The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes dynamic context markers to execute shell commands at load time for environmental awareness.
Evidence: A node script in SKILL.md reads package.json to extract specific plugin metadata and Capacitor dependencies.
Evidence: A find command in SKILL.md identifies directory structures and configuration files (e.g., capacitor.config.json, ios, android).
Context: These operations are diagnostic in nature, intended to provide the agent with the current state of the plugin project to facilitate a successful upgrade.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from local project files, which represents a potential attack surface if those files contain malicious instructions.
Ingestion points: package.json and project directory structures.
Boundary markers: The extracted data is presented as a structured snapshot in the agent's context.
Capability inventory: The skill is authorized to use npm, npx, and node through the Bash tool for project synchronization and verification.
Sanitization: The initialization script uses specific logic to filter for @capacitor/ dependencies rather than ingesting the full raw content of package.json, which significantly reduces the risk of schema confusion or instruction injection.

capacitor-plugin-upgrades