EXEMPLARExemplar is part of the NVIDIA Inception program.
NVIDIA InceptionExemplar is part of the Google for Startups program.
Google for StartupsMarshal · Reusable operating methods
Skill Management
Reusable operating methods—triggers, procedures, judgement checks, and quality bars—that turn tacit team know-how into agent-ready skills
Marshal · Skill Management
How Skill Management works
Reusable operating methods—triggers, procedures, judgement checks, and quality bars—that turn tacit team know-how into agent-ready skills
SKILLS
registry · sync · load
Authors push skills into the org registry (sky rails); IDEs and agents pull them at work time (orange rails).
Authors push skills into the org registry (sky rails); IDEs and agents pull them at work time (orange rails).
Why it matters
Built for production agents—not prompt playgrounds
How this harness capability fits the Exemplar platform: one identity, one policy layer, and one audit trail across console, MCP, and SDK.
Why Exemplar
Connected tools are not enough: production agents need repeatable judgement—how to triage, when to escalate, what “done” looks like.
Skills package that judgement so every agent run follows the same operating method, not a one-off mega-prompt.
What Exemplar delivers
Managed skill library with L1/L2/L3 disclosure—agents load metadata first and pull full instructions only when a skill applies.
Skills wired to catalog entities, policies, and tools so procedures stay grounded in your estate.
How teams use it
Author skills for incident triage, rollback readiness, security review, release notes, and governed Day 2 Ops.
Attach skills to agent profiles and orchestration steps; audit which skill version governed each run.
Capability checklist
Developer guide
Official documentation on docs.exemplar.dev for this capability.
Open developer guide (opens in a new tab)Get started
Put Skill Management to work with the rest of Marshal—or talk to us about your agent stack.
From the blog
Related posts on exemplar.dev.
- Skills make judgement reusable
Why AI agents need reusable operating methods, not just connected tools: skill files package triggers, decision checks, examples, and quality bars so teams can run production work consistently.
- Your AI Agent is Burning Money. Here's Why — and the Fix.
Token bloat from mega-prompts on Google ADK agents; how Skills progressive disclosure (L1/L2/L3) cuts cost at scale—and when a plain system prompt still wins.