The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements tools for ingesting data from external network sources, creating an attack surface for indirect prompt injection. \n
Ingestion points: Response data from fetch (referenced in references/axios-adapter.md and references/prefetching.md) and messages from NitroWebSocket (referenced in references/using-websockets.md and references/migrate-from-rn-ws.md). \n
Boundary markers: The skill does not provide or recommend the use of boundary markers or instructions to ignore embedded commands within the network payloads. \n
Capability inventory: The skill utilizes network capabilities via fetch and WebSocket connections, and it can write data to the filesystem as seen in the Hermes profiler implementation (references/perfetto-profiling.md). \n
Sanitization: No sanitization or validation routines for external content are described in the provided recipes. \n- [DATA_EXFILTRATION]: The NetworkInspector utility (detailed in references/network-inspector.md) captures complete request and response data, including sensitive authentication headers and bodies. \n
The skill provides a recipe to copy these entries as curl commands to the system clipboard (Clipboard.setString(entry.curl)), which could lead to the exposure of credentials if not handled carefully by the user. \n- [COMMAND_EXECUTION]: The skill includes instructions for executing standard platform-specific build and dependency management commands. \n
Examples include pod install for iOS and ./gradlew clean for Android, as documented in the installation and profiling sections (SKILL.md and references/perfetto-profiling.md). \n- [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of official networking packages from public registries and references documentation from known developer repositories. \n
Referenced packages include react-native-nitro-fetch and react-native-nitro-modules, which are legitimate tools provided by the skill author.

nitro-fetch