Creating Custom Personas

RAPTOR’s expert personas provide specialized methodologies for security analysis. You can create custom personas to capture domain-specific expertise and reuse successful approaches.

Understanding Personas

Personas are expert methodologies extracted from RAPTOR’s Python code and made accessible for explicit invocation. They load on-demand (0 tokens until requested) and provide specialized analysis frameworks.

Available Personas

RAPTOR includes nine built-in expert personas:

Persona	Expert	Purpose	Token Cost
Exploit Developer	Mark Dowd	Generate working PoCs	~650t
Crash Analyst	Charlie Miller / Halvar Flake	Binary crash analysis	~700t
Security Researcher	Research methodology	Vulnerability validation	~620t
Patch Engineer	Senior security engineer	Secure patch creation	~400t
Penetration Tester	Senior pentester	Web payload generation	~350t
Fuzzing Strategist	Expert strategist	Fuzzing decisions	~300t
Binary Exploitation Specialist	Binary expert	Crash exploit generation	~400t
CodeQL Dataflow Analyst	Dataflow expert	Dataflow validation	~400t
CodeQL Finding Analyst	Mark Dowd methodology	CodeQL findings	~350t

How Personas Work

Explicit Invocation

User requests persona: "Use exploit developer persona to create PoC"

Load Methodology

Claude loads persona file from tiers/personas/[name].md

Apply Framework

Persona methodology is applied to the analysis

Return Results

Structured verdict, code, or recommendations returned

Persona Structure

A persona file follows this standard format:

# Persona Name (Named Expert)
# Source: Extracted from [source_file.py]
# Purpose: Brief purpose description
# Token cost: ~XXX tokens
# Usage: "Use [persona name] to [action]"

## Identity

**Role:** Description of the expert role

**Specialization:**
- Area 1
- Area 2
- Area 3

**Purpose:** What this persona helps accomplish

---

## Analysis Framework / Methodology

### 1. First Step

Description and criteria

### 2. Second Step

Description and criteria

### 3. Third Step

Description and criteria

---

## Output Format

Expected output structure with examples

---

## Quality Standards / Checklist

Quality criteria for generated output

---

## Usage

**Invoke explicitly:**

“Use [persona name] to [action]“

**What happens:**
1. Load persona (XXX tokens)
2. Apply methodology
3. Return results

**Token cost:** 0 until invoked, ~XXX when loaded

Creating a Custom Persona

Follow these steps to create a new persona:

Step 1: Define Expertise Domain

Identify the specialized domain:

Good Persona Domains

API security analysis
Cloud configuration auditing
Cryptographic implementation review
Container security assessment
Authentication mechanism analysis
Database security configuration

Poor Persona Domains (Too Broad)

“General security” (use Security Researcher instead)
“Code review” (too generic)
“Testing” (not specific enough)

Step 2: Develop Methodology

Create a systematic approach with clear steps:

## Analysis Framework

### 1. Discovery Phase
- What to look for
- Where to find it
- How to extract it

### 2. Assessment Phase
- What to check
- What indicates security issues
- What indicates good practices

### 3. Validation Phase
- How to verify findings
- How to test exploitability
- How to rule out false positives

### 4. Reporting Phase
- What to include in report
- How to prioritize findings
- What remediation to recommend

Step 3: Create Persona File

Create a new file in tiers/personas/:

touch tiers/personas/my_custom_persona.md

Step 4: Write Persona Content

Use this template:

# My Custom Persona Name
# Source: Custom persona for [use case]
# Purpose: [Brief description]
# Token cost: ~XXX tokens (estimate)
# Usage: "Use [persona name] to [action]"

## Identity

**Role:** [Expert role description]

**Specialization:**
- [Area 1]
- [Area 2]
- [Area 3]

**Purpose:** [What this persona accomplishes]

---

## Analysis Framework

### 1. [First Step Name]

**Objective:** [What this step accomplishes]

**Process:**
- [Sub-step 1]
- [Sub-step 2]
- [Sub-step 3]

**Indicators:**
- ✅ [Good indicator]
- 🔶 [Warning indicator]
- ❌ [Bad indicator]

### 2. [Second Step Name]

[Follow same pattern]

### 3. [Third Step Name]

[Follow same pattern]

---

## Decision Criteria

**Mark as [VERDICT 1] if:**
- [Condition 1]
- [Condition 2]
- [Condition 3]

**Mark as [VERDICT 2] if:**
- [Condition 1]
- [Condition 2]

---

## Output Format

```markdown
## [PERSONA NAME] ANALYSIS

Target: [What was analyzed]

### [STEP 1 NAME]
[Results]

### [STEP 2 NAME]
[Results]

### VERDICT

**[VERDICT]**
Confidence: [High/Medium/Low]

Reasoning:
[Detailed explanation]

Recommendations:
- [Action 1]
- [Action 2]

Usage

Invoke explicitly:

"Use [persona name] to analyze [target]"
"[Persona name]: check [specific aspect]"

What happens:

Load this persona (XXX tokens)
Apply [methodology name]
Return structured analysis

Token cost: 0 until invoked, ~XXX when loaded

## Example: API Security Analyst Persona

Here's a complete example of a custom persona:

```markdown
# API Security Analyst Persona
# Source: Custom persona for REST API security
# Purpose: Systematic API endpoint security analysis
# Token cost: ~450 tokens
# Usage: "Use API security analyst to review endpoints"

## Identity

**Role:** Expert API security analyst specializing in REST API vulnerabilities

**Specialization:**
- Authentication and authorization flaws
- IDOR (Insecure Direct Object Reference)
- Rate limiting and abuse prevention
- Input validation and injection
- Mass assignment vulnerabilities
- API versioning security

**Purpose:** Identify security weaknesses in API endpoints and recommend secure implementations

---

## Analysis Framework

### 1. AUTHENTICATION ANALYSIS

**Objective:** Verify proper authentication on all endpoints

**Process:**
- Identify authentication mechanism (Bearer token, API key, OAuth, etc.)
- Check if authentication is required on sensitive endpoints
- Verify token validation happens on every request
- Assess token strength and entropy
- Check for token expiration

**Indicators:**
- ✅ Strong authentication on all sensitive endpoints
- 🔶 Authentication present but weak (API keys, basic auth)
- ❌ No authentication required on sensitive endpoints

### 2. AUTHORIZATION ANALYSIS

**Objective:** Prevent unauthorized access to resources

**Process:**
- Check for object-level authorization (IDOR prevention)
- Verify function-level authorization (role-based access)
- Test horizontal privilege escalation (access other users' data)
- Test vertical privilege escalation (access admin functions)
- Verify resource ownership validation

**Indicators:**
- ✅ Ownership checked for all resource access
- 🔶 Authorization present but may be bypassable
- ❌ No authorization checks (IDOR vulnerability)

### 3. RATE LIMITING ANALYSIS

**Objective:** Prevent abuse and brute force attacks

**Process:**
- Check rate limiting on authentication endpoints
- Check rate limiting on resource-intensive operations
- Verify rate limits are per-user or per-IP
- Test if rate limits can be bypassed
- Assess rate limit thresholds (too high = ineffective)

**Indicators:**
- ✅ Rate limiting on all sensitive endpoints
- 🔶 Rate limiting exists but thresholds too high
- ❌ No rate limiting on login/register/token endpoints

### 4. INPUT VALIDATION ANALYSIS

**Objective:** Prevent injection and validation bypass

**Process:**
- Check SQL injection in query parameters
- Check NoSQL injection in filters
- Check XSS in user-supplied fields
- Check mass assignment vulnerabilities
- Verify output encoding

**Indicators:**
- ✅ Parameterized queries, proper encoding, whitelist validation
- 🔶 Some validation but may be incomplete
- ❌ No input validation, string concatenation

---

## Decision Criteria

**Mark as CRITICAL if ANY:**
- No authentication on sensitive endpoints
- IDOR vulnerability (access any user's data)
- SQL/NoSQL injection
- Mass assignment leading to privilege escalation

**Mark as HIGH if ANY:**
- Weak authentication (API keys, basic auth over HTTP)
- Missing authorization checks on some endpoints
- No rate limiting on authentication
- XSS in stored user content

**Mark as MEDIUM if ANY:**
- Rate limits too high
- Incomplete input validation
- Information leakage in error messages
- Missing security headers

**Mark as SECURE if ALL:**
- Strong authentication (OAuth 2.0, JWT with expiration)
- Object and function-level authorization
- Rate limiting on all sensitive endpoints
- Comprehensive input validation
- Proper output encoding

---

## Output Format

```markdown
## API SECURITY ANALYSIS

Endpoint: [HTTP METHOD] [PATH]
Authentication: [Type]

### AUTHENTICATION
✅/🔶/❌ [Verdict]
Findings:
- [Finding 1]
- [Finding 2]

### AUTHORIZATION
✅/🔶/❌ [Verdict]
Findings:
- [Finding 1]
- [Finding 2]

### RATE LIMITING
✅/🔶/❌ [Verdict]
Findings:
- [Finding 1]
- [Finding 2]

### INPUT VALIDATION
✅/🔶/❌ [Verdict]
Findings:
- [Finding 1]
- [Finding 2]

### OVERALL VERDICT

**[CRITICAL/HIGH/MEDIUM/SECURE]**
Confidence: [High/Medium/Low]

Top Issues:
1. [Issue 1 with severity]
2. [Issue 2 with severity]
3. [Issue 3 with severity]

Recommendations:
1. [Fix 1]
2. [Fix 2]
3. [Fix 3]

Exploitability:
[Attack scenario if exploitable]

Usage

Invoke explicitly:

"Use API security analyst to review /api/users endpoint"
"API security analyst: check authentication on POST /api/orders"
"Analyze API security for all endpoints using API analyst persona"

What happens:

Load this persona (450 tokens)
Apply 4-step framework (auth, authz, rate limit, validation)
Return structured analysis with severity

Token cost: 0 until invoked, ~450 when loaded

## When to Create Custom Personas

Create a custom persona when:

<CardGroup cols={2}>
  <Card title="Specialized Domain" icon="bullseye">
    You have deep expertise in a specific security domain
  </Card>
  <Card title="Repetitive Analysis" icon="rotate">
    You perform the same analysis steps frequently
  </Card>
  <Card title="Complex Methodology" icon="diagram-project">
    The analysis requires multiple systematic steps
  </Card>
  <Card title="Team Knowledge" icon="users">
    You want to capture and share team expertise
  </Card>
</CardGroup>

### Good Use Cases

- **Cloud Security Auditor:** AWS/Azure/GCP configuration analysis
- **Container Security Specialist:** Docker/Kubernetes security assessment
- **Smart Contract Auditor:** Blockchain smart contract vulnerabilities
- **GraphQL Security Analyst:** GraphQL-specific security issues
- **Mobile App Security Tester:** Android/iOS security assessment
- **Infrastructure as Code Auditor:** Terraform/CloudFormation security

### Poor Use Cases

- Generic "security tester" (use Security Researcher)
- One-time analysis (not worth creating persona)
- Simple pattern matching (use Semgrep rules instead)
- Language-agnostic checks (use existing tools)

## Loading Custom Personas

Once created, invoke your persona in Claude Code:

“Use API security analyst persona to review /api/users endpoint”

Claude will:
1. Load `tiers/personas/api_security_analyst.md`
2. Apply the 4-step framework
3. Return structured analysis

## Persona Best Practices

### Keep It Focused

<Tabs>
  <Tab title="Good: Focused" icon="check">
    **API Authentication Analyst**
    - Focuses solely on API authentication
    - Deep expertise in token security
    - Clear decision criteria
  </Tab>
  <Tab title="Bad: Too Broad" icon="xmark">
    **API Analyst**
    - Tries to cover all API security
    - Shallow coverage of many topics
    - Unclear when to use
  </Tab>
</Tabs>

### Provide Clear Criteria

```markdown
## Decision Criteria

**Mark as VULNERABLE if:**
- Specific condition 1 (measurable)
- Specific condition 2 (testable)
- Specific condition 3 (objective)

**Mark as SECURE if:**
- Opposite conditions with clear thresholds

Avoid vague criteria like “if it looks insecure” or “if there might be issues.”

Include Real Examples

## Examples

### Vulnerable Code
```python
# No authorization check - IDOR vulnerability
@app.route('/api/user/<user_id>')
def get_user(user_id):
    return User.query.get(user_id)

Secure Code

# Proper ownership validation
@app.route('/api/user/<user_id>')
@login_required
def get_user(user_id):
    if current_user.id != user_id and not current_user.is_admin:
        abort(403)
    return User.query.get(user_id)

### Estimate Token Cost

Roughly estimate token count:
- ~4 characters per token
- Simple persona: 300-400 tokens
- Medium persona: 400-600 tokens
- Complex persona: 600-800 tokens

Keep under 1,000 tokens to avoid context bloat.

### Test Your Persona

Before finalizing:

1. **Test invocation:** Can Claude load and apply it?
2. **Test clarity:** Are decision criteria clear and actionable?
3. **Test usefulness:** Does it provide value over generic analysis?
4. **Test edge cases:** Does it handle unusual situations?

## Integration with Skills

<Info>
Personas provide **methodology**, while skills provide **reusable approaches**. They complement each other.
</Info>

**Persona:** Expert framework (exploit developer, crash analyst)
**Skill:** Specific technique (ROP chain building, heap exploitation)

User: Use crash analyst persona with ROP chain skill Claude: [Loads crash_analyst.md - methodology] [Loads rop_chain_skill.md - technique] [Combines both for comprehensive analysis]

## Versioning and Maintenance

### Version Your Personas

```markdown
# API Security Analyst Persona v2.0
# Last updated: 2024-03-04
# Changelog:
# - v2.0: Added GraphQL support
# - v1.1: Enhanced IDOR detection
# - v1.0: Initial release

Update Regularly

New attack techniques: Add to methodology
False positives: Refine decision criteria
User feedback: Incorporate lessons learned
Tool evolution: Update references

Share personas with your team:

Commit to repository: tiers/personas/team/
Document in README: Update persona list
Provide examples: Include usage examples
Train team: Show how to invoke

Summary

Creating custom personas:

Identify specialized domain
Develop systematic methodology
Create persona file in tiers/personas/
Test with real analysis
Refine based on results
Share with team

Personas capture expertise and make it reusable across your security practice.

Next Steps

Claude Code Usage

Learn how to invoke personas interactively

Extending RAPTOR

Add custom scanners and agents

Python CLI

Use RAPTOR from command line

Existing Personas

Reference for built-in personas

Documentation Index

​Understanding Personas

​Available Personas

​How Personas Work

​Persona Structure

​Creating a Custom Persona

​Step 1: Define Expertise Domain

​Step 2: Develop Methodology

​Step 3: Create Persona File

​Step 4: Write Persona Content

​Usage

​Usage

​Include Real Examples

​Secure Code

​Update Regularly

​Sharing Personas

​Summary

​Next Steps

Claude Code Usage

Extending RAPTOR

Python CLI

Existing Personas

Understanding Personas

Available Personas

How Personas Work

Persona Structure

Creating a Custom Persona

Step 1: Define Expertise Domain

Step 2: Develop Methodology

Step 3: Create Persona File

Step 4: Write Persona Content

Usage

Usage

Include Real Examples

Secure Code

Update Regularly

Sharing Personas

Summary

Next Steps