rafey-shell@1.0.6
Renamed to rabot. Install with npm install -g @arafeymasood/rabot
An AI-powered shell with personalized LLM assistance
AI Security Review
scanned 2d ago · by lpm-firewall-aiThe package’s chat feature sends user queries to its API, but it also includes the local working-directory path in every prompt without documenting that collection. This is suspicious runtime telemetry rather than a demonstrated malicious attack; the flagged invisible character appears in ordinary biography text and shows no hidden behavior.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
When a user submits a chat message in the CLI.
Impact
The package’s service receives the local working-directory path without disclosure in the README; this is excess runtime collection, but no credential or unrelated file collection was identified.
Mechanism
The CLI builds a prompt from the system profile, recent conversation, current working directory, timestamp, and user query, then sends it to the configured API endpoint.
AI rationale
The package’s chat feature sends user queries to its API, but it also includes the local working-directory path in every prompt without documenting that collection. This is suspicious runtime telemetry rather than a demonstrated malicious attack; the flagged invisible character appears in ordinary biography text and shows no hidden behavior.
Evidence
dist/cli.jsdist/services/LLMService.jsapi/chat.tsREADME.md
Network endpoints2
rafey-shell.vercel.app/api/chatGoogle Generative AI service (endpoint not specified in source)
Decision evidence
public snapshotAI called this Suspicious at 86.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- The CLI starts an interactive AI assistant by default, and the README describes chatting as the package’s purpose.
- The service sends the generated prompt to rafey-shell.vercel.app by default; callers can override the endpoint with RAFEY_SHELL_API_URL.
- The prompt automatically includes the current working directory and timestamp alongside the user query and conversation context; this collection is not disclosed in the README.
- The new system prompt contains an invisible Unicode character within a hospital name, but the surrounding text does not show a concealed instruction.
Evidence against
- The network request supports the package’s stated LLM chat feature, and the API implementation forwards the prompt to Google’s Generative AI service.
- The prepare script only runs the project build; no install-time data collection was identified.
- The invisible character appears within ordinary biography text, with no evidence that it changes code behavior or hides reviewer-directed instructions.
Behavioral surface
EnvironmentVars
Source & flagged code
5 flaggeddist/prompts/system.jsView file
27contains invisible/control Unicode U+200B (zero width space)
- Research Scientist at Guy's and St Thomas'<U+200B> NHS Foundation Trust (July 2018 - October 2018): Worked closely with senior physicians @ Guy's Hospital to develop a web application to detect irregularities in MRI scans. The tool analyz
Critical
Trojan Source Unicode
Source contains bidi control or invisible Unicode characters associated with Trojan Source attacks.
dist/prompts/system.jsView on unpkg · L27•Trigger-reachable chain: scripts.start -> dist/cli.js -> dist/shell/RafeyShell.js -> dist/services/LLMService.js -> dist/prompts/system.js
Reachable file contains a blocking source-risk pattern.
Critical
Trigger Reachable Dangerous Capability
A package entrypoint or install-time lifecycle script reaches a source file with blocking dangerous behavior.
dist/prompts/system.jsView on unpkgpackage.jsonView file
•Runtime dependency names matching Node built-ins: readline
Findings
2 Critical1 High5 Medium4 Low
CriticalTrojan Source Unicodedist/prompts/system.js
CriticalTrigger Reachable Dangerous Capabilitydist/prompts/system.js
HighNode Builtin Dependency Squatpackage.json
MediumNetwork
MediumEnvironment Vars
MediumStructural Risk Force Deep Review
MediumAi Review Evidencepackage.json
MediumAi Review EvidenceREADME.md
LowNon Install Lifecycle Scripts
LowScripts Present
LowHigh Entropy Strings
LowUrl Strings
Affected versions and remediation
This report applies to rafey-shell@1.0.6.
See all versions of rafey-shell for other recorded verdicts.
- Review the evidence and your use of rafey-shell@1.0.6 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.