The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The README.md file provides instructions for downloading the skill and its associated toolset from external sources, specifically the visual-reverse-engineer package on NPM and the sundny8/visual-reverse-engineer repository on GitHub. These resources are vendor-owned and part of the documented installation process.
[COMMAND_EXECUTION]: The installation documentation in README.md includes commands for the user to execute during setup, such as npx, npm install, and running local scripts like python scripts/install.py and node bin/cli.js. These are standard methods for configuring the environment for the skill's intended use.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-uploaded images, creating an ingestion surface for indirect prompt injection. This risk is minimized by the skill's instructional nature (it has no dangerous tool-calling capabilities) and the inclusion of specific guardrails in references/uncertainty-and-safety.md, which instruct the agent to use safety-sanitized visual descriptions (e.g., "dramatic fantasy creature" or "intense confrontation") for high-risk content.
Ingestion points: Image analysis workflow described in SKILL.md.
Boundary markers: Not present for image input data.
Capability inventory: No tool calls, shell execution, or network operations are present in the provided skill instructions.
Sanitization: Employs a "Safety Replacement" logic to mitigate the generation of sensitive or prohibited visual descriptions.

image-reverse-analyst