create-zuplo-api@7.8.1
Create Zuplo API with one command
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThis is an explicit user-run scaffolder with no install lifecycle hook. It still mutates first-party AI-agent control files and pulls remote skill markdown when the user selects agents, so the publish decision is a warning rather than a block.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
A user runs the create-zuplo-api bin and selects one or more agents, or passes --agents with a non-empty list.
Impact
The new project gains Zuplo agent instructions and an HTTP docs MCP server. A selected Claude setup also turns on all project MCP servers. No install-time or hidden write was identified.
Mechanism
The CLI writes project-local agent files and .mcp.json, enables Claude project MCP servers, and for non-Claude agents downloads the zuplo/tools tarball and copies only SKILL.md files under the project skill directories.
AI rationale
This is an explicit user-run scaffolder with no install lifecycle hook. It still mutates first-party AI-agent control files and pulls remote skill markdown when the user selects agents, so the publish decision is a warning rather than a block.
Evidence
package.jsondist/index.js.mcp.json.claude/settings.jsonCLAUDE.md.github/copilot-instructions.md.cursorrules.windsurfrulesAGENTS.md.claude/skills.agents/skills.cursor/skills.windsurf/skills
Network endpoints2
dev.zuplo.com/mcp/docscodeload.github.com/zuplo/tools/tar.gz/main
Decision evidence
public snapshotAI called this Suspicious at 84.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- package.json exposes a user-run bin at dist/index.js and its scripts are only dev, build, build:public, and test, with no preinstall, install, or postinstall hook.
- Agent setup runs only when the user passes --agents or answers an interactive multiselect; a truthy T path assigns an empty agent list and skips the prompt.
- A non-empty selection writes project-local instruction files, .mcp.json aimed at the Zuplo docs MCP endpoint, and Claude settings that enable project MCP servers and the zuplo/tools GitHub marketplace.
- For agents other than Claude, the CLI downloads the zuplo/tools tarball from GitHub, rejects bodies over 10MB, and copies only SKILL.md files into the new project.
Evidence against
- The None choice and an empty agent list skip those writes.
- The tarball filter keeps only SKILL.md paths and the temp directory is removed afterward.
- No credential harvesting, install-time execution, or self-dependency lifecycle chain was found in the inspected entry.
Behavioral surface
ChildProcessCrypto
Source & flagged code
8 flaggeddist/index.jsView file
1#!/usr/bin/env node
L2: (()=>{var t={7213:(t,u,p)=>{"use strict";Object.defineProperty(u,"__esModule",{value:true});u.createFileSystemAdapter=u.FILE_SYSTEM_ADAPTER=void 0;const m=p(9896);u.FILE_SYSTEM_ADA...
L3: || (${w} == "string" && ${g} && ${g} == +${g})`).assign(E,(0,b._)`+${g}`);return;case"integer":m.elseIf((0,b._)`${w} === "boolean" || ${g} === null
L4: || (${w} === "string" && ${g} && ${g} == +${g} && !(${g} % 1))`).assign(E,(0,b._)`+${g}`);return;case"boolean":m.elseIf((0,b._)`${g} === "false" || ${g} === 0 || ${g} === null`).as...
L5: || ${w} === "boolean" || ${g} === null`).assign(E,(0,b._)`[${g}]`)}}}function assignParentData({gen:t,parentData:u,parentDataProperty:p},m){t.if((0,b._)`${u} !== undefined`,(()=>t....
L6: missingProperty: ${g},
L7: depsCount: ${u},
L8: deps: ${
Critical
Credential Exfiltration
Source appears to send environment or credential material to an external endpoint.
dist/index.jsView on unpkg · L11Trigger-reachable chain: manifest.bin -> dist/index.js
L1: #!/usr/bin/env node
L2: (()=>{var t={7213:(t,u,p)=>{"use strict";Object.defineProperty(u,"__esModule",{value:true});u.createFileSystemAdapter=u.FILE_SYSTEM_ADAPTER=void 0;const m=p(9896);u.FILE_SYSTEM_ADA...
L3: || (${w} == "string" && ${g} && ${g} == +${g})`).assign(E,(0,b._)`+${g}`);return;case"integer":m.elseIf((0,b._)`${w} === "boolean" || ${g} === null
L4: || (${w} === "string" && ${g} && ${g} == +${g} && !(${g} % 1))`).assign(E,(0,b._)`+${g}`);return;case"boolean":m.elseIf((0,b._)`${g} === "false" || ${g} === 0 || ${g} === null`).as...
L5: || ${w} === "boolean" || ${g} === null`).assign(E,(0,b._)`[${g}]`)}}}function assignParentData({gen:t,parentData:u,parentDataProperty:p},m){t.if((0,b._)`${u} !== undefined`,(()=>t....
L6: m
Findings
2 Critical6 High3 Medium4 Low
CriticalCredential Exfiltrationdist/index.js
CriticalTrigger Reachable Dangerous Capabilitydist/index.js
HighChild Processdist/index.js
HighEvaldist/index.js
HighSame File Env Network Executiondist/index.js
HighCommand Output Exfiltrationdist/index.js
HighTrigger Reachable Credential Exfiltrationdist/index.js
HighTrigger Reachable Command Output Exfiltrationdist/index.js
MediumNetwork
MediumEnvironment Vars
MediumStructural Risk Force Deep Review
LowScripts Present
LowFilesystem
Low
Affected versions and remediation
This report applies to create-zuplo-api@7.8.1.
See all versions of create-zuplo-api for other recorded verdicts.
- Review the evidence and your use of create-zuplo-api@7.8.1 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.