nntrivi2001
57dc91585d
Internal SmartGift build of a Claude Code monitoring dashboard. Lanes: a durable unit of parallel agent work, one per working directory, tracked across session restarts. Managed lanes are git worktrees the dashboard provisions and can reset or remove behind a three-check destroy guard and a counted preflight; adopted lanes are directories you already own and are never destroyable. Pipelines: a lane moves through pipeline stages. A stage the agent declares with evidence renders green; a stage inferred from the tool-event stream renders dashed amber and never counts as done. Detection is forward-only within a 30-minute window, and never writes the declared stage. Workspace: one page at /run with a lane grid, the selected lane's pipeline, and a full Claude console behind a disclosure.
3.9 KiB
3.9 KiB
description
| description |
|---|
| Analyze workflow patterns using the Agent Monitor's workflow intelligence API — orchestration DAGs, tool flow transitions, subagent effectiveness, model delegation patterns, error propagation by depth, concurrency lanes, compaction impact, and agent co-occurrence. Produces prioritized optimization recommendations with quantified impact. |
Workflow Optimizer
Analyze Claude Code workflows using the Agent Monitor's workflow intelligence engine.
Input
The user provides: $ARGUMENTS
Options: "analyze", a session ID for single-session analysis, or a focus: "tools", "subagents", "cost", "errors".
Data Sources
| Endpoint | Returns |
|---|---|
GET /api/sessions?limit=100 |
Session list with metadata |
GET /api/workflows/{sessionId} |
11 workflow datasets (see below) |
GET /api/analytics |
Tool usage top 20, event types, agent types |
GET /api/pricing |
Model pricing rules for cost comparison |
Workflow Intelligence API (GET /api/workflows/{sessionId})
Returns these 11 datasets per session:
| Dataset | Content |
|---|---|
stats |
Aggregate session stats: tool count, agent depth, event count |
orchestration |
DAG: agent nodes with parent/child edges, depths, types |
toolFlow |
Transition matrix: tool A → tool B with counts (common sequences) |
effectiveness |
Subagent success: per-type completion rates, avg duration, task success |
patterns |
Recurring sequences: detected workflow patterns with frequency |
modelDelegation |
Model choices: which models are delegated which tasks |
errorPropagation |
Error flow by depth: where in the agent tree errors originate and propagate |
concurrency |
Concurrency lanes: overlapping agent execution timelines |
complexity |
Complexity score: numerical score based on depth, breadth, tool diversity |
compaction |
Compaction impact: token savings, frequency, context health |
cooccurrence |
Agent pairs: which agents frequently run together |
Optimization Analyses
1. Tool Flow Optimization
From toolFlow transition data:
- Identify the most common tool sequences (e.g., Read → Edit → Bash)
- Find redundant transitions (same tool called repeatedly = retries)
- Detect anti-patterns: high-frequency failure loops
- Recommend tool chain shortcuts
2. Subagent Strategy
From effectiveness + orchestration:
- Which subagent types (task, explore, code-review) have highest completion rates
- Average duration per subagent type — are subagents taking too long?
- Underutilized types: tasks that could benefit from delegation
- Over-spawning: too many subagents for simple tasks
3. Model Delegation Analysis
From modelDelegation:
- Which models handle which task types
- Cost-per-task comparison across models
- Opportunities to delegate simple tasks to cheaper models (Haiku/Sonnet instead of Opus)
- Calculate estimated savings from model rebalancing
4. Error Prevention
From errorPropagation:
- Where errors originate (agent depth level)
- How errors cascade to parent agents
- Error types (APIError, tool failure) by frequency
- Defensive strategies: which patterns lead to fewer errors
5. Concurrency Optimization
From concurrency:
- Which agents run in parallel vs sequential
- Bottlenecks: sequential agents that could be parallelized
- Resource contention: overlapping heavy tasks
6. Context Health
From compaction:
- How often compaction occurs per session
- Token recovery from compaction baselines
- Sessions that hit context limits — suggest breaking into smaller tasks
Output
Prioritized recommendations table:
| # | Recommendation | Source Data | Impact | Effort | Est. Savings |
|---|
Top 5 recommendations with detailed explanation, supporting data from the workflow API, and implementation steps.