ai-test-fixer@1.0.7
AI-powered test-driven code generation and automated test repair CLI and toolkit
AI Security Review
scanned 2d ago · by lpm-firewall-aiThe default path transmits project context to Google Gemini using a bundled API key, while the README presents the no-configuration path as requiring no key and does not disclose Gemini as the default. This is a material, undisclosed runtime data-sharing risk, but the transmitted context serves the package's AI repair feature and no unrelated secret collection was found.
Static reason
One or more suspicious static signals were detected.
Trigger
A user runs the default fix or plan command and a failing test leads to an AI request.
Impact
Test and implementation contents, along with failure details, leave the user's machine for an undisclosed default provider. No unrelated credential theft was identified.
Mechanism
The default configuration selects Gemini; its provider uses an embedded key when no key is configured and posts the generated prompt to Google's API.
AI rationale
The default path transmits project context to Google Gemini using a bundled API key, while the README presents the no-configuration path as requiring no key and does not disclose Gemini as the default. This is a material, undisclosed runtime data-sharing risk, but the transmitted context serves the package's AI repair feature and no unrelated secret collection was found.
Evidence
src/config.jssrc/providers/gemini.jssrc/context-builder.jssrc/providers/prompt.jsREADME.md
Network endpoints1
generativelanguage.googleapis.com
Decision evidence
public snapshotAI called this Suspicious at 91.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- The default configuration selects Gemini.
- Gemini falls back to an API key embedded in the package when no key is configured.
- Gemini sends the generated project prompt to Google's Generative Language API, including the API key in the request URL.
- The prompt includes test contents and implementation-file contents gathered from the project.
- The README describes the no-configuration default as requiring no API key, without disclosing that it uses Gemini.
Evidence against
- The network request occurs during an explicit AI repair or planning command, not during installation or module import.
- The project context sent to the model is used to generate code fixes; there is no evidence here of unrelated credential collection.
Behavioral surface
ChildProcessCryptoEnvironmentVarsFilesystemNetworkShell
HighEntropyStringsUrlStrings
Source & flagged code
6 flaggedsrc/providers/gemini.jsView file
18patternName = google_api_key
severity = high
line = 18
High
High Secret
Package contains a high-severity secret pattern.
src/providers/gemini.jsView on unpkg · L1818patternName = google_api_key
severity = high
line = 18
High
18export const DEFAULT_GEMINI_KEY = 'AIza[redacted]';
export const GEMINI_MODEL_ALIASES = {
'gemini-2.0-flash': 'gemini-3.5-flash',
'gemini-2.5-flash': 'gemini-3.5-flash',
'gemini-2.5-flash-lite': 'gemini-3.5-flash-lite',
'gemini-flash': 'gemini-3.5-flash',
'gemini-
Medium
Ai Review Evidence
Gemini falls back to an API key embedded in the package when no key is configured.
src/providers/gemini.jsView on unpkg · L1834
Findings
2 High6 Medium4 Low
HighHigh Secretsrc/providers/gemini.js
HighSecret Patternsrc/providers/gemini.js
MediumNetwork
MediumEnvironment Vars
MediumAi Review Evidencesrc/config.js
MediumAi Review Evidencesrc/providers/gemini.js
MediumAi Review Evidencesrc/providers/gemini.js
MediumAi Review Evidencesrc/providers/gemini.js
LowScripts Present
LowFilesystem
LowHigh Entropy Strings
LowUrl Strings
Affected versions and remediation
This report applies to ai-test-fixer@1.0.7.
See all versions of ai-test-fixer for other recorded verdicts.
- Review the evidence and your use of ai-test-fixer@1.0.7 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.