The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documentation defines the attack surface for indirect prompt injection via the Bounded SDK's data ingestion capabilities.
Ingestion points: Data enters the application context through SDK retrieval methods such as get, subscribe, runQuery, and functions.invoke as documented in docs/sdk-reference.md and docs/building-a-webapp.md.
Boundary markers: The documentation repeatedly highlights that all operations are governed by Bounded's server-side policy engine (policy.json), which enforces rules and invariants at the platform level, providing a security boundary separate from the client code.
Capability inventory: The SDK exposes powerful capabilities for data manipulation, including atomic multi-writes (setMany), individual document updates (set), and file storage operations (setFile).
Sanitization: The skill emphasizes the use of hosted authentication issuers (auth.bounded.sh), which prevents the application frontend from directly handling or potentially mishandling sensitive user credentials like email OTP codes.
[SAFE]: The skill provides explicit guidance on security hardening, such as encrypting mobile session stores using the OS keychain (expo-secure-store or react-native-keychain) and providing a specific resolution recipe for a moderate security advisory in a transitive dependency (uuid) within the Solana web3.js stack.
[SAFE]: All external libraries and services referenced (e.g., @bounded-sh/client, @bounded-sh/server, expo-*, Privy) are well-known, established tools or official vendor resources from the skill's authoring organization.

bounded-frontend