The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data (addresses, amounts, token symbols) to construct transaction URLs. While this creates a potential surface for indirect prompt injection, the skill includes strong defensive instructions.
Ingestion points: User-supplied addresses, amounts, and CSV payloads interpolated into URL parameters in SKILL.md.
Boundary markers: The skill explicitly instructs the agent to explain the transaction in plain words, warn the user to check all details, and advise them to verify the SafeTx hash in their wallet independently.
Capability inventory: The skill itself has no execution capabilities; it serves solely to construct strings/links for the user.
Sanitization: Instructs the use of encodeURIComponent to ensure parameters are correctly formatted within the generated links.
[COMMAND_EXECUTION]: The skill documentation includes instructions for a developer workflow involving a helper script (node scripts/tx-link.mjs) found in the vendor's repository. This is an instructional tool intended for advanced users to run locally and does not constitute a malicious execution pattern.

safe-wei