Remote multi-agent workspace CLI with deep Claude integration, native terminal streaming, and self-hosting.
Flagged as AI-agent capability risk
Allowed by default with warning: agent-facing configuration or capability changes need review before use.
AI security review
scanned · by lpm-firewall-ai
The package provides an explicit command that modifies Claude settings to enable terminal mirroring, alongside an isolated first-party assistant-home bootstrap. The integration is user-triggered and guarded, while npm postinstall only prepares package tools; classify it as a warning-level agent capability.
Static reason
One or more suspicious static signals were detected.; source matched previously finalized malicious package; routed for review; previous stored version diff introduced dangerous source
Trigger
A user runs `very-happy install-terminal-hooks`; the forwarder activates on Claude session events when the terminal supplies the Very Happy environment markers.
Impact
Claude settings gain a global callback capability for terminal mirroring, limited by runtime checks and requiring explicit setup.
Mechanism
The command merges package-owned SessionStart and SessionEnd callbacks into Claude settings. The callback exits for package-managed sessions or sessions outside a Very Happy terminal, then posts the hook payload to the daemon on loopback.
The optional CLI command registers package-owned SessionStart and SessionEnd hooks in Claude settings, preserving unrelated hooks.
Setup requires an explicit command invocation and a tmux readiness check; malformed settings stop the write. Its help limits mirroring to hand-started Claude in a Very Happy Web terminal.
The forwarder exits unless Claude is unmanaged and running in a Very Happy terminal; it sends the hook payload to the local daemon at 127.0.0.1.
Assistant mode creates missing files only under configuration.happyHomeDir/assistant and launches Claude with that directory as its working directory.
The npm postinstall runs platform-tool unpacking and a node-pty helper permission fix; it does not install the Claude hooks.
Other flagged locations select Codex session roots or derive Claude settings paths; they do not establish install-time configuration mutation.
Evidence against
Behavioral surface
SourceChild process · Crypto · Dynamic require · Environment variables · Filesystem · Network · Shell · Web socket
Supply chainHigh entropy strings · Possible obfuscation · URL strings
ManifestNo manifest risk signals triggered.
Scanned 274 files · 4.77 MB of source
19 external domains referenced
127.0.0.1
accounts.google.com
api.cluster-fluster.com
api.openai.com
auth.openai.com
claude.ai
code.claude.com
console.anthropic.com
developers.google.com
docs.openclaw.ai
github.com
goo.gle
happy.mereith.com
oauth2.googleapis.com
opencode.ai
registry.npmjs.org
veryhappy.dev
www.apple.com
www.googleapis.com
Findings
1 critical · 10 high · 7 medium · 5 low
Where they are, most severe file first. Code opens a file's flagged lines, with their evidence, below.
9import { c as claimSessionOrExit, a as applyServerSnapshot, b as applyHibernationReplay, w as withoutServerSnapshot, m as mergeReconnectMetadata } from'./sessionLock-DCKf-gjd.mjs'...