The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the find utility to search for a script named comet-env.sh across several sensitive locations, including the user's root home directory, hidden configuration folders ($HOME/.config), and AI-specific directories ($HOME/.gemini).
[REMOTE_CODE_EXECUTION]: The script found by the search is immediately executed using the shell's source command (. "$COMET_ENV"). This allows for arbitrary code execution if a malicious or tampered script exists in any of the searched paths.
[COMMAND_EXECUTION]: The skill executes multiple shell commands using dynamically defined variables ($COMET_BASH, $COMET_STATE, $COMET_GUARD, $COMET_ENV) which are initialized by the sourced script. This runtime dependency makes it difficult to verify the safety of the commands being run.
[PROMPT_INJECTION]: The skill instructs the agent to operate in a "one-time continuous execution" mode, explicitly stating it "must automatically advance through hotfix steps, without pausing to wait for user input mid-way." This configuration removes human-in-the-loop oversight for the automated shell operations.
[COMMAND_EXECUTION]: The skill automates a loop that reads a file (tasks.md) and executes code modifications, formatting commands, and tests based on the file's content, which serves as an attack surface for indirect prompt injection if the task file is influenced by untrusted external data.

comet-hotfix