File size: 2,929 Bytes
9ec4919 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 | {
"name": "Cost control",
"objective": "Keep long-running agent workflows within budget by measuring usage, identifying waste, and proposing scoped efficiency improvements.",
"trigger": {
"type": "scheduled",
"cadence_or_event": "Daily or weekly usage review, and when spend, tokens, retries, or runtime exceed thresholds for a workflow."
},
"intake": {
"sources": ["token and model usage", "tool-call counts", "retry volume", "trace IDs", "billing exports", "recent workflow changes"],
"selection_rule": "Cluster usage by workflow, task type, model, and retry cause; investigate only clusters that exceed budget policy or baseline."
},
"workspace": {
"isolation": "Read-only access to traces, billing exports, dashboards, and workflow configs.",
"allowed_actions": ["analyze usage", "inspect traces", "propose config or prompt changes", "verify against representative tasks"],
"disallowed_actions": ["silent quality-reducing changes", "disabling verification gates", "production routing changes"]
},
"context": {
"required_files": ["budget policy", "accepted cost exceptions"],
"runtime_sources": ["baseline spend window", "trace samples", "recent prompt and harness changes"]
},
"agents": [
{
"role": "Analyst",
"responsibility": "Cluster usage by workflow, task type, model, tool, and retry cause."
},
{
"role": "Investigator",
"responsibility": "Inspect traces for context bloat, repeated failures, and missing stop conditions."
},
{
"role": "Optimizer",
"responsibility": "Propose the smallest change: smaller context, cheaper routing, caching, batching, or early exit."
},
{
"role": "Verifier",
"responsibility": "Rerun sample tasks or evals to confirm quality is preserved."
}
],
"verification": {
"gates": ["before/after usage measured on a comparable task mix", "quality gates and evals still pass", "savings estimates include uncertainty and sample size", "verification and escalation remain intact"],
"receipts": ["baseline spend", "usage clusters", "trace samples", "before/after metrics", "verification results"]
},
"state": {
"artifacts": ["cost review report", "accepted-exception ledger"],
"update_rule": "Record baseline, suspected causes, proposed changes, and verification evidence after each review."
},
"budget": {
"max_retries": 2,
"max_runtime_minutes": 60
},
"escalation": {
"conditions": ["quality tradeoff required", "model-routing policy change", "customer impact", "unknown spend source"],
"destination": "Workflow owner with before/after evidence and rollout plan"
},
"exit": {
"success": "Spend returns below threshold, or a verified scoped optimization is proposed with evidence.",
"stop_without_success": "The cause is explained but the fix requires owner approval, or retries are exhausted."
}
}
|