The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze content from external URLs using tools like agent-browser and tavily/extract, which exposes the agent to untrusted data.
Ingestion points: Data enters the agent's context through browser screenshots and text extraction from competitor websites (e.g., https://competitor.com/pricing).
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present in the analysis prompts.
Capability inventory: The skill utilizes belt app run which includes a Python executor, browser automation, and search tools.
Sanitization: The skill does not implement sanitization or filtering of the retrieved web content before it is processed by the model.
[DYNAMIC_EXECUTION]: The skill includes a hardcoded Python script to generate positioning maps using the matplotlib library via the infsh/python-executor tool.
Evidence: A script in SKILL.md dynamically creates a PNG image based on competitor data points provided within the prompt context.
[COMMAND_EXECUTION]: The skill heavily relies on executing shell commands via the belt CLI tool to perform market research and product analysis.
Evidence: Multiple instances of belt app run are documented for research, funding lookups, and UX reviews.
[EXTERNAL_DOWNLOADS]: The skill references the installation of external tools and provides links to remote documentation/scripts.
Evidence: Documentation points to npx skills add belt-sh/cli and an installation guide hosted on raw.githubusercontent.com/inference-sh/skills.

competitor-teardown