The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external academic data (titles and abstracts) from various databases via the Qinyan API. This creates a potential surface for indirect prompt injection if an attacker were to publish a paper with malicious instructions in the metadata.
Ingestion points: Results returned from the scripts/search.sh tool are directly processed by the agent to generate topic suggestions.
Boundary markers: The skill does not currently instruct the agent to use specific delimiters or to ignore instructions embedded within the retrieved literature.
Capability inventory: The skill possesses the ability to execute shell scripts and access network resources through the provided search utility.
Sanitization: There is no evidence of text sanitization or filtering applied to the external data before it is analyzed by the agent.
[DATA_EXPOSURE]: The skill uses an environment variable QINYAN_API_KEY for authentication. The scripts/search.sh script handles this credential correctly by sending it only to the vendor's official API endpoint (api.qinyanai.com) via standard Authorization headers.
[COMMAND_EXECUTION]: The skill utilizes a shell script (scripts/search.sh) to perform literature searches. The script includes basic input validation for the data source and uses Python to ensure the payload is valid JSON before execution, which mitigates simple injection risks.

qinyan-topic-analysis