Security audit
Security checks for vulnerabilities and agentic risk
This skill is coherent for document comparison: it sends two user-selected local files to PaodingAI/Calliper and returns Markdown, with no hidden persistence or destructive behavior found.
Install this only if you are comfortable sending the selected document contents and filenames to the configured PaodingAI/Calliper endpoint using your API token. Verify the endpoint-related environment variables before use, avoid regulated or confidential documents unless approved, and prefer installing from a pinned version or reviewed commit when possible.
The skill invokes a Node.js script that reads environment variables and sends local files to a remote API, yet the skill manifest does not declare an explicit tool scope such as permissions or allowed-tools. This weakens reviewability and policy enforcement because consumers cannot easily tell from the manifest that the skill needs environment access and outbound network access.
The skill explicitly uploads two local documents to a remote API for comparison, but the documentation does not prominently warn that document contents leave the local system. In a document-comparison context, files may contain contracts, regulated data, or trade secrets, so missing disclosure materially increases the risk of unintended sensitive-data exfiltration.
The installation command uses npx skills add PaodingAI/skills without a pinned version, which makes installs non-reproducible and exposes users to supply-chain risk if the upstream package or referenced content changes maliciously or unexpectedly. A later compromise of the publisher account or repository could silently alter what gets installed.
The default prompt instructs the agent to always execute a bundled script, regardless of whether the user actually requested document comparison. This creates over-broad tool activation that can trigger unintended file processing and external API calls using available credentials, increasing the risk of unnecessary data exposure and misuse of local files.
The instruction says to return markdown content directly when the user asks for markdown output, and otherwise to extract only the required diff content, which enforces a specific presentation format tied to the skill. While not a strong violation, it does not present any user opt-in for alternative locale or language behavior if such preferences differ.
Detected: suspicious.env_credential_access
