gitlab-cherrypick
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using
gitandglabthat include variables derived from user input, such as the number of days for commit history (<N>) and commit identifiers for cherry-picking. While the workflow requires user confirmation, these inputs are interpolated into shell commands, presenting a risk if the execution environment does not strictly validate the input strings.\n- [INDIRECT_PROMPT_INJECTION]: The skill reads and processes external data in the form of commit subjects and author names from themainbranch. This metadata is subsequently displayed to the user and used to generate Merge Request descriptions.\n - Ingestion points: Commit history data retrieved via
git loginSKILL.md.\n - Boundary markers: Absent. There are no explicit instructions for the agent to use delimiters or ignore potentially malicious instructions embedded within commit messages.\n
- Capability inventory:
git cherry-pick(modifies local source code),git push(updates remote repositories), andglab mr create(creates Merge Requests containing the untrusted data).\n - Sanitization: Absent. Commit subjects and descriptions are used directly in formatted reports and PR descriptions without escaping or filtering.
Audit Metadata