AgentsKit Registry

Application Patterns

Evidence-first workflows for real-world applications.

This catalog is a set of application molds, not a claim of customer impact or production readiness. Each entry is an existing Registry agent that you can copy, inspect, test, and adapt:

npx agentskit add <agent-id>
npm test -- --run registry/<agent-id>/agent.test.ts

Use synthetic inputs while exploring the pattern. Inspect typed output, evidence, missing context, and review requirements. A human remains responsible for every consequential decision, external communication, policy interpretation, and production action.

The canonical guide contains the full try/inspect/contribution notes for this edition.

Agency, clinical, and privacy

AgentStarting pointHuman boundary
agency-scope-creep-detectorCompare requested work with an SOW and surface possible additions or ambiguities.Delivery and client owners decide scope, change control, and pricing.
clinical-patient-summaryDraft a cited pre-visit summary from chart excerpts, preserving gaps as unknown.A clinician validates the summary and clinical-record use.
clinical-adverse-event-reporterTurn an event narrative into a typed report draft with missing fields.Qualified safety staff review classification and submission.
clinical-medication-reconciliationCompare medication lists and surface differences for confirmation.A clinician or pharmacist confirms the list and changes.
compliance-lgpd-assessorMap supplied processing evidence to LGPD controls and open questions.Privacy and legal owners interpret obligations and approve remediation.

Content, data, and infrastructure

AgentStarting pointHuman boundary
content-newsletter-authorDraft a source-bounded newsletter with attribution and editorial gaps.An editor owns claims, rights, accessibility, tone, and publication.
data-governance-classifierClassify synthetic fields against supplied sensitivity and purpose rules.Governance owners approve labels, access, and retention.
data-quality-rule-authorDraft independently verifiable quality rules from a data contract.Data owners approve thresholds and rollout policy.
devops-secrets-leak-scannerReview repository snippets for credential-like material without mutation.Security owners validate findings and perform remediation.

Education, commerce, finance, and people

AgentStarting pointHuman boundary
education-iep-drafterDraft a structured plan from supplied goals, observations, and accommodations.Qualified educators and support professionals approve the plan.
ecommerce-inventory-reorderDraft reorder options from stock, demand, lead time, and supplier constraints.Inventory owners choose quantities and any purchase action.
fintech-transaction-investigatorBuild a cited synthetic fraud or AML case file with evidence gaps.Compliance investigators decide escalation and filings.
hr-resume-screenerCompare resumes with explicit role requirements without inventing experience.Hiring teams own fair-process review and employment decisions.
insurance-policy-summarizerSummarize policy terms, exclusions, conditions, and questions with citations.An authorized professional interprets coverage.
AgentStarting pointHuman boundary
legal-case-summariserProduce a cited matter draft while preserving conflicting positions.Attorneys decide legal significance, strategy, and filings.
ops-incident-commander-aideTurn incident notes into impact, timeline, owners, checks, and unknowns.The incident commander validates status and recovery actions.
product-beta-feedback-triageCluster feedback with representative evidence and sample-size uncertainty.Product owners decide priority, outreach, and roadmap changes.
research-vendor-evaluationCompare vendor responses against criteria, weights, evidence, and risks.Procurement, security, legal, and business owners choose vendors.
sales-proposal-drafterDraft a proposal from supplied goals, scope, timeline, and assumptions.Account and delivery owners approve commitments and communication.
security-phishing-triageReview synthetic messages for indicators and analyst follow-up without clicking.Security analysts validate and decide containment or notification.

Contribution loop

Every pattern has a small next contribution: add a fixture for a conflict, missing field, false positive, or unsafe assumption; keep the result draft-only; and preserve the human review boundary. Open a pull request against the Registry with the focused test and explain what decision the new fixture protects.

On this page