Security audit
Security checks for vulnerabilities and agentic risk
This skill openly publishes user-selected static files to agenthost.page and returns share links, with no hidden code or local persistence found.
Install only if you want an agent to publish selected files to agenthost.page. Review folder contents before publishing, avoid secrets or private documents, and treat ownerToken and shareUrl as sensitive because anyone with the right link or token may access or change the hosted content.
The skill is broadly described as a generic way to publish and share files, which can cause an agent to select it for common 'share this' requests without clearly surfacing that data will be uploaded to a third-party hosting service. That ambiguity increases the chance of unintended external disclosure of sensitive files or documents.
The skill does not prominently warn that using it transmits user-provided content to an external service and creates a hosted copy accessible via a share link. In an agent setting, this omission can lead to users or upstream orchestration invoking the skill without informed consent, causing accidental data exfiltration.
This command sends the contents of a local Markdown or HTML file directly to https://agenthost.page/publish, which is an external transmission of potentially sensitive user data. Because the service returns a pre-authenticated share link, the upload immediately creates a remotely accessible copy, raising confidentiality risks if the wrong file is selected or the action is triggered automatically.
Single file — pipe a .md or .html in with its Content-Type. Markdown renders, HTML is served as-is; the file lands at /.
curl -s --data-binary @report.md \
-H 'Content-Type: text/markdown' \
'https://agenthost.page/publish?id=report'
This example packages and uploads an entire directory to an external hosting service, which materially increases the risk of accidental bulk disclosure compared with a single-file publish. Directories often contain unintended files such as drafts, configuration files, source maps, tokens, or internal documentation, and the skill does not describe any filtering or review step before transmission.
Folder — pipe a gzipped tar of the directory.
tar czf - -C ./dist . | curl -s --data-binary @- \
-H 'Content-Type: application/gzip' \
'https://agenthost.page/publish?id=myblog'
This redeploy flow again uploads a full directory externally and additionally handles an ownerToken bearer credential that grants control over the hosted URL. Exposure or misuse of that token could allow unauthorized redeployments or persistent modification of shared content, compounding the confidentiality and integrity risks of the upload itself.
Reuse the ownerToken and username from the first publish:
tar czf - -C ./dist . | curl -s --data-binary @- \
-H 'Content-Type: application/gzip' \
-H 'Authorization: Bearer <ownerToken>' \
'https://agenthost.page/publish?id=myblog&username=<username>'
The documentation explicitly describes accessKey being embedded in shareUrl as a query parameter, which creates a credential-bearing URL that can leak via browser history, copy/paste, screenshots, logs, analytics, referrers, and chat transcripts. In this skill's context, the share link is intended to be handed to humans, which increases the likelihood of broad secret exposure and unintended access to supposedly private content.
The file documents authenticated requests using username in query parameters and elsewhere shows credential use in curl commands, but it does not warn users to avoid placing sensitive values in ways that may be logged or persisted. Since this is API documentation for security-sensitive operations, omission of a user-facing warning about secret handling is a meaningful disclosure gap.
The documentation notes that POST /key/rotate invalidates all existing share links, which is a potentially disruptive and irreversible user-impacting action. Although the behavior is described, there is no explicit warning or caution callout to alert users before performing it.
No suspicious patterns detected.