The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation script scripts/install.sh and the SKILL.md documentation utilize curl | bash to download and execute code from https://raw.githubusercontent.com/HKUDS/Upskill/. This allows arbitrary code execution from a repository outside of the trusted vendor list.
[COMMAND_EXECUTION]: The skill installs multiple slash commands that execute shell scripts located in the user's home directory (~/.claude/hooks/). It also manages a build pipeline that creates git worktrees and executes code autonomously to validate model performance.
[COMMAND_EXECUTION]: The /upskill-configure command and scripts/install.sh modify ~/.claude/settings.local.json to register persistent SessionStart, SessionEnd, and UserPromptSubmit hooks. These hooks execute custom logic automatically at the beginning and end of every agent session, creating a persistence mechanism within the agent environment.
[EXTERNAL_DOWNLOADS]: Fetches implementation scripts and configuration resources from the third-party HKUDS/UpSkill GitHub repository during installation and updates.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes session.log files containing potentially untrusted data from previous agent interactions. If a session log contains malicious instructions, the 'Teacher' model analyzing the failure could be influenced to generate compromised skills.

upskill