registry / process-ai / 0.5.0

process-ai@0.5.0

Framework method-agnóstico de mapeamento de processos ponta-a-ponta — equipe de 6 agentes IA (Déa+Bento+Miguel+Júlia+Zanoni+Laura) que conduz o usuário por Q&A até documentar o processo completo (ingestão documental→SIPOC→hierarquia→BPMN→POPs) com confian

AI Security Review

scanned 8d ago · by lpm-firewall-ai

LPM flags this version as an AI-agent control-surface risk. Install-time code mutates the consumer project and Claude Code's .claude/skills control surface without an explicit user command. It also invokes pip to install Python dependencies.

Static reason

One or more suspicious static signals were detected.

Trigger

npm install process-ai@0.5.0

Impact

Registers process-ai agent skills and writes configuration/content into the consuming project.

Mechanism

postinstall launches pip and the package installer

Policy narrative

On npm installation, postinstall runs pip dependency installation and executes the package CLI using the consumer directory as target. The CLI installs package-controlled skills into the consumer's .claude/skills directory, enabling Claude Code slash-command behavior, and writes additional project configuration and method packs. This is an unconsented install-time mutation of an AI-agent control surface.

Rationale

The behavior is concrete and automatic at postinstall, not merely a user-invoked setup path. Although no data theft was found, the package mutates a foreign AI-agent control surface during dependency installation.

Evidence

package.jsonbin/postinstall.jsdist/bin/process-ai.jsdist/toolkit/adapters/claude-code/skill-copy.jsdist/toolkit/src/installer/orchestrator.jsscripts/requirements-ingest.txt<consumer>/.claude/skills/process-ai*/SKILL.md<consumer>/.process-ai/config<consumer>/method-packs/

process-ai@0.5.0

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Source & flagged code