finishing-a-development-branch

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION] (MEDIUM): The skill constructs shell commands by directly interpolating potentially untrusted strings such as branch names and PR titles.\n
  • Ingestion points: The skill uses variables like <feature-branch>, <base-branch>, and <title> derived from the environment or user input.\n
  • Capability inventory: Executes git merge, git branch -d, gh pr create, and git worktree remove using subprocess calls.\n
  • Sanitization: No escaping or validation is performed on these variables before they are passed to the shell.\n
  • Risk: An attacker could craft a malicious branch name (e.g., "; rm -rf /;") to execute arbitrary commands when the agent attempts to merge or delete the branch.\n- [DATA_EXFILTRATION] (LOW): The skill performs network operations to push local data to external repositories.\n
  • Evidence: Option 2 uses git push -u origin <feature-branch> and gh pr create to move local code and commit history to a remote server.\n
  • Risk: While this is the intended purpose of the skill, it represents a data exfiltration vector if an agent is tricked into pushing sensitive information to an attacker-controlled repository.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 12:46 PM