mittuled annual-comp-review-runner
Agent: Agent Configuration Manager
L2 Agent Configuration Manager (1x) responsible for model selection per agent, compute budget allocation, context window sizing, tool access policies, and API key management.
Department ethos: ideal-agent-operations.md
Skill Description
Reviews whether each agent's model, compute budget, and tool access are appropriately sized through an annual configuration audit.
When to Use
- When the annual configuration review cycle is due.
- When fleet compute costs have drifted significantly from the allocated budget.
- When new model tiers or tools have become available and existing agent configurations may be suboptimal.
Workflow
- Gather Configuration Baselines: Collect each agent's current configuration: model tier, compute allocation, context window size, tool access list, and API key scope. Deliverable: configuration inventory.
- Collect Performance and Cost Data: Pull per-agent performance metrics and cost data from the past year. Include ROI assessments and health monitoring trends. Deliverable: annual performance and cost dataset.
- Assess Right-Sizing: Compare each agent's configuration to its actual usage. Identify over-provisioned agents (paying for capacity not used) and under-provisioned agents (performance degraded by resource constraints). Deliverable: right-sizing assessment per agent.
- Propose Configuration Changes: For each misaligned agent, propose specific changes -- model upgrade/downgrade, compute budget adjustment, tool access addition/removal. Include cost impact estimates. Deliverable: configuration change proposals with cost impact.
- Execute Approved Changes: Implement approved configuration changes. Verify each change does not regress agent performance. Deliverable: updated configurations with post-change validation.
Anti-Patterns
- Review without data: Making configuration decisions based on intuition rather than actual usage and performance data. Why: subjective assessments lead to either wasteful over-provisioning or harmful under-provisioning.
- Annual-only reviews: Treating the annual review as the only time configurations can change. Why: significant events (model releases, cost spikes, new tools) require mid-cycle adjustments.
- Blanket changes: Applying the same configuration adjustment to all agents without per-agent analysis. Why: each agent has different workload patterns and performance requirements; uniform changes inevitably misfit some agents.
Output
On success: Produces an annual configuration review report containing right-sizing assessments, approved configuration changes, and post-change validation results. Delivered to the VP Agent Operations.
On failure: Report which agents could not be reviewed (missing data, unavailable metrics), what partial assessment was completed, and what data gaps need to be closed.
Related Skills
409a-valuation-commissioner-- ROI data feeds into annual right-sizing decisions.compensation-benchmarking-- Model benchmarking informs upgrade/downgrade recommendations.benefits-programme-administrator-- Tool access changes are executed through the benefits programme.