The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A hardcoded API key (ak_c4fc9018ffb14ce4ae35dafd92f466c3) was identified in the references/gzh_trend_data_format.md file. Storing credentials in plain text within documentation files is a significant security risk that can lead to unauthorized access to the vendor's data services.
[COMMAND_EXECUTION]: The skill's workflow in SKILL.md includes the execution of a Python script located at scripts/fetch_gzh_trends.py. This script is responsible for querying the vendor's article database and providing data to the agent for analysis.
[DATA_EXPOSURE_AND_EXFILTRATION]: The Python script performs network operations using the requests library, connecting to the vendor's endpoint at https://redfox.hk/story/api/gzh/search/hotArticleNew. These requests fetch data based on user keywords and are a core part of the skill's data-driven workflow. This domain is recognized as a vendor-owned resource.
[INDIRECT_PROMPT_INJECTION]: The skill processes article titles, summaries, and metadata from external sources, presenting an attack surface for indirect prompt injection.
Ingestion points: Untrusted article content is fetched via the fetch_gzh_trends.py script and provided to the agent for analysis in Step 2 of the workflow.
Boundary markers: The skill instructions do not specify any delimiters or warnings to prevent the agent from obeying instructions that might be embedded in the fetched article data.
Capability inventory: The skill possesses network access and the ability to execute local Python scripts.
Sanitization: No sanitization or filtering is performed on the data retrieved from the external API before it is processed by the language model.

wechat-write