The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses a modular architecture where 'sub-skills' (additional instruction sets) are loaded dynamically at runtime. It resolves local filesystem paths based on environment variables and directory searches, then uses cat to include the sub-skill content into the prompt context. This enables flexible design logic but depends on the integrity of the local skill directory.
[INDIRECT_PROMPT_INJECTION]: The skill performs project-wide scanning of configuration files and source code to identify frameworks and libraries. This behavior involves ingesting untrusted data from the project files into the agent's context for processing.
Ingestion points: Reads package.json, build.gradle, Package.swift, and various source files (.swift, .html, .css, etc.) to determine dependencies and platforms.
Boundary markers: Not explicitly defined in the detection scripts.
Capability inventory: The agent has access to Bash, Read, Write, Edit, and WebSearch tools.
Sanitization: The skill utilizes grep and ls for pattern matching to map the tech stack rather than direct execution of file contents.

paint