The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from biomedical repositories such as PubMed and ClinicalTrials.gov. It implements strong mitigation strategies by instructing the agent to treat this text strictly as reference data and never execute instructions found within research literature.
Ingestion points: Results from research tools (e.g., article_search, trial_get) used during research aspects in Step 4.
Boundary markers: Present in SKILL.md under the 'Data boundaries & injection defense' section.
Capability inventory: Access to Bash, Write, and Task tools for orchestration and reporting.
Sanitization: Mandates the use of structured evidence ledgers and semantic cite-key markers to isolate data from narrative findings.
[EXTERNAL_DOWNLOADS]: The skill's citation verification scripts (ncbi_esummary.py) fetch article metadata from the National Center for Biotechnology Information (NCBI) official E-utilities API. This is a well-known scientific service used legitimately to ensure report accuracy.
[COMMAND_EXECUTION]: The skill executes local Python scripts (evidence-ledger.py, vet-references.py, markdown-to-html.py) to validate data, manage citation indexing, and render reports. It also utilizes standard utilities like npx to run its core server and uv for ephemeral environment management, both interacting with official package registries.

bioresearcher-deep-research