The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts and standard system utilities such as git and gh to handle project cloning, branch management, and the creation of pull requests on GitHub.
[DATA_EXFILTRATION]: The skill reads conversation history logs from ~/.claude and ~/.codex to include them in the research submission. This data is transmitted to a repository owned by a well-known organization. The workflow explicitly informs the user about this collection and requires their consent before proceeding.
[INDIRECT_PROMPT_INJECTION]: The prepare-submission.sh script parses the RESEARCH_CHALLENGE.md file to identify which project artifacts to copy. Although this creates a potential surface for path traversal if the file content is manipulated, the skill includes a dedicated workflow for generating this file and includes path filters.
[SAFE]: The skill targets an official repository of the authoring organization. The inclusion of hardcoded excludes for directories like .git and .venv helps prevent accidental exposure of credentials or environment secrets during the submission process.

research-challenge