The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates sensitive blockchain treasury operations and accepts untrusted data (such as names, descriptions, and addresses) which could be used to manipulate agent behavior. This is a inherent surface for indirect prompt injection.
Ingestion points: User-supplied values for the --description, --name, and --metadata flags, as well as configuration files in scripts/grant-config.sh, scripts/create.sh, and scripts/fee-config.sh.
Boundary markers: None explicitly implemented in the shell scripts for text delimitation.
Capability inventory: Full treasury lifecycle management on the Xion blockchain, including funding, withdrawing tokens, and managing Authz/Fee grants via the xion-toolkit CLI.
Sanitization: Input validation is handled by the xion-toolkit CLI. Additionally, the fund.sh and withdraw.sh scripts implement hardcoded shell-level confirmation prompts to ensure human-in-the-loop verification for financial transactions.

xion-treasury