The Agent Skills Directory

[PROMPT_INJECTION]: The reference file references/prompt-engineering.md contains a 'Safety Filter Rephrase Strategies' section. This section provides systematic instructions for the AI to circumvent its own safety guardrails (specifically identified as Layer 2 output filters) by using techniques such as abstraction, metaphor, and 'artistic framing' to bypass blocks on public figures, violence, and other restricted content categories.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands via magick (ImageMagick) and ffmpeg for post-processing and format conversion. The instructions (e.g., in SKILL.md and references/post-processing.md) demonstrate direct interpolation of variables into shell strings, such as file naming based on user-provided keywords. This creates a risk of command injection if the agent processes malicious user-supplied strings without proper sanitization.
[EXTERNAL_DOWNLOADS]: The skill requires the execution of a local installation script (./extensions/banana/install.sh) and relies on a third-party MCP server (@ycse/nanobanana-mcp) hosted on GitHub. While these are for functional purposes, they represent a dependency on external code that is not bundled with the skill itself.
[COMMAND_EXECUTION]: The skill refers to and expects the presence of several local Python scripts (cost_tracker.py, presets.py) that are used to manage usage logs and brand presets via the command line.

seo-image-gen