The Agent Skills Directory

[COMMAND_EXECUTION]: Automatically executes git add . to stage all modified and untracked files when the skill is triggered. While the skill includes logic to intelligently judge file safety and unstage known bad patterns (like .env or id_rsa), the initial aggressive staging of all files without prior review increases the surface area for accidental commitment of sensitive data if the filtering logic fails or is incomplete.
[DATA_EXFILTRATION]: The skill attempts to collect metadata about the agent's environment by reading the configuration file ~/.trae/traecli.yaml and checking environment variables like $ANTHROPIC_MODEL and $ANTHROPIC_BASE_URL. This data is formatted into a Co-authored-by trailer and included in the final commit message. If the user proceeds with the git push option, this internal environment metadata is exfiltrated to the remote Git server.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests untrusted code content from the repository through git diff --cached (referenced in SKILL.md).
Boundary markers: The skill does not define explicit delimiters or instructions to ignore embedded prompts within the code changes being analyzed.
Capability inventory: Uses Bash for shell execution and AskUserQuestion for user interaction.
Sanitization: There is no documented process for sanitizing or escaping the content retrieved from code diffs before passing it to the LLM for analysis and message generation.

git-cmsg