The Agent Skills Directory

[SAFE]: The skill is primarily instructional documentation (scope: dev, internal: true) intended for developers. It provides code examples for React hooks and agent scripts that interact with a localized application state.
[PROMPT_INJECTION]: The skill describes an architecture where agents ingest data from the UI (URL and navigation state). This presents a surface for indirect prompt injection (Category 8). However, the skill explicitly mentions the use of framework-level boundary markers (<current-screen> and <current-url> blocks) which serve to isolate untrusted data from system instructions.
Ingestion points: readAppState calls in actions/view-screen.ts.
Boundary markers: Explicitly mentioned as auto-injected <current-screen> and <current-url> blocks.
Capability inventory: Data fetching via domain actions (fetchEmailList) and state modification via writeAppState in navigate scripts.
Sanitization: Not explicitly detailed in the templates, but the design relies on the agent platform's internal block-parsing logic.
[COMMAND_EXECUTION]: The provided code snippets use standard JavaScript/TypeScript and React patterns. No unsafe command execution, subprocess spawning, or shell injections were found. Operations are limited to internal state management and logging.

context-awareness