The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents workflows involving untrusted external data, such as XRPL payment references and FDC attestation payloads.
Ingestion points: Untrusted data enters the agent context from XRPL transaction memos and FDC verifier responses as described in SKILL.md and direct-minting-guide.md.
Boundary markers: The SKILL.md file contains explicit security sections warning developers to treat this data as opaque binary protocol information and never as free-form text or AI input.
Capability inventory: The skill includes scripts for executing blockchain transactions (reserve-collateral.ts, execute-minting.ts, redeem-fassets.ts) and performing network requests to verifier APIs.
Sanitization: Instructions emphasize decoding data strictly according to fixed binary formats and contract ABIs, ensuring a clear separation between external data and execution logic.
[COMMAND_EXECUTION]: The skill provides various TypeScript resource scripts for protocol interaction. These are intended for developer use and follow best practices, such as requiring environment variables for sensitive keys and providing dry-run modes for state-changing transactions.
[EXTERNAL_DOWNLOADS]: The execute-minting.ts script performs network requests to fetch attestation proofs from Flare Data Connector (FDC) verifiers. The target URLs are configured via environment variables (VERIFIER_URL_TESTNET, COSTON2_DA_LAYER_URL), representing standard protocol communication with infrastructure services.

flare-fassets