The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection as it ingests untrusted data from external files which are then processed by an LLM.
Ingestion points: The script scripts/gemini_image_gen.py reads content from files specified by the --input-image, --prompt-file, and --jobs (JSONL) command-line arguments.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to help it distinguish between system instructions and content within the ingested files.
Capability inventory: The skill has the capability to write files to the filesystem and make network requests to the Google Gemini API.
Sanitization: The skill does not perform sanitization of input file content or image data before passing it to the SDK.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of a third-party Python library at runtime.
Evidence: SKILL.md and references/cli.md instruct the use of uv run --with google-genai to execute the skill's logic.
Source: The package is the official SDK from a well-known technology service provider and is required for the skill to communicate with the Gemini API.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a bundled Python script to perform its tasks.
Evidence: Documentation in SKILL.md and references/cli.md provides multiple examples of executing scripts/gemini_image_gen.py via uv run.
Context: These executions are scoped to the project and are used to provide the main functionality of the skill without requesting elevated system privileges.

gemini-imagegen