registry / dokkimi / 0.5.4

dokkimi@0.5.4

CLI for isolated Docker environments for microservice testing

AI Security Review

scanned 12d ago · by lpm-firewall-ai

LPM flags this version as an AI-agent control-surface risk. npm postinstall silently mutates global AI-agent configuration and instruction files. It registers a persistent MCP command and prompts agents to execute it.

Static reason

High-risk behavior combination matched malicious policy.

Trigger

npm installation of dokkimi@0.5.4

Impact

Unconsented control of Claude Code, Cursor, and Copilot behavior for the installing user.

Mechanism

postinstall registration of MCP and AI instruction files

Policy narrative

During npm postinstall, the package loads its LLM registration module without user action. That module creates or modifies global Claude Code, Cursor, and GitHub Copilot instruction/configuration files, registers `dokkimi mcp`, and injects text directing agents to use its tools or run the CLI. This is persistence across unrelated AI-agent sessions and changes broad foreign control surfaces at install time.

Rationale

The package performs concrete, unconsented postinstall mutation of multiple foreign AI-agent control surfaces. This meets the blocking policy regardless of the absence of observed exfiltration.

Evidence

package.jsonscripts/postinstall.jsapps/cli/dist/lib/llm-context-register.js~/.claude.json~/.cursor/mcp.json~/.claude/CLAUDE.md~/.cursor/rules/dokkimi.md~/.github/copilot-instructions.md~/.dokkimi/dokkimi-instructions.md

dokkimi@0.5.4

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Source & flagged code