The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs shell commands by interpolating variables like ${CONTEXT} and ${USER_QUESTION} directly into double-quoted strings (e.g., gemini -p "${CONTEXT}"). This pattern is vulnerable to command injection if the interpolated content contains shell metacharacters such as double quotes, backticks, or dollar signs. Since ${CONTEXT} includes output from git diff, which reflects the content of local files, an attacker could potentially execute arbitrary commands by modifying a file and triggering a review.
[PROMPT_INJECTION]: The skill ingests data from external sources, including git diff output and information from advertising or SEO tools, and includes this data directly in the prompt sent to the Gemini model.
Ingestion points: git diff HEAD and MCP tool outputs (e.g., mcp__google_ads_mcp__*).
Boundary markers: None. The content is placed within a text-based prompt without delimiters or "ignore embedded instructions" warnings.
Capability inventory: Execution of the gemini CLI tool and various git commands.
Sanitization: None. Data is passed as raw strings to the model.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the @google/gemini-cli package from NPM. This is a required dependency for the skill's core functionality.

gemini