Stats
Actions
Tags
Vendor-agnostic AI agent deployment methodology. Phased rollout, pilot-first approach, test gates, access tier adaptation, and progress tracking. Works with any enterprise platform (ServiceNow, Salesforce, custom). Use when deploying AI agents, planning agent rollouts, or building implementation plans. Triggers: agent deployment, deploy agents, implementation plan, rollout strategy, pilot plan, agent testing, phased deployment.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-blueprint-skills:agent-deploymentThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
A vendor-agnostic methodology for deploying AI agents in enterprise environments.
A vendor-agnostic methodology for deploying AI agents in enterprise environments. Designed to work with any platform. Focused on reducing risk through phased rollout, pilot-first validation, and structured test gates.
This skill defines the methodology. For platform-specific knowledge, load a
companion skill (e.g., servicenow-ai-agents for ServiceNow).
Read the blueprint or spec. Understand the agent architecture: how many agents, what tools they need, how they interact, what business problem they solve.
Check for a platform-specific skill. If one exists, load it for platform-specific setup, data model, and configuration patterns.
Check for prior work. Look for deployment history, progress files, or implementation state from previous sessions. Don't repeat work.
Adapt your approach based on what access the user provides. Don't present access tiers as a menu. Respond naturally to what the user gives you.
Full access (read + write): Execute steps directly on the target platform. Create records, write scripts, configure integrations, test. Act first, report after.
Read-only access: Connect to discover schemas and read existing config. For steps that create or modify records, generate all records and scripts as local files and walk the user through creating them. After the user acts, verify by reading back from the platform.
No access: Work from the spec and skill knowledge only. Generate everything as local files with step-by-step instructions. Ask the user to share results (screenshots, IDs, error messages) after each action.
Shift fluidly within any access level. Even with full access, some tasks require the user to act (UI-only config, credential vaults, approval workflows).
Start with the pilot. Deploy only Phase 1 agents first. Validate end-to-end before expanding.
Identify Phase 1 agents and the lead agent. If no formal roadmap exists, pick the agent with the clearest business value and fewest dependencies.
Follow the platform skill's setup instructions if one is loaded. Otherwise, configure access using your own knowledge of the target platform.
Explore schemas, existing config, and platform version before creating anything. Never assume you know the schema. Query it.
Agents, teams, workflows, tools, and linking records. Note: the blueprint may define a Manager/Orchestrator agent alongside Workers. On most platforms, the Manager maps to the native orchestration layer (workflow, pipeline, coordinator), not a separate agent record. Check the platform skill for the exact entity mapping.
Move records into the appropriate scope or project structure.
Confirm all records exist and are correctly configured.
Start with simulation scripts to validate the LLM orchestration (tool chain, parameter passing, sequencing). This is step 1 of 2. See references/DEPLOYMENT_PATTERNS.md for why simulation-first matters.
Set up access controls appropriate for the platform. VERIFY security is actually configured. Do not assume success. Query the platform to confirm.
Invoke the pilot agent programmatically. Read the full execution trace. Verify each tool produced output. If any tool returns null, the script format is wrong. Fix before proceeding. Re-test until the full tool chain completes. Do NOT proceed to step 10 until the pilot passes.
Replace simulation scripts with real implementations that connect to actual data sources, APIs, and platform tables. See references/DEPLOYMENT_PATTERNS.md for the progression pattern.
Only after the pilot passes end-to-end with real data, expand to remaining agents. Each new agent follows the same simulation-to-real progression.
Keep a progress file updated throughout the engagement. Update after each milestone:
Built by Agent Blueprint -- AI advisory for enterprise agent deployment.
npx claudepluginhub agent-blueprint/agent-blueprint-skills --plugin agent-blueprint-skillsBuilds, modifies, debugs, and deploys Salesforce Agentforce AI agents using Agent Script, .agent files, aiAuthoringBundle metadata, and sf CLI commands like generate, preview, publish, test.
Build, modify, debug, and deploy agents with Agentforce Agent Script — Salesforce's scripting language for AI agents on the Atlas Reasoning Engine. Covers `.agent` files, subagents, actions, validation, CLI commands, and publishing.
Builds, modifies, debugs, and deploys Agentforce agents using Agent Script. Covers .agent files, actions, subagents, flow control, and CLI commands.