The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads external binaries and scripts during setup, including the claude binary from claude.ai and the codex CLI from OpenAI.
[REMOTE_CODE_EXECUTION]: The cloud setup script executes a remote shell script from GitHub via curl | sh to install the worktrunk maintainer tool.
[COMMAND_EXECUTION]: The agent executes numerous shell commands using the gh and git CLIs to perform repository operations, and Python's subprocess module is used extensively in helper scripts.
[PRIVILEGE_ESCALATION]: proxy/setup_sandbox.py uses sudo to manage a dedicated non-privileged user and restricted file system permissions as part of its security sandbox implementation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from PRs and issues. It implements a security model involving configuration pinning and file restoration from the base branch to mitigate the risk of a malicious PR overriding agent instructions.
[DYNAMIC_CONTEXT_INJECTION]: The repository includes a specific pre-commit hook designed to detect and prevent unauthorized shell command execution through preprocessor syntax in skill markdown files.

fable-tend