Cost Optimize

Analyze recent token usage across agents and models, identify waste, and recommend specific optimizations with estimated dollar savings.

When to use

When costs are higher than expected or you want to proactively reduce spending. Analyzes model selection efficiency, cache utilization, agent redundancy, and prompt efficiency.

Steps

1. Load usage data -- call mcp__claude-flow__memory_search on the cost-tracking namespace (last 7 days). The memory_* tools route by namespace; use them — not agentdb_hierarchical-* (which routes by tier).

2. Analyze model fit -- for each agent, assess whether the model tier matches task complexity:

   • Agents doing simple tasks (formatting, linting) on Sonnet/Opus → suggest Haiku or Agent Booster
   • Agents doing complex tasks (architecture, security) on Haiku → flag quality risk

3. Check cache rates -- compute cache hit rate per agent; if below 60%, recommend enabling or improving prompt caching (90% cost reduction on cache reads)

4. Detect redundancy -- look for multiple agents performing overlapping tasks, or agents being spawned for work that could be batched

5. Estimate savings -- for each recommendation, calculate: current cost, projected cost after optimization, dollar savings, percentage reduction

6. Search prior optimization patterns -- call mcp__claude-flow__agentdb_pattern-search (ReasoningBank-routed; don't pass a namespace argument — pattern-* tools ignore it).

7. Store the optimization pattern -- two paths:

   • Pattern store (typed, recommended): mcp__claude-flow__agentdb_pattern-store with type: 'cost-optimization'. Don't pass a namespace arg — ReasoningBank routes it; on bridge unavailability the fallback writes to the reserved pattern namespace with controller: 'memory-store-fallback' (see ruflo-agentdb ADR-0001).
   • Plain store (namespace-routable): mcp__claude-flow__memory_store --namespace cost-patterns — this DOES respect the cost-patterns namespace because memory_* is namespace-routed.

8. Close the routing feedback loop — auto-emit hooks_model-outcome -- for each downgrade recommendation, format the outcome-emit command as part of the recommendation table so it can be run directly:

   # success path (downgrade worked)
   node plugins/ruflo-cost-tracker/scripts/outcome.mjs "<task-description>" <model> success
   
   # escalated path (had to upgrade after downgrade attempt)
   node plugins/ruflo-cost-tracker/scripts/outcome.mjs "<task-description>" <model> escalated
   

   The script wraps npx @claude-flow/cli hooks model-outcome -t ... -m ... -o ... with explicit-argv spawnSync so quoting is safe. Without this signal the router does not learn from cost-tracker's recommendations and the booster bypass rate (see cost-booster-route skill) does not improve over time. This is the typed equivalent of the legacy routing-outcomes namespace (see ruflo-intelligence ADR-0001 §"Neutral").

9. Report -- display: ranked recommendations with savings estimate, total potential savings, implementation priority (quick wins first), and any model-outcome events emitted in step 8

CLI alternative

npx @claude-flow/cli@latest memory search --query "cost optimization strategies" --namespace cost-patterns
npx @claude-flow/cli@latest memory store --key "opt-2026-05-04" --value '{...}' --namespace cost-patterns