agent-cards@0.8.2
CLI for managing virtual debit cards for AI agents
AI Security Review
scanned 25d ago · by lpm-firewall-aiThe scanner flags map to real agent configuration capabilities, but the inspected behavior is tied to user-run setup and no concrete malicious chain is established. Classify this as a warning-level agent capability, not malware.
Static reason
One or more suspicious static signals were detected.; previous stored version diff introduced dangerous source
Trigger
A user runs the agents setup or engineer initialization commands.
Impact
The setup changes future agent guidance for the selected project or user configuration, but inspected evidence does not show an unconsented install-hook mutation or malicious payload.
Mechanism
The CLI writes managed content into agent instruction files and a package-owned Claude skill directory.
AI rationale
The scanner flags map to real agent configuration capabilities, but the inspected behavior is tied to user-run setup and no concrete malicious chain is established. Classify this as a warning-level agent capability, not malware.
Evidence
dist/src/commands/agents.jsdist/src/engineer/commands/connect.jsdist/src/engineer/lib/skill.jsdist/src/engineer/lib/snippet.js
Decision evidence
public snapshotAI called this Suspicious at 83.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- The manifest defines only a prepublishOnly build hook, not an install lifecycle hook.
- The agents command selects agent instruction files and writes a managed steering block that directs agents to AgentCard commands.
- The engineer integration can write a package-owned skill into the repository’s Claude skill directory and a managed instruction snippet.
- These configuration changes are exposed through user-run CLI setup; no concrete malicious behavior was established.
Evidence against
- The inspected code does not establish install-time execution, credential theft, or a malicious network recipient.
- The Git helper creates a local review branch, and the API command dynamically loads the catalog after setting the configured API URL.
Behavioral surface
ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemNetworkShell
HighEntropyStrings
Source & flagged code
6 flaggeddist/src/integrate/git.jsView file
1import { execFileSync } from 'child_process';
L2: import { branchNameFor } from './session.js';
High
Child Process
Package source references child process execution.
dist/src/integrate/git.jsView on unpkg · L1dist/src/commands/api.jsView file
•matchType = previous_version_dangerous_delta
matchedPackage = agent-cards@0.8.0
matchedIdentity = npm:YWdlbnQtY2FyZHM:0.8.0
similarity = 0.773
summary = stored previous version shares package body but lacks this dangerous source file
Critical
Previous Version Dangerous Delta
This package version adds a dangerous source file absent from the previous stored version; route for source-aware review.
dist/src/commands/api.jsView on unpkg23if (existsSync(p)) {
L24: return (await import(pathToFileURL(p).href));
L25: }
Medium
Findings
1 Critical5 High4 Medium7 Low
CriticalPrevious Version Dangerous Deltadist/src/commands/api.js
HighChild Processdist/src/integrate/git.js
HighShell
HighTrigger Reachable External Ai Agent Control Surface Mutationdist/src/commands/agents.js
HighRuntime Package Installdist/src/lib/global-install.js
HighSemantic Analysis Limiteddist/src/commands/api.js#virtual:normalized:round1
MediumDynamic Requiredist/src/commands/api.js
MediumNetwork
MediumEnvironment Vars
MediumStructural Risk Force Deep Review
LowNon Install Lifecycle Scripts
LowScripts Present
LowFilesystem
Affected versions and remediation
This report applies to agent-cards@0.8.2.
See all versions of agent-cards for other recorded verdicts.
- Review the evidence and your use of agent-cards@0.8.2 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.