        "status": result["status"],
        "total": result["total"],
        "items": result["items"],
        "return_eligible": result["return_eligible"]
    }
```

--------------------------------

### Agent Scratchpad for Long Investigations

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

A Markdown example of an agent's scratchpad file used to store key findings during long investigations. This allows the agent to consult findings without re-running discovery.

```markdown
# investigation-scratchpad.md
## Key findings
- PaymentProcessor in src/payments/processor.ts inherits from BaseProcessor
- refund() is called from 3 places: OrderController, AdminPanel, CronJob
- External PaymentGateway API has a rate limit of 100 req/min
- Migration #47 added refund_reason (NOT NULL) — 2024-12-01
```

--------------------------------

### PostToolUse Hook for Data Normalization

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

The `PostToolUse` hook intercepts tool results before they are returned to the model. This example demonstrates normalizing date formats from different tools into a consistent ISO 8601 format.

```python
# Example: normalize date formats from different MCP tools
@hook("PostToolUse")
def normalize_dates(tool_result):
    # Convert Unix timestamp -> ISO 8601
    # Convert "Mar 5, 2025" -> "2025-03-05"
    return normalized_result
```

--------------------------------

### Severity Criteria with Examples for Code Issues

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Define clear severity levels (CRITICAL, HIGH, MEDIUM, LOW) for code issues, accompanied by concrete examples to guide assessment.

```text
CRITICAL: Runtime failure for users
  Example: NullPointerException while processing a payment

HIGH: Security vulnerability
  Example: SQL injection, XSS, missing authorization checks

MEDIUM: Logic bug without immediate impact
  Example: Wrong sorting, off-by-one error

LOW: Code quality
  Example: Duplication, suboptimal algorithm for small data
```

--------------------------------

### Attribution for Claim Provenance

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Contrasts a bad example of a claim without source information to a good example that includes claim details, source, and confidence.

```text
Bad: "The AI music market is estimated at $3.2B." (No source, no year.)
```

```json
Good:
{
  "claim": "The AI music market is estimated at $3.2B.",
  "source_url": "https://example.com/report",
  "source_name": "Global AI Music Report 2024",
  "publication_date": "2024-06-15",
  "confidence": 0.9
}
```

--------------------------------

### Organizing Rules with .claude/rules/ and Paths

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Shows how to use the .claude/rules/ directory with YAML frontmatter and 'paths' to conditionally load rules based on file types or locations.

```yaml
---
paths: ["src/api/**/*"]
---

For API files, use async/await with explicit error handling.
Each endpoint must return a standard response wrapper.
```

```yaml
---
paths: ["**/*.test.tsx", "**/*.test.ts"]
---

Tests must use describe/it blocks.
Use data factories instead of hardcoding.
Do not mock the database—use a test database.
```

--------------------------------

### Escalation for Policy Gap

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Provides an example of escalating a customer request when the current policy does not cover the situation, such as competitor price matching.

```text
Customer: "Competitor X has this item 30% cheaper—give me a discount"
Policy: covers price adjustments only on your own site
Agent: [escalates — policy does not cover competitor price matching]
```

--------------------------------

### Interpreting Temporal Differences

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Highlights the importance of including dates to avoid misinterpreting temporal differences as contradictions. Shows 'bad' and 'good' examples.

```text
Bad: "Source A says 10%, source B says 15%. Contradiction."
Good: "Source A (2023) says 10%, source B (2024) says 15%. Likely +5% growth over a year."
```

--------------------------------

### Using @path for File Imports in CLAUDE.md

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Demonstrates how to use the @path syntax in a project's CLAUDE.md file to reference external configuration files, promoting modularity.

```markdown
# Project CLAUDE.md

Coding standards are described in @./standards/coding-style.md
Test requirements are in @./standards/testing-requirements.md
Project overview is in @README.md and dependencies are in @package.json
```

--------------------------------

### Initialize Agent with Task Tool

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

When initializing an AgentDefinition, ensure the 'Task' tool is included in the `allowed_tools` list if you intend to spawn subagents.

```python
# The coordinator's allowedTools must include "Task"
coordinator_agent = AgentDefinition(
    allowed_tools=["Task", "get_customer"]
)
```

--------------------------------

### Self-Correction Pattern JSON Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

This JSON structure helps detect internal contradictions by comparing stated and calculated values, flagging discrepancies.

```json
{
  "stated_total": "$150.00",
  "calculated_total": "$145.00",
  "conflict_detected": true,
  "line_items": [
    {"name": "Widget A", "price": 75.00},
    {"name": "Widget B", "price": 70.00}
  ]
}
```

--------------------------------

### Tool Choice Any Configuration

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Sets the tool_choice parameter to 'any', forcing the model to call some tool. This is useful when guaranteed structured output is required.

```json
{"type": "any"}
```

--------------------------------

### Legacy .claude/commands/ Structure

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Illustrates the legacy directory structure for defining custom slash commands using individual Markdown files.

```tree
.claude/commands/
  review.md        -- /review -- standard code review
  test-gen.md      -- /test-gen -- test generation
```

--------------------------------

### Edit Tool Usage Example

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Explains the Edit tool's purpose: precisely modifying an existing file by replacing a specific snippet identified by unique text.

```text
Replace a specific snippet via unique text match
```

--------------------------------

### Coordinator Manifest for Resume

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Illustrates the JSON manifest file used by a coordinator to resume agent tasks after a crash. It tracks the status of different agents.

```json
// agent-state/manifest.json
{
  "web-search": "completed",
  "doc-analysis": "in_progress",
  "synthesis": "not_started"
}
```

--------------------------------

### Prompt Chaining for Complex Task Decomposition

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Illustrate prompt chaining by breaking down a complex task, like code review, into sequential, focused steps to avoid attention dilution and ensure consistent analysis.

```text
Step 1: Analyze auth.ts (local issues only)
       -> Output: list of issues in auth.ts

Step 2: Analyze database.ts (local issues only)
       -> Output: list of issues in database.ts

Step 3: Integration pass (cross-file dependencies)
       -> Output: issues at module boundaries
```

--------------------------------

### Annotating Coverage Gaps in Synthesis

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Demonstrates how to use Markdown to indicate partial coverage in a synthesized report, specifically noting the cause of the limitation (e.g., agent timeout).

```markdown
## Report: AI Impact on Creative Industries

### Visual Art (FULL COVERAGE)
[research results]

### Music (PARTIAL COVERAGE — search agent timeout)
[partial results]
⚠️ Note: coverage for this section is limited due to a timeout in the search agent.

### Literature (FULL COVERAGE)
[research results]
```

--------------------------------

### Agentic Loop Steps

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Illustrates the core pattern for autonomous task execution using Claude. The loop involves sending requests with tools, receiving responses, checking the stop reason, and repeating until the task is complete.

```text
1. Send a request to Claude with tools
2. Receive a response
3. Check stop_reason:
   - "tool_use" -> execute the tool, append the result to history, go back to step 1
   - "end_turn" -> the task is complete, show the result to the user
4. Repeat until completion
```

--------------------------------

### PreToolUse Hook for Policy Enforcement

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

The `PreToolUse` hook allows interception of tool calls before they are executed. This example shows how to enforce a policy by blocking refund operations exceeding a specified amount.

```python
# Example: block refunds above $500
@hook("PreToolUse")
def enforce_refund_limit(tool_call):
    if tool_call.name == "process_refund" and tool_call.args.amount > 500:
        return redirect_to_escalation(tool_call)
```

--------------------------------

### Current .claude/skills/ Structure

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Shows the current directory structure for defining custom slash commands and skills, using subdirectories for each skill.

```tree
.claude/skills/
  review/SKILL.md  -- /review -- with frontmatter configuration
  test-gen/SKILL.md
```

--------------------------------

### Immediate Escalation Protocol

Source: https://github.com/paullarionov/claude-certified-architect/blob/main/guide_en.MD

Demonstrates an immediate escalation scenario where the agent directly calls an 'escalate_to_human' function upon a customer's explicit request for a manager.

```text
Customer: "I want to speak to a manager"
Agent: [immediately calls escalate_to_human]
NOT: "I can help with your issue, let me..."
```