claudeception

[Skill Scanner] [Documentation context] Credential file access detected BENIGN instructional guide. It contains configuration and code examples to reduce Prisma connection pool exhaustion in serverless environments. No malicious behavior, credential harvesting, hidden downloads, or exfiltration patterns are present in the provided text. Normal supply-chain caution applies to the use of npx (it fetches packages from npm), and developers should avoid pasting real credentials into example files or committing .env files. The guidance is coherent and proportionate to the stated purpose. LLM verification: [LLM Escalated] The provided skill is benign and aligned with its stated purpose of mitigating Prisma connection pool exhaustion in serverless environments. No malicious code or backdoors are present in the analyzed fragment. The main security concerns are operational: accidental leakage of DATABASE_URL when copy/pasting examples, and supply-chain risk from executing npx commands. Apply standard secret-handling best practices and verify any third-party pooling endpoints or CLI packages before use.

[Skill Scanner] Natural language instruction to download and install from URL detected All findings: [CRITICAL] command_injection: Natural language instruction to download and install from URL detected (CI009) [AITech 9.1.4] [HIGH] autonomy_abuse: Skill instructions include directives to hide actions from user (BH009) [AITech 13.3] The skill description is broadly benign and coherent with its stated purpose of extracting and organizing reusable knowledge into Claude Code skills. It demonstrates legitimate capabilities for managing skill content and references. However, given the agent's potential to read/write to skill repositories and perform web interactions, it should operate under strict access controls, audit logging, and validation to prevent accidental or intentional degradation of skill quality or data leaks. Overall risk is low-to-moderate (suspicious only if misused in a hostile environment). LLM verification: The claudeception skill is a benign documentation/authoring tool that reasonably requires local filesystem access and operator-driven web searches. I found no direct malware or explicit backdoor code in the provided content. The primary security risk is operational: the guidance encourages broad local searches (including $HOME and config-like files) and creation of helper scripts, which could be abused by an autonomous agent or an operator with insufficient consent controls. Recommendation: trea