The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill logic in SKILL.md interpolates user-provided variables ${BRANCH} and ${CUSTOM_DIR} directly into a shell command template used by the Bash tool. Because there is no sanitization of these strings, an attacker can provide input containing shell metacharacters (e.g., ;, &, |) to execute unauthorized commands on the host system.
[EXTERNAL_DOWNLOADS] (MEDIUM): The scripts/download_openharmony.sh script downloads the repo tool from https://raw.gitcode.com/gitcode-dev/repo/raw/main/repo-py3. GitCode is not on the list of trusted external sources provided in the security guidelines, introducing a third-party dependency risk.
[REMOTE_CODE_EXECUTION] (HIGH): The skill downloads the repo utility, marks it as executable using chmod a+x, and runs it via repo init. This pattern of downloading and executing unverified remote code is a high-risk behavior that bypasses static analysis of the skill's actual runtime actions.
[PRIVILEGE_ESCALATION] (MEDIUM): The script scripts/download_openharmony.sh performs a chmod a+x on the downloaded repo tool. Additionally, the skill documentation prompts users to execute sudo apt-get install commands, which involves privilege elevation.

openharmony-download