The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download a setup script from https://tempo.xyz/install to /tmp/tempo_install.sh and execute it using bash in SKILL.md. This is done to install the Tempo CLI, which is necessary for the skill's primary operation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted response data from external HTTP 402 challenges to extract parameters like token addresses, amounts, and recipients.
Ingestion points: Handled in SKILL.md and references/credential-construction.md, where the agent parses the WWW-Authenticate header or JSON body of a 402 response.
Boundary markers: The skill contains explicit instructions to enforce strict validation rules on all parsed fields before usage.
Capability inventory: Uses tempo, cast, curl, and jq to execute requests, check balances, and sign transactions based on external inputs.
Sanitization: Employs explicit regex validation patterns for addresses, chain IDs, and amounts, and mandates the rejection of any input containing shell metacharacters like ;, |, &, or newlines. Additionally, it implements mandatory human-in-the-loop confirmation gates (AskUserQuestion) before any transaction execution or cryptographic signing.

pay-with-any-token