registry / @varveai/adit-cli / 1.1.3
@varveai/adit-cli@1.1.3
AI Development Intent Tracker — Transparent Time Machine for your codebase
AI Security Review
scanned 6d ago · by lpm-firewall-aiThe postinstall hook automatically installs ADIT's own skills into user-level agent skill directories and starts its local daemon. This is a real, package-related setup action, so it warrants a warning under the agent-extension lifecycle rule; the reviewed install path does not show broader agent-control changes or unrelated exfiltration.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
npm package installation, unless CI or the install-setup deferral condition skips it.
Impact
Agent sessions may discover ADIT skills without a separate setup command, and the local service begins running after installation.
Mechanism
The lifecycle script runs the package installer, which writes ADIT skill files to the user's Claude and Agents skill directories and starts the daemon.
AI rationale
The postinstall hook automatically installs ADIT's own skills into user-level agent skill directories and starts its local daemon. This is a real, package-related setup action, so it warrants a warning under the agent-extension lifecycle rule; the reviewed install path does not show broader agent-control changes or unrelated exfiltration.
Evidence
package.jsoninstall-skill.cjsdist/installation.jsdist/commands/skills.jsdist/local/server.js~/.agents/skills/<adit-skill>/SKILL.md~/.agents/skills/<adit-skill>/.adit-managed.json~/.claude/skills/<adit-skill>/SKILL.md~/.claude/skills/<adit-skill>/.adit-managed.json
Network endpoints1
127.0.0.1 (local daemon)
Decision evidence
public snapshotAI called this Suspicious at 94.0% confidence as Unknown with medium false-positive risk.
Evidence for warning
- The npm postinstall hook runs install-skill.cjs.
- For published installs, that script invokes install-skills.js with --open-browser; workspace installs use --skills-only.
- The install flow writes ADIT skills under the user's .agents/skills and .claude/skills directories, skipping customized copies.
- Published install setup also starts the ADIT background service.
- The service listens on 127.0.0.1.
Evidence against
- The automatic agent-file changes are limited to ADIT skill files; this install path does not install the separately defined Claude or Codex hooks.
- The PR command's reported requests use the local daemon and require an explicit PR command invocation.
Behavioral surface
ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemNetworkShellWebSocket
HighEntropyStringsMinified
Source & flagged code
9 flaggedpackage.jsonView file
•scripts.postinstall = node install-skill.cjs
High
Install Time Lifecycle Scripts
Package defines install-time lifecycle scripts.
package.jsonView on unpkgdist/local/attribution-analysis.jsView file
1import { createHash, randomUUID } from "node:crypto";
L2: import { execFile } from "node:child_process";
L3: import { promisify } from "node:util";
High
Child Process
Package source references child process execution.
dist/local/attribution-analysis.jsView on unpkg · L1dist/commands/cli-agent/cli-process.jsView file
80
Findings
1 Critical7 High4 Medium5 Low
CriticalAi Agent Control Hijackdist/main.js
HighInstall Time Lifecycle Scriptspackage.json
HighChild Processdist/local/attribution-analysis.js
HighShelldist/commands/cli-agent/cli-process.js
HighCommand Output Exfiltrationdist/commands/pr.js
HighTrigger Reachable Command Output Exfiltrationdist/commands/pr.js
HighRuntime Package Installdist/commands/self-update.js
HighSemantic Analysis Limiteddist/local/attribution-evidence.js#virtual:normalized:round1
MediumDynamic Requiredist/version.js
MediumNetwork
MediumEnvironment Vars
MediumStructural Risk Force Deep Review
Affected versions and remediation
This report applies to @varveai/adit-cli@1.1.3.
See all versions of @varveai/adit-cli for other recorded verdicts.
- Review the evidence and your use of @varveai/adit-cli@1.1.3 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.