The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses obfuscation by substituting the '$' character for the standard '/' path separator in multiple URLs and package import statements. This is a common technique used to bypass automated scanners. Additionally, it references 'npx agentdb@latest', which downloads and executes unversioned code from the npm registry. Evidence: 'https:/$github.com$ruvnet$agentic-flow$tree$main$packages$agentdb' and 'import { ... } from "agentic-flow$reasoningbank"'.
[COMMAND_EXECUTION]: The troubleshooting section contains instructions to execute a command with elevated privileges using 'sudo' to open network ports on the host system ('sudo ufw allow 4433$udp').
[PROMPT_INJECTION]: The skill processes 'patterns' and 'memories' through functions like 'insertPattern' and 'retrieveWithReasoning'. This creates an indirect prompt injection surface where untrusted data from external databases or peer nodes could influence the agent's reasoning. 1. Ingestion points: 'insertPattern' calls in 'SKILL.md'. 2. Boundary markers: No markers or 'ignore' instructions are present in the provided snippets. 3. Capability inventory: 'retrieveWithReasoning' and 'synthesizeContext' provide capabilities to process and summarize this data into agent context. 4. Sanitization: No sanitization logic is documented.

AgentDB Advanced Features