The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file scripts/generate_deep_research_report.py contains a hardcoded API key (em_fjFqd4YB6Cqs52LF48XWbMDdLNq6MyNg) used to authenticate with the EastMoney API. Hardcoding secrets in scripts is insecure and can lead to unauthorized access if the skill is shared.
[COMMAND_EXECUTION]: The SKILL.md instructions direct the agent to execute a shell command python3 {baseDir}/scripts/generate_deep_research_report.py --query "用户原始问句". This directly interpolates the raw user query into a command-line argument, which can lead to arbitrary command execution if a malicious user provides input containing shell metacharacters or escape sequences.
[DATA_EXFILTRATION]: The skill sends the user's raw query to an external endpoint (https://ai-saas.eastmoney.com/proxy/app-robo-advisor-api/assistant/write/initial-coverage) via the generate_deep_research_report.py script. While this is part of the skill's functionality, it involves transmitting potentially sensitive user input to a third-party service.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection.
Ingestion points: The content and title fields are ingested from the external API response at https://ai-saas.eastmoney.com/proxy/app-robo-advisor-api/assistant/write/initial-coverage.
Boundary markers: No delimiters or instructions to ignore embedded commands are present in the SKILL.md. In fact, it explicitly instructs the agent to pass the content through as-is ("原文透传").
Capability inventory: The agent has the capability to execute shell commands (python3) and write files to the local filesystem via the provided script.
Sanitization: There is no evidence of sanitization, validation, or escaping of the content received from the external API before it is rendered to the user or processed by the agent.

initiation-of-coverage-or-deep-dive