The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill fetches code from an external repository and executes scripts contained within it.
Evidence: In Step 4, the skill uses git clone to download content from https://github.com/garrytan/gstack.git and then executes ./setup.
Evidence: In Step 4.75, the skill iterates through scripts in the gstack-upgrade/migrations directory and executes them using bash "$migration".
[COMMAND_EXECUTION]: The skill performs extensive shell command execution to manage the installation environment and upgrade flow.
Evidence: Multiple sections use bash to perform system-level operations such as mktemp, rm -rf, mv, and cp -Rf on directories within the user's home and project paths.
[EXTERNAL_DOWNLOADS]: The skill downloads external resources from a third-party source during the upgrade process.
Evidence: Step 4 uses git fetch and git clone targeting https://github.com/garrytan/gstack.git to retrieve new versions of the software.
[PROMPT_INJECTION]: The skill processes external data that could potentially influence agent behavior through indirect injection.
Evidence: Step 6 instructs the agent to read CHANGELOG.md from the downloaded repository and summarize it for the user. If this file contains malicious instructions, the agent might follow them while processing the text.

gstack-upgrade