memory-bridge
Warn
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: Accesses sensitive files within the Claude Code configuration directory (
~/.claude/projects/*/memory/*.md). These files contain session history and project-specific memories. The data is processed and stored in an external database (AgentDB) via MCP tools. - [COMMAND_EXECUTION]: Provides instructions to execute
node .claude/helpers/auto-memory-hook.mjs import-all, which runs a local script outside the skill's distribution package. - [DYNAMIC_EXECUTION]: Utilizes an external helper script for automation. The content and security of
.claude/helpers/auto-memory-hook.mjscannot be verified from the skill source. - [INDIRECT_PROMPT_INJECTION]: Ingests potentially untrusted content from memory files which could contain malicious instructions from previous sessions.
- Ingestion points: Markdown files in
~/.claude/projects/*/memory/. - Boundary markers: None present; the instructions do not specify any delimiters or safety warnings for the imported content.
- Capability inventory: Includes file system access (
Read), shell execution (Bash), and database storage tools (mcp__plugin_ruflo-core_ruflo__memory_import_claude). - Sanitization: No sanitization or validation of the ingested markdown content is described.
Audit Metadata