The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands such as grep, pnpm install, and pnpm tsc. These are used for surveying the project codebase, installing library dependencies, and validating migration progress through type checking. These operations are standard for a developer-oriented coding agent and are restricted to project-related tasks.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of legitimate packages including @sodax/wallet-sdk-react, @tanstack/react-query, and @sodax/dapp-kit from the official NPM registry. It also includes examples that connect to well-known public blockchain RPC endpoints (e.g., Sonic, Ethereum, Solana). No downloads from untrusted or unknown sources were detected.
[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read and process untrusted user source code during migration and integration tasks. This presents a surface for indirect prompt injection where malicious instructions embedded in a user's codebase could attempt to subvert the agent's migration logic.
Ingestion points: migration-v1-to-v2/knowledge/ai-rules.md (Survey and Typecheck workflows) and integration/knowledge/ai-rules.md (Survey workflow).
Boundary markers: Not explicitly defined within the instruction files.
Capability inventory: The agent has permissions to execute terminal commands, modify the filesystem, and install packages.
Sanitization: The skill does not implement specific data sanitization but provides a strict protocol (Stop Conditions) requiring the agent to halt and ask the user for confirmation when encountering ambiguous or custom implementation patterns.

sodax-wallet-sdk-react