The Agent Skills Directory

[SAFE]: The skill integrates with the Birdeye API using standard REST and WebSocket protocols. No malicious commands, persistence mechanisms, or unauthorized data exfiltration patterns were detected.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from the Birdeye API, such as token names, symbols, trader tags, and transaction descriptions. This constitutes a surface for indirect prompt injection.
Ingestion points: Birdeye API response objects processed in templates/birdeye-client.ts and examples/.
Boundary markers: None explicitly defined in the provided client for string data.
Capability inventory: Performs network operations via fetch and ws to established Birdeye endpoints.
Sanitization: Relies on standard JSON parsing; no specific string sanitization for instruction markers is implemented.
[CREDENTIALS_UNSAFE]: The skill uses environment variables (BIRDEYE_API_KEY, SOLANA_PRIVATE_KEY) for authentication. The instructions in SKILL.md explicitly warn against exposing API keys in agent responses, which is a standard security best practice.
[EXTERNAL_DOWNLOADS]: The documentation and examples reference official libraries (@x402/fetch, @solana/web3.js, ws) from public registries to support DeFi and WebSocket functionality. These are legitimate dependencies for the skill's stated purpose.

birdeye