The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The instructions in SKILL.md command the agent to download a JavaScript file from https://graceful-tiger-715.convex.site/check-updates.mjs and execute it using node. This facilitates the execution of unverified remote code on the local environment.
[COMMAND_EXECUTION]: The skill invokes shell commands such as curl, node, and git. It also explicitly instructs the agent to run "migration.gate" commands that are provided by the external remote script, allowing for the execution of arbitrary shell strings.
[EXTERNAL_DOWNLOADS]: The skill performs downloads from a user-controlled subdomain (graceful-tiger-715.convex.site), which is not a trusted software distribution channel.
[INDIRECT_PROMPT_INJECTION]: The skill processes complex migration steps and command sequences defined by a remote source. This design is vulnerable to manipulation by the external resource to control agent behavior.
Ingestion points: The content retrieved from the remote URL https://graceful-tiger-715.convex.site/check-updates.mjs.
Boundary markers: None present; the instructions do not include delimiters or warnings to ignore malicious instructions within the data.
Capability inventory: Shell access for downloading, executing scripts, and performing git operations.
Sanitization: No validation or filtering is applied to the commands or steps received from the remote source before they are acted upon.

convex-check-updates