ai-prompt-leaking

Systematically extract hidden system prompts, core directives, and invisible context intentionally concealed within Large Language Model (LLM) applications. This skill utilizes targeted linguistic e

SKILL.md

GitHub

AI Prompt Leaking

When to Use

  • When analyzing an AI-powered system (customer support bot, coding assistant, data analyst) to uncover its proprietary internal instructions, hidden API keys, or pre-configured biases.
  • To demonstrate how seemingly secure conversational agents can be tricked into revealing their foundational programming.

Prerequisites

  • Access to target AI/ML system or local model deployment for testing
  • Python 3.9+ with relevant ML libraries (transformers, torch, openai)
  • Understanding of LLM architecture and prompt processing pipelines
  • Authorized scope and rules of engagement for AI red team testing

Workflow

Phase 1: Context Boundary Testing

# Concept: The LLM ```

### Phase 2: Targeted Extraction Prompts

```text
# ```

### Phase 3: Translation and Obfuscation Exploitation

```text
# ```

### Phase 4: Summarization Attacks

```text
# ```

#### Decision Point 🔀
```mermaid
flowchart TD
    A[Formulate Prompt ] --> B{Prompt Leaked ]}
    B -->|Yes| C[Document System ]
    B -->|No| D[Refine ]
    C --> E[Exploit Further ]

🔵 Blue Team Detection & Defense

  • Strict Delimiters: Heuristic Output Filtering: Key Concepts
    Concept Description

Output Format

Ai Prompt Leaking — Assessment Report
============================================================
Target: [Target identifier]
Assessor: [Operator name]
Date: [Assessment date]
Scope: [Authorized scope]
MITRE ATT&CK: [Relevant technique IDs]

Findings Summary:
  [Finding 1]: [Severity] — [Brief description]
  [Finding 2]: [Severity] — [Brief description]

Detailed Results:
  Phase 1: [Phase name]
    - Result: [Outcome]
    - Evidence: [Screenshot/log reference]
    - Impact: [Business impact assessment]

  Phase 2: [Phase name]
    - Result: [Outcome]
    - Evidence: [Screenshot/log reference]
    - Impact: [Business impact assessment]

Risk Rating: [Critical/High/Medium/Low/Informational]
Recommendations:
  1. [Immediate remediation step]
  2. [Long-term hardening measure]
  3. [Monitoring/detection improvement]

📚 Shared Resources

For cross-cutting methodology applicable to all vulnerability classes, see:

References