skills-from-claude-blog

Choose the simplest loop primitive (turn-based, /goal, /loop, /schedule) and make success criteria explicit and verifiable.

Verify any UI change end-to-end before declaring it done.

Plan and roll out Claude Code via a self-hosted apps gateway, using centralized SSO login, managed settings enforcement, and per-user usage attribution when running on Amazon Bedrock or Google Cloud.

Decide whether to run Claude via Microsoft Foundry hosted on Azure or hosted on Anthropic, based on Azure-native requirements vs. needing full API feature/model availability.

A practical playbook for designing and operating effective human–agent teams: working in public, defining roles and tools, setting a north star, and building trust through verification and gradual autonomy.

Configure and roll out workspace-owned “agent identities” for team-wide AI agents with least privilege, strong boundaries, and reliable auditing.

Roll out Claude Desktop across an organization with cloud-hosted inference and enterprise controls, based on a provider-backed deployment (AWS, Google Cloud, or Microsoft Foundry).

Centralize authorization for MCP connectors via an identity provider so admins provision once and users inherit access via IdP groups and roles.

Request and iteratively publish shareable Claude Code artifacts (live, interactive pages) built from full session context.

Decide where to place Claude Code instructions (CLAUDE.md, rules, skills, subagents, hooks, output styles, or appended system prompt) based on load timing, compaction behavior, context cost, and authority.

Plan and review a Workload Identity Federation (WIF) setup for Claude Platform workloads, using service accounts, federation rules, and least-privilege scopes.

A repeatable workflow to import design systems into Claude Design, keep output on-brand, and hand off designs to Claude Code without restarting from screenshots.

Patterns observed in Build Day projects for planning, verification with independent sub-agents, and cost/latency optimization under tight time constraints.

Reusable planning and iteration patterns drawn from multiple winning projects in Claude’s “Built with Opus 4.7” hackathon.

Use Claude with Microsoft 365 context (SharePoint/OneDrive, Outlook, Teams) by framing questions for cross-tool enterprise search and producing source-backed reports.

Decide when to use Skills vs prompts, Projects, MCP, and subagents, and combine them into a coherent workflow.

Produce structured first-pass finance deliverables and iterate with human validation, keeping work close to spreadsheets, decks, and source files.

Dispatch a multi-pass, high-signal pull request review and deliver a consolidated overview plus actionable inline findings.

Workflow patterns for using Claude Artifacts (GA) to iteratively build and share code, documents, and visualizations across Free/Pro/Team and mobile.

Guidance for using Claude models with a generally available 1M-token context window (Opus 4.6 and Sonnet 4.6), including practical considerations for very large inputs.

A checklist and guidance for taking a tool-using Claude agent from prototype to production by separating the harness (agent runner) from the execution sandbox, with emphasis on sessions, credentials, and observability.

Use Claude Managed Agents scheduled deployments and vault environment variables to run routine work on a cron schedule and securely authenticate CLI tools without exposing secrets to the model.

Use Apple Foundation Models for fast on-device tasks, then hand off to Claude for multi-step reasoning, code generation, web search, or code execution—while keeping a single, continuous user experience.

Monitor, debug, and improve a published Claude connector using directory observability signals like adoption, error rates, and latency.

Use Claude Cowork for multi-step knowledge work across local files and connected apps, aiming for concrete deliverables with traceable citations.

Patterns for building Claude-powered GTM workflows (email drafting, pre-call briefs, end-of-day recaps) that pull fresh context from source systems and can be packaged for team adoption.

Best practices, checklists, and reusable prompts for delegating multi-step knowledge work in Claude Cowork to produce deliverables.

Use this skill when the user asks for business analytics that requires querying a data warehouse (SQL and/or governed semantic layer), and the work must be reliable under entity ambiguity, freshness/staleness risk, and retrieval gaps.

Use this skill when you are writing, reviewing, or refactoring Claude Code skills and want proven structure, content patterns (Gotchas, references, scripts, templates), and distribution guidance.

Patterns and reusable prompt templates for creating “dynamic workflows” in Claude Code to orchestrate multi-agent harnesses for complex, high-value tasks.

Practices for adapting engineering planning, context sharing, code review, and team operating norms once agentic coding becomes the default.

Decide when to use Claude Code dynamic workflows and structure requests so Claude can coordinate parallel subagents, resume long runs, and verify work before presenting results.

Add a structured planning-orchestration layer before code generation to make outcomes, assumptions, edge cases, and validation criteria explicit, producing a reviewable PRD-style plan.

Use this skill to decide whether to run Claude Managed Agents with self-hosted sandboxes and/or MCP tunnels, and to collect the minimum deployment details and security considerations for each option.

Keep finance narratives (board decks, monthly reviews) coherent as numbers refresh by using Claude as an integrity layer for reconciliation, consistency checks, and first-pass variance commentary.

Plan Claude governance using the Claude Compliance API and the 28 launch integrations Anthropic announced on May 21, 2026. Use when an IT or security team is wiring Claude Enterprise and the Claude Platform into existing DLP, SASE, data-security, SIEM/SecOps, identity, eDiscovery, AI-SPM, or AI observability/telemetry stacks — or when a security/compliance platform vendor is deciding whether to apply to the integration network.

Evaluate and shortlist the Claude Opus-powered defensive cybersecurity partner offerings that Anthropic publicized on May 21, 2026. Use when a security buyer or CISO is choosing among the three families of offerings — continuous offensive testing at production scale, closing the find-to-fix gap, and getting AI into production with governance — and needs a structured comparison against the partners Anthropic explicitly named (Wiz, Palo Alto Networks/Unit 42, CrowdStrike, Accenture, TrendAI, Deloitte, PwC), plus an awareness flag for BCG, Infosys, and SentinelOne.

Turn recurring sales leadership routines (meeting prep, weekly forecast rollups, and large-scale account scoring) into scheduled, repeatable workflows that assemble data from multiple systems and produce standardized outputs with human approval.

Use HTML artifacts (instead of Markdown) in Claude Code to produce richer, more readable, and more shareable outputs for planning, code review/understanding, prototyping, and reporting.

Keep Claude Managed Agents tool execution and connectivity inside your security perimeter using self-hosted sandboxes and MCP tunnels.