Skip to content
Testing
Skill

/kanban

Use when work has several deliverables, tickets, or delegates: board, WIP 1 writer + 2 review, pull rule, Definition of Done, findings fixed or filed.

BOOST
From plugin
shaft-engine
41720 skills
Install
$ npx -y skills add ShaftHQ/SHAFT_ENGINE --skill kanban --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/kanban

Context preview

The summary Claude sees to decide when to auto-load this skill.

Use when work has several deliverables, tickets, or delegates: board, WIP 1 writer + 2 review, pull rule, Definition of Done, findings fixed or filed.

SKILL.md

kanban.SKILL.md
name: kanban
description: >-
  Use when work has several deliverables, tickets, or delegates: board, WIP 1
  writer + 2 review, pull rule, Definition of Done, findings fixed or filed.
license: MIT

Kanban task flow

Use when work has more than one deliverable, more than one ticket, or any delegate. Flow beats busyness: finish before starting.

Board

Keep one board per task as JSON (models corrupt JSON less than prose) in the session state directory, never in tracked files:

{"wip": {"doing": 1, "review": 2},
 "cards": [{"id": "T1", "title": "...", "column": "ready",
            "acceptance": ["..."], "proof": "command that proves it",
            "evidence": null, "pr": null}]}

Columns: `backlog` -> `ready` -> `doing` -> `review` -> `done`.

  • `ready` requires acceptance criteria and a proof command. No criteria, no pull.
  • `doing` WIP **1 writer** per repository: one card being edited at a time.
  • `review` WIP **2**: at two cards waiting for review or CI, pull nothing new.
  • `done` only when the Definition of Done holds.

Pull rule

Every session, turn, and wake starts by draining right to left: open PRs and `review` (red or `ACTION_REQUIRED` checks, behind base, conflicts, review comments, auto-merge off), then `doing`, then shipped bugs or red default branch. Pull a new `ready` card only when nothing downstream is actionable, and only while `review` has capacity; waiting on CI counts as idle. Never leave a started card to start another. Never push work forward. Group related cards into the fewest safe pull requests. Avoid PRs that depend on other open PRs; land them in order and merge base once before the final push.

Definition of Done

A card is done when all hold:

1. Every acceptance criterion is met and its proof command output is attached. 2. Tests for new behavior were red first, then green. 3. One fresh-context review is clean; a second round only for blocker findings. 4. Required checks are green on the exact head after one blocking wait per push (`gh pr checks --watch --fail-fast` or `gh run watch`); retry once, then stop and report. Never re-verify what already passed. 5. Every finding raised while working is closed (see below), including static-analysis and code-quality review threads (fix or reply, then resolve) before the PR counts as ready. 6. Merged or handed off as instructed; worktree cleaned. 7. After any CE harness change merges, every agent and delegate reloads the latest harness from main before continuing. After each completed delivery run `python3 .chaos-engine/tool.py maintain` (fast-forward with local edits kept, reinstall, doctor, stale-store refresh, reload). 8. Ship to-deliver items in the fewest pull requests; group related issues.

Check items 2 and 4 plus labels and links with `python3 .chaos-engine/tool.py dod <PR>` ([`dod.py`](../../dod.py)) (table; exit 1 names each gap).

Findings policy

Every finding ends as exactly one of:

  • **fixed** in the current delivery, or
  • **filed** as an issue, allowed only when it is both outside the task's scope

and not critical, blocker, or high severity.

Nothing stays only in chat, a report, or a local file. The final report lists each finding as fixed (with PR) or filed (with issue).

Waste checklist

Before each pull, drop anything that is partially done work, extra process, extra features, task switching, waiting, handoffs without context, or defects passed downstream (Poppendieck's seven wastes).

Status

Every status report is a Markdown table, one row per card (`ID | Issue(s) | PR | Status | Proof or blocker`), then Risks. Never prose or bullet lists. No separate status rituals. Report only rows that changed since the last report; say nothing when nothing changed.

Read more
Ships withshaft-engine

One Java engine for web, mobile, API, CLI, database, and desktop. Drivers, waits, and evidence are handled. Every run ends in an Allure report. &nbsp; Upgrade to SHAFT_ENGINE · ▶ Watch the 26-second clip with sound and captions · 100-second overview

Get the whole plugin
Stats
416
Stars
155
Forks
Active
Maintenance
Java
Language
MIT
License
2h ago
Last commit
8y ago
Created
6h ago
Added

Repo: ShaftHQ/SHAFT_ENGINE

Other skills on shaft-engine.

ci-cd
Skill

ci-cd

Use when PR merge waits are long or CI/CD is being designed, audited, or sped up: targets, PR/main/nightly tiers, opt-in labels,…