The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a custom Python script (scripts/collect.py) to invoke the longbridge command-line interface. The implementation employs subprocess.run with arguments passed as a structured list, effectively preventing shell injection vulnerabilities. The script also verifies the presence of the required CLI tool before execution to ensure environment safety.
[EXTERNAL_DOWNLOADS]: The agent fetches financial datasets (income statements, consensus estimates, news, and segments) from the Longbridge platform using its official CLI and MCP server tools. These operations are limited to data acquisition for financial reporting and analysis purposes.
[INDIRECT_PROMPT_INJECTION]: The skill processes external information, including financial statements from the longbridge CLI and earnings call transcripts obtained via web searches. This creates a potential surface for indirect prompt injection if the external data were to contain adversarial instructions.
Ingestion points: Data collected via scripts/collect.py and call transcripts retrieved through web searches.
Boundary markers: No explicit delimiters or instructions to ignore embedded prompts are applied to the external content before it is processed into the final reports.
Capability inventory: The skill possesses file system access (writing to and reading from a temporary directory for raw JSON data) and the ability to execute the pre-installed longbridge CLI binary.
Sanitization: The symbol argument provided by the user is sanitized in the Python script to remove whitespace and format it correctly for file paths (e.g., replacing dots with underscores) and CLI arguments, mitigating potential path-based attacks.

stock-earnings