registry / @sema-agent/core / 1.301.0

@sema-agent/core@1.301.0

Stateless, task-oriented AI agent core

AI Security Review

scanned 2d ago · by lpm-firewall-ai

No install-time, import-time, or covert attack surface was confirmed. The package is an AI-agent core with explicit runtime shell and web capabilities.

Static reason

High-risk behavior combination matched malicious policy.

Trigger

Explicit CLI/API task execution by the integrating application or user.

Impact

Requested agent tasks can run with the host permissions supplied by the integrator; no automatic credential exfiltration, persistence, or foreign control-surface mutation was found.

Mechanism

User-configured agent execution, provider API calls, and package-owned session storage.

Rationale

Scanner reverse-shell evidence is a false positive caused by security-policy text and denylist literals. Source inspection finds package-aligned, explicitly invoked agent functionality without a concrete malicious chain.

Evidence

package.jsondist/core/auto-mode-prompt-assets.jsdist/core/auto-promote.jsdist/bin/sema-tb.jsdist/engine/execution-env/node-execution-env.jsdist/stores/file/fs-atomic.js

Network endpoints3

api.openai.com/v1api.anthropic.comapi.deepseek.com

@sema-agent/core@1.301.0

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Source & flagged code