The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes data from external sources like the Hyperliquid leaderboard (trader tags, asset names, momentum events). This represents an indirect prompt injection surface where a malicious actor on the exchange could attempt to influence agent behavior through crafted metadata.
Ingestion points: scripts/bald-eagle-scanner.py retrieves data using mcporter_call for markets, top traders, and momentum events.
Boundary markers: Findings are output as structured JSON, but string-based data from the exchange (e.g., trader tags) is not explicitly delimited to prevent the agent from interpreting the content as instructions.
Capability inventory: The skill has the capability to initiate trades, set leverage, and manage risk parameters based on the scanner's output.
Sanitization: Numerical data is sanitized via safe_float and safe_int, but string-based metadata lacks explicit sanitization against potential injection patterns.
[EXTERNAL_DOWNLOADS]: scripts/bald_eagle_config.py implements a fallback mechanism for MCP tool calls using the requests library. This mechanism sends data to an API base defined in environment variables, which is consistent with the vendor's (Senpi-ai) infrastructure for skill execution.

bald-eagle