The Agent Skills Directory

Command Execution: The skill executes several local scripts and tools to verify project state, check versions, and perform type-checking using the shell.
Evidence: node "${PLUGIN_ROOT}/scripts/check-version.js" is used in SKILL.md to ensure the environment is ready.
Evidence: The skill uses npx --no-install tsc --noEmit to validate that generated TypeScript code compiles correctly without installing new dependencies.
Dynamic Context Injection: The main skill file utilizes a shell command execution pattern that runs at load time to replace content with script output.
Evidence: !node "${PLUGIN_ROOT}/scripts/check-version.js" in SKILL.md executes a version check script when the skill is accessed.
Indirect Prompt Injection Surface: The skill ingests external data from project configuration files and remote Dataverse entity definitions to drive its logic.
Ingestion points: Data is read from package.json, power.config.json, and fetched from Dataverse EntityDefinitions via a helper script.
Boundary markers: The skill processes this data primarily as structured JSON and enforces strict schema checks on the resulting metadata.
Capability inventory: The skill has the ability to execute shell commands (bash), create or edit files (write, edit), and invoke other internal skills.
Sanitization: Inputs such as capability names are normalized to a standard lowercase hyphenated format before being used for routing or file naming.

add-native