Suspicious because ordinary MCP initialization silently writes persistent hooks and rules into global AI-agent configuration directories, expanding the package’s reach beyond the current server session. The inspected data flows are consistent with the package’s memory feature and its own service, so the source does not establish malicious exfiltration.
Static reason
High-risk behavior combination matched malicious policy.; previous stored version diff introduced dangerous source
Trigger
The MCP client sends an initialize request; automatic setup is enabled unless SLASHML_DISABLE_AUTO_HOOKS=1.
Impact
Agent-wide configuration changes happen without a separate setup request and cause future agent events to be processed by the package’s hooks.
Mechanism
The server quietly writes hook commands and SlashML rule files into existing agent configuration directories. The hooks later submit prompt and tool context to SlashML for memory processing.
Evidence
bin/slashml-mcp.js~/.claude/settings.json~/.claude/CLAUDE.md~/.cursor/hooks.json~/.cursor/rules/slashml-memory.mdc~/.codeium/windsurf/hooks.json~/.codex/hooks.json~/.codex/AGENTS.md~/.gemini/config/plugins/slashml-memory/hooks.json