From the RiskyPlugins AI security review of the observed evidence. The review covers version 0.1.19, not the analysed version 0.1.37.
Analysis record
- Analysed
- 1 month ago 18 October 2026
- Version
- v0.1.37
- Artifact
- SHA256 ab86248b…dca492
- Detail
- Full list
Is @7ots/cli safe?
@7ots/cli is the command line side of 7ots.com, a desktop pet that sits next to your AI coding agent. It hooks into Claude Code and shell sessions, keeps reminders and notes, and can watch your screen if you turn that on. It declares no browser permissions and no host permissions. What it does instead is make ordinary network calls from files like src/auth/AuthManager.js and src/llm/ProxyLLM.js, and one real service in its endpoint list is api.letta.com, a known agent framework.
The most serious label in the scan is a critical secret finding, repeated six times, for a hardcoded token in src/i18n/messages/server.js. That file holds translated interface text. A rule that hunts for fixed tokens inside server code will match translation keys and placeholder strings, so the finding points at a file rather than showing a live credential. If it were real, it would mean a working credential shipped inside a text file anyone can open.
Nothing in the scan shows the pattern that matters for agent tools: reading your AWS or SSH keys and posting them somewhere. The network calls line up with what the description advertises, sign-in, model traffic through a proxy, and the chat and mail integrations.
The scanner also produced a long endpoint list that is mostly noise. Entries like body.desktop.menu, card.bio and 3ώ.mx came from code being read as if it were a domain name. 3ώ.mx is not a hostname. What you are weighing here is a young 0.1.19 release from an individual developer that asks for screen access and shell hooks.
Evidence ledger
Ranked by severity · findings with a source location link to the code viewer
YARA Rule Matches
16 rules| Severity | Rule | Hits | Files | Metadata |
|---|---|---|---|---|
| LOW | LocalStorageShouldNotBeUsed | 7 | dist/7ots.esm.js.mapcli/pet/pet.htmlsrc/core/storage.js +4 more | - |
| LOW | UsingCommandLineArguments | 1 | cli/7ots.mjs | - |
| LOW | credential env files | 48 | server/sdk.mjscli/lib/config.mjsserver/demo.mjs +45 more | - |
| LOW | postinstall persistence mechanism | 14 | cli/pet/settings.htmlsrc/actions/PageTools.jsdist/7ots.iife.js.map +11 more | - |
| LOW | postinstall file download | 54 | src/integrations/apumail.jscli/lib/tooluse.mjssrc/character/more/acc-head.js +51 more | - |
| LOW | NoUseWeakRandom | 27 | cli/pet/electron/director.cjscli/pet/pet.htmlsrc/character3d/room-themes.js +24 more | - |
| LOW | HavingAPermissiveCrossOriginResourceSharingPolicy | 1 | server/server.mjs | - |
| LOW | SQLInjection | 1 | cli/lib/mcp.mjs | - |
| LOW | postinstall network communication | 65 | src/avatar/AvatarStage.jscli/lib/meet.mjscli/pet/electron/main.cjs +62 more | - |
| LOW | postinstall system command | 77 | src/llm/ProxyLLM.jssrc/core/AgentBrain.jscli/lib/integrations.mjs +74 more | - |
| LOW | UsingShellInterpreterWhenExecutingOSCommands | 17 | cli/7ots.mjscli/lib/stt.mjscli/lib/integrations.mjs +14 more | - |
| LOW | postinstall crypto operations | 30 | cli/lib/account.mjscli/lib/assistant.mjscli/lib/orquesta.mjs +27 more | - |
| LOW | postinstall obfuscation | 33 | cli/pet/pet.htmlsrc/character/motion.jssrc/character/mods.js +30 more | - |
| LOW | postinstall registry modification | 9 | src/core/AgentBrain.jssrc/index.jssrc/core/AgentWidget.js +6 more | - |
| LOW | AlertStatementsShouldNotBeUsed | 1 | cli/lib/senses/index.mjs | - |
| LOW | postinstall file manipulation | 72 | src/actions/PageTools.jscli/lib/wake.mjscli/lib/desktop-sync.mjs +69 more | - |
Finding Categories
YARA Rules Matched
16 rules(457 hits)MCP Server Analysis
MCP servers expose tools and resources to AI assistants. Unlike browser extensions, they run as standalone processes with direct system access. Tool definitions are analyzed for prompt injection, data exfiltration, and tool poisoning patterns.
AI Security Report
AI Security Review
Evidence context: threat category none; evidence quality moderate.
@7ots/cli is the command line half of 7ots.com, a desktop pet that attaches to Claude Code and shell hooks, keeps reminders and notes, and can look at your screen when you allow it. The findings track that job description closely.
Tool poisoning: zero. The package does ship src/mcp/McpClient.js, which makes network calls at lines 73, 94 and 103, but that file is an MCP client reaching out to servers. It does not define tool descriptions that steer an agent, and the scan found no hidden directives, no instruction tags and no invisible Unicode anywhere (obfuscation findings: none).
Credential and network scope: no finding touches .ssh/, .aws/credentials, .kube/config or application_default_credentials.json. The sixteen fetches sit in files that match advertised features: src/auth/AuthManager.js:96-101 for sign-in, src/llm/ProxyLLM.js:49 for model traffic through a proxy, src/integrations/apuchat.js and src/integrations/apumail.js for the chat and mail hooks, and src/identity/ContactCard.js plus src/identity/Face.js for rendering the pet's identity card. The one recognisable host in the endpoint list is api.letta.com, a known agent framework. The rest of that list is source code parsed as hostnames: body.desktop.menu, card.bio, call.pin, boot.pid, audio.play. c2.cx reads badly, but the same extractor emitted 3ώ.mx and 1-l.sy, which are not domains.
The six critical hits all read MCP-TRANSPORT-HARDCODED-TOKEN-src/i18n/messages/server.js. That path is a translation catalog for a message namespace called server. A transport rule looking for static bearer strings in server code fires on message keys and placeholder text. It deserves a glance at the file, and there is no companion finding of reading a credential store, so it is not theft architecture.
Strongest counterargument: the risk here is capability, not code. Shell hooks, screen capture, an LLM proxy and an MCP client in one package, published by an individual account, at version 0.1.19 with no recorded users. A tool built to watch your coding agent can do damage without a single malicious line. What keeps this out of malicious territory is the shape of the evidence: 489 low-severity code-smell hits, 425 IoC strings dominated by parsed source, sixteen fetches that map to declared features, zero malware signatures, zero obfuscation and zero tool-poisoning. There is no file that both harvests named secrets and posts them to an undocumented host.
Key Reasons
- Zero tool-poisoning findings; src/mcp/McpClient.js is an MCP client making outbound calls, not a set of tool descriptions carrying hidden AI directives
- No findings against .ssh, .aws/credentials, .kube/config or application_default_credentials.json, so there is no credential-harvest half of an exfiltration pair
- All sixteen NET-FETCH sites sit in files matching declared features: AuthManager.js, ProxyLLM.js, apuchat.js, apumail.js, Face.js, ContactCard.js and McpClient.js
- The six critical MCP-TRANSPORT-HARDCODED-TOKEN hits all point at src/i18n/messages/server.js, a translation catalog, which is where a string-matching transport rule misfires
- Endpoint list is dominated by parsed property chains (body.desktop.menu, card.bio, c2.cx, 3ώ.mx) rather than resolvable hosts; the one real host, api.letta.com, is a known agent framework
False Positive Considerations
- IoC extractor reading object property chains and short strings as domains (body.desktop.menu, call.pin, 3ώ.mx, 1-l.sy)
- 489 low-severity code-smell hits that fire on any non-trivial Node.js source
- MCP transport token rule matching translation keys and placeholders in src/i18n/messages/server.js
- Duplicate reporting of the same i18n file as six separate critical secret findings
Reviewed 2026-10-17 · version 0.1.19; recommended action: monitor; model confidence 72%.
MCP version history
Risk trend by version
3 analyzed versions. Each point is the latest successful scan for that version; failed zero-score scans are hidden. Dates: First seen by Risky Plugins.
Pick any point on the chart, then open the Source tab to read that version's code.
Source Code Not Available
Source code is not available for this version of the extension.
About This Extension
Frequently Asked Questions
Similar Extensions
Related extensions from the same publisher or marketplace