The Agent Skills Directory

[DYNAMIC_EXECUTION]: The reference implementations for mathematical calculation utilize the unsafe JavaScript eval() function, which can execute arbitrary code if the input strings are not strictly controlled.
Evidence: references/algebra.md includes eval(result) and eval(current) within the evaluateExpression and showSteps functions. If an agent incorporates these templates into generated artifacts using user-provided input, it could lead to Cross-Site Scripting (XSS) or local code execution in the browser context.
[DATA_EXFILTRATION]: The skill configuration discloses sensitive host-level information by including absolute file paths related to the author's local development environment.
Evidence: SKILL.md contains hardcoded local paths such as /Users/jamesrochabrun/Desktop/skills/skills/math-teacher/scripts/, which reveals the internal directory structure and the username of the developer.
[COMMAND_EXECUTION]: The skill instructions and scripts enable the agent to perform file system operations and interact with the host operating system's application layer.
Evidence: SKILL.md explicitly directs the agent to use the Bash tool to execute local shell scripts.
Evidence: The scripts scripts/generate_game.sh and scripts/generate_playground.sh perform directory creation (mkdir), file writing (cat >), and use the open command to launch external applications (typically a web browser) on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by processing untrusted user input to generate standalone executable artifacts without sufficient sanitization or boundary enforcement.
Ingestion points: User-provided math concepts and queries defined in the usage patterns of SKILL.md.
Boundary markers: None are specified to delimit user content or prevent the agent from misinterpreting adversarial input during artifact generation.
Capability inventory: The skill can write files to the local disk and trigger the opening of these files in a browser, where the aforementioned eval() risks may be triggered.
Sanitization: No sanitization logic or validation steps are present in the scripts or the reference code templates to verify the integrity of the math expressions before they reach executable contexts.

math-teacher