onboarding-programme-builder
Agent: Agent Trainer & Skill Optimizer
L2 Agent Trainer and Skill Optimizer (1x) responsible for evaluating existing skills, running evals, optimising prompts, improving agent performance, and training new agent configurations.
Department ethos: ideal-agent-operations.md
Skill Description
Creates the evaluation and warm-up procedure for newly provisioned agents to ensure consistent performance from the first production task.
When to Use
- When a newly provisioned agent needs to be validated before handling production workload.
- When the onboarding process is ad-hoc and needs standardization across agent types.
- When an agent has been reconfigured (new model, updated prompt) and needs re-onboarding to verify performance.
Workflow
- Define Onboarding Criteria: Establish what "production-ready" means for the agent's role -- accuracy minimums, latency ceilings, and skill coverage requirements. Deliverable: onboarding success criteria.
- Design Warm-Up Scenarios: Create a set of representative tasks that cover the agent's full skill range, from routine to edge cases. Include known failure modes from similar agents. Deliverable: warm-up scenario suite.
- Execute Warm-Up: Run the agent through all warm-up scenarios. Score outputs against the success criteria. Identify areas where the agent underperforms. Deliverable: warm-up results with per-scenario scores.
- Tune and Iterate: For underperforming areas, adjust prompts, context, or configuration and re-run the relevant scenarios until criteria are met. Deliverable: tuned agent configuration with passing scores.
- Certify for Production: Issue production readiness certification. Document the agent's tested capabilities and known limitations. Hand off to the Agent Operations Manager. Deliverable: production readiness certificate with capability documentation.
Anti-Patterns
- Skipping warm-up for urgency: Sending agents directly to production because the workload is urgent. Why: untested agents produce unpredictable outputs that create more work through error correction than the warm-up time would have cost.
- Generic onboarding for all roles: Using the same warm-up scenarios regardless of agent role and complexity. Why: a data analysis agent and a customer communication agent have entirely different failure modes; generic scenarios miss role-specific risks.
- Onboarding without documentation: Completing warm-up but not recording the agent's tested capabilities and limitations. Why: undocumented capabilities lead to the agent being assigned tasks outside its validated range.
Output
On success: Produces a production readiness certificate with warm-up results, tuned configuration, and documented capabilities and limitations. Delivered to the Agent Operations Manager.
On failure: Report which onboarding criteria the agent could not meet, what tuning was attempted, and whether the agent needs fundamental reconfiguration or is unsuitable for the role.
Related Skills
first-hire-process-builder-- Deployment procedure that precedes onboarding.manager-development-programme-- Specialized onboarding for orchestration agents.compliance-training-manager-- Compliance validation is a component of onboarding for agents handling sensitive data.