The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it ingests and processes external data from blockchain RPC logs, pricing oracle feeds, and hosted API responses.
1. Ingestion points: Data enters the agent's context through transaction logs (debugging.md), market data feeds (pricing-oracle.md), and Payments API responses (private-payments.md).
1. Boundary markers: Textual boundary markers are absent in the prompts, but the skill mandates rigorous code-level validation of all external data.
1. Capability inventory: Capabilities include generating code diffs, executing shell commands for local development/testing, and constructing/submitting blockchain transactions (typescript-setup.md).
1. Sanitization: The skill provides extensive sanitization guidance, such as mandatory freshness and identity checks for oracle feeds and strict escrow authority verification for post-commit actions to prevent unauthorized instruction execution.
[EXTERNAL_DOWNLOADS]: The skill references official MagicBlock development SDKs and tools, including @magicblock-labs/ephemeral-rollups-sdk, @magicblock-labs/ephemeral-validator, and @magicblock-labs/magicsvm. These resources originate from the vendor's verified NPM and GitHub repositories and are standard dependencies for the described development environment.

magicblock