registry / devez-vibe / 1.10.18
devez-vibe@1.10.18
Stable terminal UI for Codex and Claude Agent SDK
AI Security Review
scanned 1d ago · by lpm-firewall-aiSource inspection confirms an automatic postinstall change to user-level agent skill directories, so this warrants a warning for agent extension lifecycle risk. The inspected behavior is limited to installing the package’s bundled skills and does not establish a publish-blocking attack.
Static reason
High-risk behavior combination matched malicious policy.; source matched previously finalized malicious package; routed for review
Trigger
Installing the package through npm triggers its postinstall hook.
Impact
Existing matching skill directories can be changed during installation, affecting later agent behavior.
Mechanism
The hook runs install-skills.mjs, which copies bundled skill trees into both agent skill directories and prunes destination files absent from the package copy.
AI rationale
Source inspection confirms an automatic postinstall change to user-level agent skill directories, so this warrants a warning for agent extension lifecycle risk. The inspected behavior is limited to installing the package’s bundled skills and does not establish a publish-blocking attack.
Evidence
package.jsoninstall-skills.mjs~/.codex/skills/luna-loop~/.codex/skills/insane-search~/.claude/skills/luna-loop~/.claude/skills/insane-search
Decision evidence
public snapshotAI called this Suspicious at 92.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- The postinstall hook automatically runs the skill installer.
- The installer targets user-level Codex and Claude skill directories.
- It copies and prunes the package’s bundled skill trees in those directories.
- The bridge’s separate permission update is exposed as a runtime method.
Evidence against
- The inspected skill installation is limited to the package’s two named bundled skills.
- The bridge’s settings paths are selected by destination scope.
Behavioral surface
ChildProcessCryptoEnvironmentVarsFilesystem
HighEntropyStringsUrlStrings
Source & flagged code
7 flagged · loading sourcepackage.jsonView file
•scripts.postinstall = node install-skills.mjs && node prepare-search-deps.mjs
High
Install Time Lifecycle Scripts
Package defines install-time lifecycle scripts.
package.jsonView on unpkg•scripts.postinstall = node install-skills.mjs && node prepare-search-deps.mjs
Medium
Ambiguous Install Lifecycle Script
Install-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkginstall-skills.mjsView file
1import {
L2: copyFileSync,
L3: existsSync,
L4: mkdirSync,
L5: readdirSync,
L6: rmSync,
L7: writeFileSync,
L8: } from "node:fs";
...
L16: const userHome = homedir();
L17: const codexHome = process.env.CODEX_HOME?.trim() || join(userHome, ".codex");
L18: const claudeHome = process.env.CLAUDE_CONFIG_DIR?.trim() || join(userHome, ".claude");
L19:
Critical
Ai Agent Control Hijack
Source creates an unconsented AI-agent control surface through install-time mutation or a default unauthenticated remote skill channel.
Findings
1 Critical3 High5 Medium4 Low
CriticalAi Agent Control Hijackinstall-skills.mjs
HighInstall Time Lifecycle Scriptspackage.json
HighPayload In Excluded Dirskills/insane-search/tests/live_benchmark/bench.py
HighKnown Malware Source Similaritybin/dvz.js
MediumAmbiguous Install Lifecycle Scriptpackage.json
MediumEnvironment Vars
MediumShips Native Binarybin/dvz.exe
MediumShips Build Helperskills/insane-search/tests/live_benchmark/bench.py
MediumStructural Risk Force Deep Review
LowScripts Present
LowFilesystem
LowHigh Entropy Strings
LowUrl Strings
Affected versions and remediation
This report applies to devez-vibe@1.10.18.
See all versions of devez-vibe for other recorded verdicts.
- Review the evidence and your use of devez-vibe@1.10.18 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.