The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes the !command`` syntax in SKILL.md to execute shell commands automatically when the skill is accessed.
Evidence: The skill executes node -e to parse package.json and find to discover configuration files within the project directory.
This mechanism is used to provide the agent with a 'Live Project Snapshot' without manual user input.
[INDIRECT_PROMPT_INJECTION]: The automated environment scanning process creates a vector where malicious content in local project files could influence agent behavior.
Ingestion points: The shell command in SKILL.md reads the dependencies, devDependencies, and scripts sections of package.json and prints them directly into the agent context.
Boundary markers: There are no explicit delimiters or instructions to the agent to treat the command output as untrusted data.
Capability inventory: The skill specifies allowed-tools including Bash with node -e and find, allowing the agent to perform further file system operations based on the ingested data.
Sanitization: The ingested content from package.json is not sanitized or validated before being injected into the prompt context.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute various package manager and Capacitor CLI commands which modify the local system.
Evidence: Instructions include npm install, npx cap init, npx cap sync, and npx cap run across SKILL.md and the framework-specific reference files.

framework-to-capacitor