The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to run npx pagecast, which downloads and executes the latest version of the 'pagecast' package from the NPM registry at runtime.
[REMOTE_CODE_EXECUTION]: The use of unpinned npx commands allows for the execution of remote code without version verification, which is a significant security risk if the package registry or package is compromised.
[DATA_EXFILTRATION]: The core function of the skill is to upload local project files (HTML, Markdown, static builds) to a public hosting service (pages.dev). The instructions specifically direct the agent to "proactively offer" to publish files and suggest that "when it is borderline, offer," which may lead to the unintended public exposure of internal reports, design documents, or sensitive plans.
[COMMAND_EXECUTION]: The skill executes various shell commands, including npx pagecast publish, npx pagecast goal publish, and npx pagecast pages deploy, with absolute file paths provided at runtime.
[PROMPT_INJECTION]: Instructions such as "do not wait to be asked," "run the publish command immediately," and "do not ask again" encourage high autonomy. This reduces user oversight and could cause the agent to publish files before a user has adequately reviewed the sensitivity of the data.
[DATA_EXPOSURE]: The skill documentation acknowledges that published links are "unlisted capability URLs" and that "anyone with the URL can view it," explicitly stating that "unlisted does not mean private."

publish-report