The Agent Skills Directory

[SAFE]: The skill relies on gmgn-cli, an external command-line tool. Since the tool is produced by the same vendor as the skill author (gmgnai), this is classified as standard vendor-provided functionality.
[SAFE]: The skill uses curl https://ipv6.icanhazip.com for network troubleshooting. This is a well-known and legitimate service used to determine the host's external IP address for connectivity debugging.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from blockchain records (e.g., token symbols, caller bios, and wallet names) that could theoretically contain prompt injection payloads. However, the skill implements strong mitigation strategies:
Ingestion points: Data enters the context via gmgn-cli output in SKILL.md.
Boundary markers: The skill mandates wrapping external strings in specific delimiters (「」) and strictly forbids printing the call_thesis field which contains raw caller-written text.
Capability inventory: The skill primarily uses gmgn-cli for data retrieval and does not expose high-risk capabilities to the untrusted data.
Sanitization: The instructions include a comprehensive clean_text() protocol that strips control characters, bidi overrides, zero-width characters, and markdown metacharacters before any external data is presented to the agent.

gmgn-callout-score