The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a series of shell commands and scripts to manage remotes, branches, and GitCode operations.
[DYNAMIC_EXECUTION]: The skill constructs shell commands using dynamic variables derived from the local environment and remote repository metadata. Evidence: Variables such as $BRANCH_NAME, $FORK_OWNER, and $ISSUE_NUMBER are interpolated into shell commands (e.g., git push -u $FORK_REMOTE <branch-name> and sed operations). If these values (especially branch names) contain shell metacharacters, they could facilitate command injection attacks when executed by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from the repository environment, which is then used in security-sensitive operations.
Ingestion points: Data is collected from git log (commit messages), git diff (code changes), and remote PR templates fetched via git show $UPSTREAM_REMOTE/master:.gitee/PULL_REQUEST_TEMPLATE.zh-CN.md.
Boundary markers: The instructions do not define clear delimiters or "ignore" instructions for processing these external inputs.
Capability inventory: The skill has the capability to execute shell commands, perform network operations via git push, and modify repository state through gitcode MCP tools.
Sanitization: There is no evidence of sanitization or validation of the ingested strings before they are used as parameters for creating issues or pull requests.
[EXTERNAL_DOWNLOADS]: The skill fetches external content from remote repositories to populate PR templates. Evidence: The workflow attempts to retrieve .gitee/PULL_REQUEST_TEMPLATE.zh-CN.md from the upstream OpenHarmony repository using git fetch and git show if a local copy is not present.

gitcode-pr