The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the codex command-line tool using a dedicated wrapper script (scripts/imagegen.sh). This tool execution is the primary documented function of the skill and relies on a trusted prerequisite CLI tool.
[INDIRECT_PROMPT_INJECTION]: The skill features an indirect prompt injection surface, as it interpolates user-supplied product details into complex prompt templates for image generation.
Ingestion points: User descriptions and style preferences are ingested in the Step 1 (Intent Recognition) and Step 3 (Prompt Assembly) workflows defined in SKILL.md.
Boundary markers: The skill employs a structured JSON format and task-specific instructional wrappers to isolate user input from the model's core generation instructions.
Capability inventory: The skill uses Bash to run a local script, which in turn interacts with the codex tool and a local HTTP service (127.0.0.1:4312).
Sanitization: The scripts/imagegen.sh script specifically utilizes jq -Rs . to properly escape user-provided strings before inclusion in JSON payloads, effectively preventing control character injections into the backend generation engine.

ecom-image2