registry / greprag / 5.74.0

greprag@5.74.0

GrepRAG — agent memory for Claude Code, Codex, and OpenCode.

AI Security Review

scanned 29d ago · by lpm-firewall-ai

LPM treats this as warn-only first-party agent extension lifecycle risk. Install-time setup copies this package's skill into detected Claude/Codex skill directories. Explicit setup registers agent hooks that upload turn data and inject cloud inbox content into agent context.

Static reason

High-risk behavior combination matched malicious policy.

Trigger

npm install for skill copy; explicit greprag init --codex plus trusted hooks for runtime behavior

Impact

A configured user can have prompts, responses, tool summaries, and inbound cloud messages processed by GrepRAG.

Mechanism

first-party agent skill setup, lifecycle hooks, and authenticated cloud memory/inbox relay

Rationale

This is not concrete malware under the install-control policy: automatic installation is limited to a package-owned skill, while hook configuration requires an explicit command and platform trust review. The package still warrants a warning because its configured hooks can relay sensitive agent context and cloud-delivered instructions.

Evidence

package.jsonscripts/postinstall.jsdist/commands/init.jsdist/hook.jsdist/codex-steering.jsdist/codex-prompt-cache.js~/.claude/skills/greprag~/.codex/skills/greprag~/.codex/hooks.json~/.greprag/.env~/.greprag/codex/*.json

Network endpoints2

/v1/memory/turn

greprag@5.74.0

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Source & flagged code