The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The file references/data-queries.md instructs the agent to run pip install -r ../findata-toolkit-cn/requirements.txt. Because this file is located outside the skill's own directory, its contents are not verifiable during analysis, presenting a supply chain risk where malicious packages could be installed.- [COMMAND_EXECUTION]: The skill relies on executing Python scripts from a sibling directory (../findata-toolkit-cn/scripts/views_runner.py) to function. Executing code that is not bundled with the skill itself is a security risk if the external environment is compromised or if the scripts do not properly sanitize user-provided inputs.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze text from financial reports and notes fetched from external websites like Sina and Eastmoney. These external sources are untrusted and could contain hidden malicious instructions designed to bias the analysis. 1. Ingestion points: Financial statement data fetched via external scripts. 2. Boundary markers: The instructions lack explicit delimiters to separate untrusted external text from the agent's instructions. 3. Capability inventory: The agent can perform complex financial modeling and generate long-form reports. 4. Sanitization: No sanitization or filtering of external content is specified before the agent processes it.- [METADATA_POISONING]: The references/data-queries.md file contains a redundant copy of the skill's description field embedded in the body of the documentation. This repetition could be used to manipulate the model's focus or reflects poor control over the skill's metadata and instructions.

financial-statement-analyzer