hunt-llm-ai
Hunt LLM/AI feature bugs — prompt injection, indirect injection, exfiltration via tool-use/markdown, ASCII smuggling, agentic AI security (OWASP Agentic Apps 2026, ASI01-ASI10). Patterns: direct injection ('ignore previous instructions'), indirect injection via documents/web pages/email the model reads, ASCII smuggling (Unicode Tags block U+E0000-U+E007F, invisible to humans, decoded by the model), tool-use exfiltration (model has fetch/browse tool, attacker injects OOB URL, model exfils chat history/secrets), markdown-image zero-click exfil, system-prompt extraction, IDOR-via-AI (cross-tenant data). Targets: chatbots, RAG, summarizers, agentic copilots, MCP tools. Detection: any LLM-backed endpoint, doc upload triggering AI processing, autonomous agent with tools. Validate: OOB/Collaborator callback for exfil, verbatim-reproducible system-prompt leak (run twice), verifiable cross-tenant leak or RCE. Confabulation is NOT a finding. Use when hunting AI features, chatbots, RAG, agentic systems, MCP.
70
86%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
View guide
Critical
Do not install without reviewing
Security
2 findings: 1 critical severity, 1 high severity. Installing this skill is not recommended: please review these findings carefully if you do intend to do so.
E004: Prompt injection detected in skill instructions.
What this means
Detected a prompt injection in the skill instructions. The skill contains hidden or deceptive instructions that fall outside its stated purpose and attempt to override the agent’s safety guidelines or intended behavior.
Why it was flagged
The document contains explicit, actionable instructions for secret exfiltration and attack techniques (markdown/image zero-click, tool-use/browse SSRF, DNS exfil, ASCII/Unicode smuggling, system-prompt extraction, etc.), which is clearly malicious/deceptive guidance despite a defensive framing.
W007: Insecure credential handling detected in skill instructions.
What this means
The skill handles credentials insecurely by requiring the agent to include secret values verbatim in its generated output. This exposes credentials in the agent’s context and conversation history, creating a risk of data exfiltration.
Why it was flagged
The skill contains explicit templates and examples that instruct an agent to insert user secrets (API keys, emails, tokens) verbatim into image URLs, fetch/tool calls, or OOB callbacks, which requires the LLM to output secret values directly.
Low
Low-risk findings.
1 low severity finding. Worth noting, but not necessarily harmful.
W011: Third-party content exposure detected (indirect prompt injection risk).
What this means
The skill exposes the agent to untrusted, user-generated content from public third-party sources, creating a risk of indirect prompt injection. This includes browsing arbitrary URLs, reading social media posts or forum comments, and analyzing content from unknown websites.
Why it was flagged
The skill document describes indirect-injection channels (e.g., “a web page the summarize this URL feature fetches” and “an email/calendar invite/Jira ticket/PR description an agentic assistant processes”) that would require the runtime workflow to ingest outsider-authored free text via those monitored inputs, creating a reachable injection surface.
- Repository
- elementalsouls/Claude-BugHunter
- Commit
646b72e
- Audited
- Security analysis
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.