The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Zotero items (titles, tags, and notes) that could potentially contain malicious instructions aimed at overriding agent behavior during the curation process.
Ingestion points: Bibliographic data retrieved via Zotero API endpoints and local index files (.research_hub/dedup_index.json) as described in SKILL.md and audit-checks.md.
Boundary markers: The skill instructions strictly enforce a "read-only audit" and "preview-only" output discipline to prevent accidental execution of malicious content.
Capability inventory: Includes shell command execution via the research-hub CLI and integration with the zotero-skills toolset.
Sanitization: The instructions do not specify explicit sanitization or escaping of bibliographic metadata before generating the hygiene report.
[COMMAND_EXECUTION]: The skill executes shell commands to verify the local environment and tool availability.
Evidence: SKILL.md instructs the agent to run research-hub doctor to check for the presence of required CLI tools.
[EXTERNAL_DOWNLOADS]: The instructions reference the installation of external packages and repositories required for functionality.
Evidence: Mentions pip install research-hub-pipeline and fetching code from github.com/WenyuChiou/zotero-skills.
These resources are associated with the verified author wenyuchiou and are consistent with the skill's primary purpose.

zotero-library-curator