Imported from DiegoNogueiraDev/agf (.agents/skills/kanban-orchestrator/SKILL.md). Install upstream with npx skills add DiegoNogueiraDev/agf --skill kanban-orchestrator. Copyright stays with the author.

kanban-orchestrator

Transform the mcp-graph execution graph into a visual Kanban board with intelligent orchestration. Deterministic-first: all decisions are made via graph traversal, no LLM calls needed.

When to Use

• Visualizing project progress as a Kanban board
• Managing WIP (Work In Progress) limits
• Identifying bottlenecks and blocked tasks
• Getting smart suggestions for next actions
• Tracking flow metrics (throughput, cycle time, lead time)
• During any lifecycle phase (cross-cutting skill)

CLI Usage — zero MCP

View the Kanban Board

agf kanban                           # status columns, WIP, flow metrics
agf kanban --swimlane epic           # group the lanes
agf kanban --swimlane sprint
agf kanban --sprint <id>             # narrow the board to one sprint's cards

Move a Card

A move is a status transition, and status_flow validates it — an illegal move is refused, not logged:

agf kanban validate-move <id> <status>   # dry-run: unresolved deps, WIP overflow (advisory)
agf node status <id> in_progress         # then commit the move
agf done <id>                            # done goes through the DoD gate, never a raw status write

Get Suggestions

agf kanban --suggestions --select data.suggestions   # auto-promote, unblock, WIP/bottleneck, next task

Read the Flow

The board shows state; agf insights explains it:

agf insights wip           # WIP count and alert
agf insights bottlenecks   # where the queue accumulates (TOC: fix the bottleneck first)
agf insights flow          # flow_on vs flow_off verdict
agf insights summary       # cycle time, lead time, throughput

Dashboard

The Kanban tab in the dashboard provides:

• 5 status columns: Backlog, Ready, In Progress, Blocked, Done
• Drag-and-drop cards between columns
• WIP limit indicators (red when exceeded)
• Swimlane grouping by Epic or Sprint
• Suggestions sidebar with auto-apply
• Flow metrics bar (throughput, cycle time, blocked %)

Methodology

WIP Limits (Little's Law)

• Default: In Progress = 3, Ready = 10
• cycle_time = WIP / throughput — lower WIP = faster delivery
• Visual warning when limits are exceeded

Pull System

• Use next tool or Kanban suggestions to pull the next task
• Never push tasks to In Progress without finishing current work

Bottleneck-First (Theory of Constraints)

• If Blocked column is growing, stop adding new work
• Focus on resolving blockers before starting new tasks
• Automated detection when blocked > 30% of total tasks

Suggestion Types

Flow Metrics

• Throughput: Total done tasks
• Avg Cycle Time: Average hours from creation to completion
• Avg Lead Time: Average hours from first status change to done
• Blocked %: Percentage of tasks currently blocked
• WIP Violations: Count of columns exceeding limits

Integration with Other Skills

• graph-implement: Use kanban(action: "board") before start_task to see the big picture
• graph-validate: Check Kanban metrics after validation phase
• graph-review: Export Kanban state for review handoff
• graph-plan: Use swimlane view by sprint during planning

Comparison with Anthropic Agent Teams

Codex Notes

• In Codex Plan Mode, use this skill for planning only and do not mutate files.
• During implementation, follow the project AGENTS.md rules and use apply_patch for manual edits.

Token Economy

Economia de tokens. Os levers compartilhados por todas as skills — --select, agf retrieve-command, agf exec chain, reuso antes de criação — vivem em _shared.md → Token Economy. Fonte única: um parágrafo repetido em trinta arquivos é o trigésimo primeiro que envelhece sozinho.

O board é lido muito mais vezes do que é escrito, e cada leitura completa custa o envelope inteiro. Peça só a coluna que você vai usar.

• agf … --select data.<caminho> — projeta o envelope antes de ele chegar ao contexto. Ler o board inteiro para descobrir uma contagem é o desperdício mais comum desta skill.
• agf retrieve-command "<intenção>" — recupera o comando exato a partir da intenção, em vez de fixar um catálogo de comandos no contexto.
• agf exec chain "cmd1; cmd2" — encadeia num único round-trip quando você precisa de várias leituras; N envelopes viram um.

Reuse antes de criar: um card que já existe não precisa ser redescrito.