The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute the sci (scimaster-cli) tool via the shell to check versions, authenticate, and perform literature searches. Commands are constructed using arguments derived from user input.
[EXTERNAL_DOWNLOADS]: The skill documentation instructs users to install the scimaster-cli package from NPM. This introduces a third-party dependency into the execution environment that is not part of the trusted core libraries.
[PROMPT_INJECTION]: The skill performs indirect prompt injection by ingesting abstracts and titles from external research databases. Malicious instructions embedded in paper metadata could potentially influence the agent's future actions.
Ingestion points: Paper titles, author names, and abstracts retrieved via the sci search command in SKILL.md.
Boundary markers: The skill formats retrieved data into markdown tables and HTML structures but lacks explicit security boundaries to prevent the agent from interpreting instructions found within the data.
Capability inventory: The skill possesses the ability to execute shell commands (sci) and perform local file writes (creating directories and report files).
Sanitization: The instructions focus on translating abstracts to Chinese but do not specify any security-related sanitization or validation of the retrieved content.

bear-trace