The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests market and narrative data from third-party crypto providers, creating an indirect prompt injection attack surface. This is mitigated by best-practice instructions and script-level validation. Evidence: 1. Ingestion points: External API responses from CoinGecko, DefiLlama, and AIXBT (SKILL.md, references/aixbt.md). 2. Boundary markers: Instructions to distinguish aggregator observations from executable quotes and handle missing data as unknown (SKILL.md, references/methodology.md). 3. Capability inventory: A local Node.js script (scripts/price-check.mjs) performs network fetches, and the skill references external tools for broader data discovery. 4. Sanitization: The diagnostic script implements a strict identifier regex, caps responses at 64KiB, and uses fatal text decoding to prevent malformed payload processing.
[COMMAND_EXECUTION]: The skill documentation instructs the agent to execute a provided local utility, scripts/price-check.mjs, for price verification. The script is part of the vendor-supplied skill and contains safety constraints including timeouts, redirect blocking, and explicit credential omission.
[SAFE]: The skill demonstrates a high security posture by emphasizing credential safety, redacting network errors to prevent data leaks, and providing deterministic diagnostic tools. No obfuscation, persistence mechanisms, or unauthorized privilege escalation patterns were detected.

galleon-defi-data