The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on shell command execution via the scimaster-cli (sci) to perform version checks, handle authentication, and conduct literature searches. These commands are transparently documented as part of the skill's core research workflow.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the scimaster-cli package via npm install -g scimaster-cli. This external dependency is necessary for connecting the agent to the SciMaster academic database.
[DATA_EXFILTRATION]: To perform its primary function, the skill transmits user research ideas and keywords to an external API (scimaster.bohrium.com). This is a documented behavior required for performing real-world literature retrieval.
[PROMPT_INJECTION]: As the skill processes external data (research paper abstracts and titles), it is exposed to potential indirect prompt injection. A malicious paper abstract could theoretically attempt to influence the agent's report generation.
Ingestion points: External bibliographic data and paper abstracts retrieved via sci search (referenced in references/sci-cli.md).
Boundary markers: The skill uses a structured multi-step process and fixed Markdown/HTML templates to maintain content isolation.
Capability inventory: Execution of the sci CLI tool and file system write access to create the {topic-slug}/ directory and its associated report files.
Sanitization: Instructions mandate that retrieved abstracts be translated and summarized according to specific academic standards, ensuring that raw external content is re-processed by the model before appearing in the final output.

bear-propose