skills/semgrep/skills/llm-security/Runlayer

llm-security

Fail

Audited by Runlayer on Mar 8, 2026

Risk Level: HIGH

Scan Summary

Max Score

93%

Files

Flagged

Chunks

Flagged Files (12)

llm-security/AGENTS.mdHIGH

93.4%

Tool passed security scan

Risky tool definition detected

Current: ${current_spend}, Limit: ${budget}" # Warning at 80% usage if current_spend / budget > Decimal("0.8"): log_warning(f"User {user_id} at {current_spend/budget*100}% of budget") return True, "" def record_cost( self, user_id: str, model: str, input_tokens: int, output_tokens: int ): """Record actual cost after request.""" config = COST_CONFIGS.get(model) actual_cost = ( config.input_cost_per_1k * (input_tokens / 1000) + config.output_cost_per_1k * (output_tokens / 1000) ) self.db.record_us

llm-security/rules/system-prompt-leakage.mdHIGH

92.4%

Risky tool definition detected

**Minimize prompt sensitivity** - Only include necessary instructions **References:** - [OWASP LLM07:2025 System Prompt Leakage](https://genai.owasp.org/llmrisk/llm07-system-prompt-leakage/) - [MITRE ATLAS T0051 - Prompt Injection (Meta Prompt Extraction)](https://atlas.mitre.org/techniques/AML.T0051)

llm-security/rules/vector-embedding.mdHIGH

91.2%

Risky tool definition detected

**Sanitize content** - Remove hidden characters and suspicious patterns **References:** - [OWASP LLM08:2025 Vector and Embedding Weaknesses](https://genai.owasp.org/llmrisk/llm08-vector-and-embedding-weaknesses/) - [RAG Security Best Practices](https://docs.aws.amazon.com/prescriptive-guidance/latest/rag-llm-application-patterns/security.html)

llm-security/rules/unbounded-consumption.mdHIGH

91.0%

Risky tool definition detected

Tool: llm-security/rules/unbounded-consumption.md [2/2] Description: l in lengths) / len(lengths) if length_variance < 100: # Very consistent lengths return True return False def _is_adversarial_structure(self, query: str) -> bool: """Detect adversarial query structures.""" # Check for unusual character patterns if len(set(query)) < len(query) * 0.3: # Low character diversity return True # Check for token manipulation patterns if re.search(r'(.)\1{10,}', query): # Repeated characters return True

llm-security/rules/excessive-agency.mdMEDIUM

89.8%

Tool passed security scan

Risky tool definition detected

**Default deny** - Reject unknown or unvalidated actions **References:** - [OWASP LLM06:2025 Excessive Agency](https://genai.owasp.org/llmrisk/llm06-excessive-agency/) - [Principle of Least Privilege](https://csrc.nist.gov/glossary/term/least_privilege) - [NeMo Guardrails](https://github.com/NVIDIA/NeMo-Guardrails)

llm-security/rules/data-poisoning.mdMEDIUM

88.6%

Risky tool definition detected

**Red team testing** - Test models for hidden backdoors and biases **References:** - [OWASP LLM04:2025 Data and Model Poisoning](https://genai.owasp.org/llmrisk/llm04-data-and-model-poisoning/) - [MITRE ATLAS T0018 - Backdoor ML Model](https://atlas.mitre.org/techniques/AML.T0018) - [Poisoning Attacks on Machine Learning](https://arxiv.org/abs/2007.08199)

llm-security/rules/supply-chain.mdMEDIUM

88.6%

Risky tool definition detected

llm-security/rules/misinformation.mdMEDIUM

79.7%

Tool passed security scan

Risky tool definition detected

llm-security/rules/output-handling.mdLOW

72.9%

Risky tool definition detected

**Never use shell=True** - Avoid shell execution with LLM-derived input 5.

llm-security/rules/sensitive-disclosure.mdLOW

71.5%

Risky tool definition detected

llm-security/README.mdLOW

60.3%

Tool passed security scan

llm-security/rules/_sections.mdLOW

52.2%

Tool passed security scan

Passed Files (2)Click to expand

llm-security/rules/prompt-injection.mdOK

36.7%

Tool passed security scan

llm-security/SKILL.mdOK

32.5%

Tool passed security scan

Audit Metadata

Max File Score

93%

Classification

KNOWN_SERVER_ALL_UNKNOWN

Files Scanned

Files Flagged

Chunks Analyzed

Analyzed

Mar 8, 2026, 12:05 PM