ReasoningBank with AgentDB
Pass
Audited by Gen Agent Trust Hub on Feb 28, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill utilizes
npx agentdb@latestto fetch and execute the AgentDB command-line tool for setup and migration tasks. - [COMMAND_EXECUTION]: Provides instructions to run CLI commands for initializing databases, managing exports, and configuring MCP servers for agent integration.
- [DATA_EXFILTRATION]: Operates on local files and directories, specifically reading from and writing to
./.agentdb/reasoningbank.dband./.swarm/memory.dbfor memory persistence. - [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface due to its core function of retrieving and synthesizing previously recorded 'experiences' into the agent's context.
- Ingestion points: Untrusted data can enter via
insertPatternand themigrateCLI command found in SKILL.md. - Boundary markers: There are no explicit markers or 'ignore' instructions wrapping the retrieved memories in the provided context synthesis examples.
- Capability inventory: The skill utilizes the
agentdbCLI for system-level operations and has access to local database storage. - Sanitization: The implementation does not show any escaping or validation of the ingested pattern data before it is processed by the reasoning modules.
Audit Metadata