The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to use the Bash tool to execute a command-line interface (flyer-maker), which delegates to local Python scripts.
[DYNAMIC_EXECUTION]: The entry point script scripts/run.py implements process replacement via os.execv(). It dynamically calculates the path to a virtual environment Python interpreter (ROOT / ".runners-venv" / "bin" / "python3") and re-executes itself within that environment. This is a common pattern for local tool runners but involves runtime path computation for executable code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input through various CLI flags such as --title, --date, and --location. This data is interpolated into prompt templates for external image generation models.
Ingestion points: Data enters the agent context via command-line arguments --title, --subtitle, --date, --time, --location, and --cta as described in references/event-details.md.
Boundary markers: Prompt templates in references/composition-zones.md use double quotes as delimiters (e.g., Headline: "<TITLE>"), which may not be sufficient to prevent instruction override if the user provides malicious content.
Capability inventory: The skill has the capability to execute shell commands via the Bash tool, write files to the local filesystem, and make network requests to multiple third-party image generation APIs (Ideogram, OpenAI, Google, BFL, ByteDance).
Sanitization: The skill documentation mentions length-based validation to prevent visual distortion, but there is no evidence of sanitization or filtering of control characters or instructions within the user-provided strings.

flyer-maker