Irbskill
A managed evaluation and execution harness that versions research skills, data authority, study approvals, tool permissions and human review for bounded clinical-research tasks.
Academic medical centers and research organizations may want reusable agent skills without asking every coordinator to install and govern code independently. Irbskill proposes a managed harness for a small verified skill set, with study-specific data scope, tool restrictions and review. The central source claim is unresolved: the supplied research could not confirm the referenced community corpus of hundreds of medical-research skills and found only a much smaller official repository. Corpus existence, provenance and license are therefore a pre-build kill gate, not a market fact.
A business associate agreement allocates contractual responsibilities; it does not make software compliant or authorize a research use. Institutional review board approval, privacy authorization or waiver, Common Rule applicability, electronic-record controls, protocol scope, data-use agreements and site policy remain external authorities. A skill can contain unsafe code, prompt injection, hidden network access, incorrect scientific logic or actions beyond its declared purpose. Outputs about eligibility, adverse events, study type or cohorts remain candidates for qualified review.
Skill source, license, code version, security review, benchmark result, study protocol, institutional approval, participant authorization, data-use permission, dataset, execution request, policy decision, sandbox run, tool call, output candidate, researcher finding, approved record, destination write, acknowledgment, correction, audit finding and research outcome remain separate. Irbskill should make controlled evaluation repeatable without issuing compliance, scientific-validity or institutional-approval claims.
A research-informatics, clinical-trial operations, institutional-review-board operations or quality leader at an academic medical center or research organization.
Current clinical-agent interest and reported demand for contractual protected-data support create a timely evaluation need.
Reusable skills can reduce repeated implementation, but portability can hide institution-specific approvals, data rights and scientific invalidity.
One cross-reference, two inbound connections and two direct connections provide limited supplied convergence.
The input identifies a specific regulated execution problem, plausible research-operations buyers, available research-data infrastructure and no confirmed managed skill runtime in the supplied search.
The claimed large skill corpus is unverified, the buyer quartet is incomplete, scientific and regulatory semantics are institution-specific and no structural incumbent copying cost is evidenced.
Discussion
No comments yet — be the first to weigh in.