AI Security Review
scanned 18m ago · by lpm-firewall-aiUser-invoked initialization writes persistent IMH-Code guidance and skills into multiple global AI assistant configuration directories, and later live execution can run local AI CLIs with permission bypass flags. This is dangerous agent-platform behavior but not npm lifecycle hijacking.
Decision evidence
public snapshot- bin/imhcode.js default init copies package CLAUDE.md, AGENTS.md, and skills into ~/.claude, ~/.gemini, and ~/.copilot.
- bin/imhcode.js registerCliGlobally appends PATH exports to .zshrc/.bashrc/.bash_profile/.profile and writes shims under ~/.imhcode/bin and ~/.local/bin.
- bin/imhcode.js ensureCavemanAndGraphify can run npm install -g skills and npx skills add juliusbrussee/caveman during init.
- src/orchestrator/executor.ts launches opencode with --dangerously-skip-permissions and codex with --dangerously-bypass-approvals-and-sandbox when live agent execution is requested.
- AGENTS.md contains hard-gate agent workflow instructions intended to steer AI coding agents.
- package.json has no npm lifecycle scripts, so behavior is not install-time triggered.
- Broad control-surface writes are reached by explicitly running the imhcode CLI init path, not by import or npm install.
- No credential harvesting or data exfiltration logic found in inspected entrypoints.
- Network activity observed is package-aligned setup/model CLI behavior, not a hardcoded exfiltration endpoint.
- CLAUDE.md/AGENTS.md content is product workflow guidance rather than covert payload instructions.
Source & flagged code
22 flagged · loading sourcePackage source references a known benign dynamic code generation pattern.
skills/typeui-main/src/prompts/designSystem.tsView on unpkg · L18Source writes persistence or remote-access backdoor material.
bin/imhcode.jsView on unpkg · L24A package entrypoint or install-time lifecycle script reaches a source file with blocking dangerous behavior.
bin/imhcode.jsView on unpkg · L24This package version adds a dangerous source file absent from the previous stored version; route for source-aware review.
bin/imhcode.jsView on unpkgManifest entrypoint contains risky behavior absent from dist/build output.
bin/imhcode.jsView on unpkg · L24Package source references dynamic require/import behavior.
bin/imhcode.jsView on unpkg · L22Package hides binary, compressed, or executable-looking payloads in test/fixture/hidden paths.
skills/ui-ux-pro-max/.claude/skills/design/scripts/cip/generate.pyView on unpkgPackage ships non-JavaScript build or shell helper files.
skills/ui-ux-pro-max/.claude/skills/design/scripts/cip/generate.pyView on unpkgPackage ships high-entropy non-source blobs.
skills/theme-factory/theme-showcase.pdfView on unpkgHardcoded password in skills/graphify/skill-windows.md
skills/graphify/skill-windows.mdView on unpkg · L681Hardcoded password in skills/graphify/skill-trae.md
skills/graphify/skill-trae.mdView on unpkg · L592Hardcoded password in skills/graphify/skill-pi.md
skills/graphify/skill-pi.mdView on unpkg · L543Hardcoded password in skills/graphify/skill-codex.md
skills/graphify/skill-codex.mdView on unpkg · L605Hardcoded password in skills/graphify/skill-kiro.md
skills/graphify/skill-kiro.mdView on unpkg · L543Hardcoded password in skills/graphify/skill-copilot.md
skills/graphify/skill-copilot.mdView on unpkg · L603Hardcoded password in skills/graphify/skill-opencode.md
skills/graphify/skill-opencode.mdView on unpkg · L655Hardcoded password in skills/graphify/skill-aider.md
skills/graphify/skill-aider.mdView on unpkg · L544Hardcoded password in skills/graphify/skill-claw.md
skills/graphify/skill-claw.mdView on unpkg · L544Hardcoded password in skills/graphify/skill-droid.md
skills/graphify/skill-droid.mdView on unpkg · L600Hardcoded password in skills/django-tdd/SKILL.md
skills/django-tdd/SKILL.mdView on unpkg · L25Hardcoded password in skills/django-tdd/SKILL.md
skills/django-tdd/SKILL.mdView on unpkg · L117Hardcoded password in skills/django-tdd/SKILL.md
skills/django-tdd/SKILL.mdView on unpkg · L126