The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands (ls, find) to locate a workflow script. These commands target specific configuration directories (~/.claude, ~/.codex) and perform deep searches in the current working directory.
[REMOTE_CODE_EXECUTION]: The skill uses the Workflow tool to execute a JavaScript file (improve.js) at a path determined dynamically at runtime. This search-and-execute pattern could lead to the execution of unauthorized code if a malicious script is placed in a directory scanned by the find command, particularly within the project root.
[PROMPT_INJECTION]: The skill functions as a wrapper for an autonomous loop that ingests external code. It lacks explicit instructions for sanitization or boundary markers (such as XML tags or delimiters) to prevent malicious instructions within the target code from influencing the reviewer or fixer agents.
[REMOTE_CODE_EXECUTION]: The skill gives the agent high autonomy to poll task outputs and continue execution without user intervention. This autonomous behavior, combined with the dynamic loading of scripts, increases the impact of any potential code execution vulnerability.

code-improver