Back to catalog
Static reviewDevelopment

Skill dossier

testing-for-system-prompt-leakage

Source declaration

Extracts LLM system prompts using direct requests, jailbreak/instruction-override framing, translation/encoding tricks, and few-shot replay, combining manual payloads with automated garak and Promptfoo scanners to surface embedded secrets, routing logic, and policy leakage (OWASP LLM07:2025). Use during LLM application red-team engagements or when validating that no credentials or authorization logic live in the system prompt.

The author's original description, preserved in English. It describes intent, not verified performance.

Extracts LLM system prompts using direct requests, jailbreak/instruction-override framing, translation/encoding tricks, and few-shot replay, combining manual payloads with automated garak and Promptfoo scanners to surface embedded secrets, routing logic, and policy leakage (OWASP LLM07:2025). Use during LLM application red-team engagements or when validating that no credentials or authorization logic live in the system prompt.

developmentreferencesscripts

Research summary

Machine-generated analysis; pending human review

These observations come from deterministic source parsing, not a human security review or controlled task evaluation.

Extracts LLM system prompts using direct requests, jailbreak/instruction-override framing, translation/encoding tricks, and few-shot replay, combining manual payloads with automated garak and Promptfoo scanners to surface embedded secrets, routing logic, and policy leakage (OWASP LLM07:2025). Use during LLM application red-team engagements or when validating that no credentials or authorization logic live in the system prompt. This bilingual summary is generated deterministically from collected source text and file structure; it is pending human editorial review.

Review scope

Deterministic static analysis of collected SKILL.md and supporting text/code. Machine-assisted candidates and findings await human editorial review; no dynamic behavior or effect is assessed.

Observable capabilities

  • Workflow [Source statement; provisional, not tested. Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L82]
  • Step 1: Baseline with direct extraction requests [Source statement; provisional, not tested. Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L84]
  • Step 2: Instruction-override and role-play framing [Source statement; provisional, not tested. Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L95]
  • Step 3: Completion and delimiter attacks [Source statement; provisional, not tested. Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L103]
  • Step 4: Encoding / format-shift attacks [Source statement; provisional, not tested. Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L113]
  • Step 5: Automated probing with garak [Source statement; provisional, not tested. Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L121]
  • Step 6: Build a Promptfoo red-team extraction suite [Source statement; provisional, not tested. Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L142]
  • Step 7: Triage leaked content for secrets and logic [Source statement; provisional, not tested. Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L177]
  • Step 8: Remediate and re-test [Source statement; provisional, not tested. Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L190]

Limitations & scope

  • Static analysis cannot assess runtime behavior, command success, network reachability, or model output quality.
  • No commands, scripts, network requests, model calls, or controlled skill executions were performed.
  • Category, tags, capabilities, workflows, and bilingual editorial summaries are deterministic candidates pending human review, not verified outcomes.
  • Exact token counts are unavailable until a later pipeline performs and records an official model-specific count.

Relevant for

  • Practitioners evaluating whether to inspect and install the source skill; this audience is inferred from the product context, not measured usage. [Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L1-L26]

What ships with the skill

The file inventory comes from the collected source at the recorded commit. Paths and sizes are structural observations, not a measure of quality.

files
5
Markdown files
3
scripts
1
references
2
Size
28.8 KB
Structural complexity
Low

File inventory

PathTypeBytes
skills/testing-for-system-prompt-leakage/LICENSEOther11 KB
skills/testing-for-system-prompt-leakage/references/api-reference.mdReference2.6 KB
skills/testing-for-system-prompt-leakage/references/standards.mdReference1.5 KB
skills/testing-for-system-prompt-leakage/scripts/agent.pyScript4.4 KB
skills/testing-for-system-prompt-leakage/SKILL.mdInstruction9.2 KB

SKILL.md frontmatter

{
  "name": "testing-for-system-prompt-leakage",
  "description": "Extracts LLM system prompts using direct requests, jailbreak/instruction-override framing, translation/encoding tricks, and few-shot replay, combining manual payloads with automated garak and Promptfoo scanners to surface embedded secrets, routing logic, and policy leakage (OWASP LLM07:2025). Use during LLM application red-team engagements or when validating that no credentials or authorization logic live in the system prompt.",
  "domain": "cybersecurity",
  "subdomain": "ai-security",
  "tags": [
    "ai-security",
    "system-prompt-leakage",
    "prompt-extraction",
    "llm-red-team",
    "owasp-llm07",
    "garak",
    "promptfoo",
    "data-leakage"
  ],
  "version": "1.0",
  "author": "mahipal",
  "license": "Apache-2.0",
  "nist_ai_rmf": [
    "MEASURE-2.7"
  ],
  "atlas_techniques": [
    "AML.T0057"
  ]
}

A reading of the operating sequence

This workflow reconstructs the apparent sequence from source material. Unless controlled evaluation is explicitly recorded, it is provisional—not an observed trace.

  1. 01
    Action

    Overview

    Source-described action candidate under “Overview”; not verified execution.

  2. 02
    Decision

    The system prompt must never be treated as a secret or used as a security control. If leaking it breaks your security model, the security model is wrong. The real findings during a leakage test are the secrets and logic embedded in the prompt that should have been enforced server-side.

    Source-described decision candidate under “Overview”; not verified execution.

  3. 03
    Action

    System prompts are extractable. Through direct requests, instruction-override (jailbreak) framing, translation/encoding tricks, completion attacks, and few-shot replay, adversaries can reliably recover preambles.

    Source-described action candidate under “Overview”; not verified execution.

  4. 04
    Decision

    When to Use

    Source-described decision candidate under “When to Use”; not verified execution.

  5. 05
    Action

    Objectives

    Source-described action candidate under “Objectives”; not verified execution.

  6. 06
    Action

    MITRE ATT&CK Mapping

    Source-described action candidate under “MITRE ATT&CK Mapping”; not verified execution.

Inspect all extracted workflow candidates (21 candidate steps)
  1. 07
    Action

    Workflow

    Source-described action candidate under “Workflow”; not verified execution.

  2. 08
    Action

    Step 1: Baseline with direct extraction requests

    Source-described action candidate under “Step 1: Baseline with direct extraction requests”; not verified execution.

  3. 09
    Action

    Step 2: Instruction-override and role-play framing

    Source-described action candidate under “Step 2: Instruction-override and role-play framing”; not verified execution.

  4. 10
    Action

    Step 3: Completion and delimiter attacks

    Source-described action candidate under “Step 3: Completion and delimiter attacks”; not verified execution.

  5. 11
    Action

    Step 4: Encoding / format-shift attacks

    Source-described action candidate under “Step 4: Encoding / format-shift attacks”; not verified execution.

  6. 12
    Action

    Step 5: Automated probing with garak

    Source-described action candidate under “Step 5: Automated probing with garak”; not verified execution.

  7. 13
    Action

    Step 6: Build a Promptfoo red-team extraction suite

    Source-described action candidate under “Step 6: Build a Promptfoo red-team extraction suite”; not verified execution.

  8. 14
    Action

    Step 7: Triage leaked content for secrets and logic

    Source-described action candidate under “Step 7: Triage leaked content for secrets and logic”; not verified execution.

  9. 15
    Action

    Step 8: Remediate and re-test

    Source-described action candidate under “Step 8: Remediate and re-test”; not verified execution.

  10. 16
    Action

    Move all secrets out of the prompt into a secrets manager; the model should never see them.

    Source-described action candidate under “Step 8: Remediate and re-test”; not verified execution.

  11. 17
    Action

    Enforce authorization and routing decisions in application code, not in the preamble.

    Source-described action candidate under “Step 8: Remediate and re-test”; not verified execution.

  12. 18
    Output

    Add an output guardrail (see `defending-llms-with-guardrails`) that blocks responses echoing the preamble.

    Source-described output candidate under “Step 8: Remediate and re-test”; not verified execution.

  13. 19
    Output

    Re-run garak and Promptfoo; the extraction plugins must report zero successful leaks.

    Source-described output candidate under “Step 8: Remediate and re-test”; not verified execution.

  14. 20
    Action

    Tools and Resources

    Source-described action candidate under “Tools and Resources”; not verified execution.

  15. 21
    Action

    Validation Criteria

    Source-described action candidate under “Validation Criteria”; not verified execution.

Potential risk surfaces

Static findings identify points to inspect before installation. They do not establish that a skill is safe, malicious, or effective.

Attention recommended

15 deterministic lexical finding(s) require review across the collected text. Findings can occur in documentation and do not prove execution or exploitability; binary-only files were not scanned.

Network accessObserved
File writesObserved
Credential useObserved
Medium

Network access is mentioned

9 occurrences

A URL or network client appears in this source line. Review destinations, data sent, and trust boundaries; static text cannot establish actual network activity. This rule matched 9 location(s) in the collected text; 3 representative location(s) are cited. Matching text may be documentation, and the analyzer did not execute it.

Note

File modification is described or implemented

2 occurrences

This line contains a possible file-write or redirection operation. Check destination paths and overwrite behavior; the static review did not write any file. This rule matched 2 location(s) in the collected text; 2 representative location(s) are cited. Matching text may be documentation, and the analyzer did not execute it.

High

Credential or secret material is referenced

4 occurrences

This line resembles a credential literal or reads a sensitive environment value. Do not expose its contents; verify whether it is a placeholder, required secret, or accidentally committed material. This rule matched 4 location(s) in the collected text; 3 representative location(s) are cited. Matching text may be documentation, and the analyzer did not execute it.

Where context can be spent

Discovery, activation, and resource layers have different meanings. Exact values appear only when the official token-counting method was run with a named model.

Exact token counts unavailable

All token values, including per-file inventory, are null until the pipeline performs an official exact count with an explicit model.

No token measurement was recorded with the required official method. This dossier does not substitute character counts, file sizes, or estimates.

Controlled evaluation

No controlled evaluation

This skill has only the evidence identified above. No task success, token efficiency, latency, or quality advantage is claimed.

Not tested

Use with the source in view

Install command
Inspect the pinned source at https://github.com/mukul975/Anthropic-Cybersecurity-Skills/blob/54a798831d2266a3ca61ce68a7acb80b81160d57/skills/testing-for-system-prompt-leakage/SKILL.md and follow the source repository's installation instructions. No installation was performed or verified by this static analyzer. Source license: Apache-2.0.

Prerequisites

  • Static analysis cannot assess runtime behavior, command success, network reachability, or model output quality.
  • No commands, scripts, network requests, model calls, or controlled skill executions were performed.
  • Category, tags, capabilities, workflows, and bilingual editorial summaries are deterministic candidates pending human review, not verified outcomes.
  • Exact token counts are unavailable until a later pipeline performs and records an official model-specific count.

Example prompts

Not provided

A decision, with its limits attached

Strengths

  • The snapshot contains 5 collected file(s), 3 Markdown file(s), and 1 script file(s), enabling source-level inspection. [Evidence: skills/testing-for-system-prompt-leakage/SKILL.md:L1]

Considerations

  • No functional, security, or model-quality result has been established. Human review must inspect source context and all cited findings before installation or publication.
Research conclusion

Provisional deterministic static inventory only; pending human editorial review and, separately, controlled evaluation. No install recommendation, security assurance, or effectiveness claim is made.

Traceable by construction

The dossier is a time-bound reading of a public artifact, not an endorsement. Use the source link to inspect the exact upstream material.

mukul975Anthropic-Cybersecurity-Skillsskills/testing-for-system-prompt-leakage
Source type
Community
Commit
54a798831d2266a3ca61ce68a7acb80b81160d57
Collected
20 Nov 2026
Last checked
20 Nov 2026
Upstream license
Apache-2.0
Repository stars
36,446
Content basis

Deterministic lexical parsing, YAML frontmatter validation, file inventory, source-described workflow extraction, and heuristic provisional classification of the supplied files only. Generated by machine; no LLM, execution, dynamic testing, or human editorial approval.