agentdb-learning
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs the agent to use
npx agentdb@latest, which downloads and executes the AgentDB CLI tool from the npm registry at runtime. These are vendor-provided resources from the author, ruvnet. - [COMMAND_EXECUTION]: The skill provides several shell commands using the
npxutility for creating plugins, listing templates, and managing learning agents. These commands involve local file operations such as writing to the./pluginsdirectory and managing a local database at.agentdb/learning.db. - [DYNAMIC_EXECUTION]: The skill utilizes WebAssembly (WASM) for high-performance neural inference during the training of models. This involves executing binary code to accelerate machine learning tasks at runtime.
- [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by ingesting external data into the agent's reinforcement learning loop.
- Ingestion points: Untrusted experience data (state, action, reward) is ingested via the
adapter.insertPatternmethod described inSKILL.md. - Boundary markers: No explicit boundary markers or instructions to ignore embedded content within the experience data are present in the provided examples.
- Capability inventory: The skill possesses capabilities including command execution via
npx agentdband training/retrieval operations that influence the agent's future decision-making process. - Sanitization: There is no evidence of sanitization or filtering for the
pattern_dataingested from external environment interactions.
Audit Metadata