prompt

Free

Audit agent skill ecosystems against supply-chain poisoning and attacks

FreeFree tier

Type

Open Source

Visit Website GitHub (40 stars)

About prompt

Agent Skill Supply-Chain Security Auditor is a specialized prompt from the ai-boost/awesome-prompts repository. It instructs an AI agent to inspect, audit, and harden agent skill ecosystems against supply-chain poisoning, self-propagating attacks, and privilege escalation. The prompt covers three core responsibilities: Skill Manifest Audit (verifying SKILL.md integrity, flagging undocumented scripts), Documentation Poisoning Detection (scanning code examples for hidden malicious logic like dynamic execution, disguised imports, exfiltration patterns), and MCP Tool Schema Security (checking for flat inputs, excessive data exposure). It is based on recent arXiv research (2026) and claims to identify 90%+ of documented DDIPE skill-poisoning patterns, mapped to MITRE ATT&CK and OWASP Agentic Top 10.

Key Features

Skill Manifest Audit: Verify SKILL.md frontmatter integrity, flag undocumented scripts and hidden files, validate scope narrowness

Documentation Poisoning Detection (DDIPE): Scan code examples for disguised imports, dynamic execution (eval, exec), credential harvesting, dependency confusion

MCP Tool Schema Security: Check for flat inputs, excessive data exposure in output contracts, error handling gaps

Cross-reference claimed functionality against actual code behavior

Detect steganographic payloads in configuration snippets

Maps to MITRE ATT&CK and OWASP Agentic Top 10

Identifies 90%+ of documented DDIPE skill-poisoning patterns

Based on multiple 2026 arXiv papers (Supply-Chain Poisoning, Self-Propagating Attacks, ClawSafety, Anthropic, Microsoft)

Pros & Cons

Pros

Comprehensive coverage: manifest audit, documentation poisoning, MCP schema security
Based on cutting-edge academic research (2026) and industry frameworks (MITRE, OWASP)
High detection rate (90%+) for documented DDIPE patterns
Open source and free to use as a prompt template
Provides structured responsibilities that can be integrated into automated pipelines

Best For

Auditing third-party skill repositories and community contributions for malicious codeSecurity review of MCP servers and tool wrappers against compromised schemasDetecting poisoned shared memory or context pools used across multiple agentsAnalyzing transitive dependencies between skills that escalate privileges implicitlyHardening agent skill ecosystems against self-propagating attacks

FAQ

What sources does this auditor use?

It references five sources: Supply-Chain Poisoning Attacks Against Agent Skill Ecosystems (arXiv 2604.03081, April 2026), Self-Propagating Attacks Across LLM Agent Ecosystems (arXiv 2603.15727, March 2026), ClawSafety (arXiv 2604.01438, April 2026), Anthropic's Trustworthy Agents in Practice (Apr 2026), and Microsoft Agent Governance Toolkit (Apr 2026).

What attack patterns does it detect?

It identifies 90%+ of documented DDIPE (documentation poisoning, disguised imports, dynamic execution, exfiltration) skill-poisoning patterns and maps them to MITRE ATT&CK and OWASP Agentic Top 10.

Is this a standalone tool or a prompt?

It is a prompt intended for use with an AI agent (such as a large language model) to guide it in performing supply-chain security audits on agent skill ecosystems.