Comparison
Kanban skill vs.
GitHub Issues
Not a replacement β a different tool for a different bottleneck. GitHub Issues is a shared, durable, public system of record. The kanban skill is a private, local, agent-native working surface. Pick by what's actually slowing you down.
Plain Markdown in your repo. The agent's fast local scratch-board.
A database behind an API. The shared, public system of record.
So why not just use GitHub Issues?
You can. Almost everything the kanban skill does, you could do with GitHub Issues plus the gh CLI or a GitHub MCP server. The difference is what it costs to get there.
The same task on GitHub Issues means more noise, more turns, more tokens, higher latency, and heavier prompting to get the agent to reach for it at all. The kanban skill trades GitHub's reach for local speed β and for a solo builder driving an agent, speed is usually the thing in short supply.
Kanban skill vs. GitHub Issues
Fourteen dimensions. A is a clear win; a dash is a deliberate trade-off that just comes down to what you need. The kanban skill takes the speed and locality rows; GitHub Issues takes the scale and collaboration ones.
Plain Markdown in your repo, in git.
GitHub's database, behind an API.
Yes β it's just files on disk.
No β needs network and auth.
Native fs tools: Read, Grep, Glob.
gh CLI or MCP round-trips.
Low β grep returns only the matching lines.
High β JSON payloads and tool schemas.
Local disk, effectively instant.
A network round-trip per call.
One prompt: a skill file and a small script.
Account, auth token, MCP config.
None β the board travels with the repo.
Lives on GitHub.
Minimal by design: priority + effort β all a solo builder needs.
Labels, milestones, assignees, projects β for coordinating a team.
None β id clashes if two people add #1894.
Server-assigned ids, safe for teams.
Pruned to the decisions that steer the next task β why an idea was rejected, what shipped β so the agent proposes forward, never re-doing done or dead work.
Full comment history and edits kept, nothing dropped.
Archive the task once its items are checked off.
Auto-closes issues from linked PRs and CI.
grep β quick on a small board, unwieldy as it grows.
Indexed full-text search and saved filters.
Possible, but only by committing to the Markdown β no lightweight filing.
Anyone can file, comment, and react without a commit.
Every card stays visible in the repo β only the memory hub is pruned to essentials.
Public and linkable β the open-source default.
Where each one wins
Neither is strictly better. The kanban skill optimizes for one agent moving fast; GitHub Issues optimizes for many people staying in sync.
Kanban skill
Token-light and instant
No MCP, no network. The agent greps local Markdown instead of paging a remote API β fewer tokens, lower latency, no auth to refresh mid-task.
Agents actually use it
Agents are reluctant to search GitHub Issues; they reach for filesystem tools by default. A Markdown board meets them where they already are β less prompting, fewer hallucinated task states.
Offline and yours
Plain files in git. Works on a plane, works when GitHub is down. No SaaS dependency, no vendor lock-in β clone the repo and the whole board comes with you.
Memory tuned for proposing
It records the decisions that steer the next task: why an idea was rejected, what got shipped, the gap to the goal. So the agent proposes forward β not re-doing done work or re-floating what you killed.
GitHub Issues
Built for teams
Server-assigned ids, safe concurrent edits, assignees. The kanban skill has no database β two people can both mint #1894 and conflict.
Transparency and reach
Public and linkable, with external contributors filing, commenting, and reacting. The right home when openness matters more than raw speed.
Full context, forever
The kanban skill deliberately compresses β an archived card shrinks to a line. On GitHub every comment, edit, and cross-link stays intact.
Deep integration
Auto-closing from PRs, commit links, project boards, labels, milestones, and a whole ecosystem of third-party tools and indexed search at scale.
Why agents prefer files
The real difference shows up when an agent does the work. Ask the same thing β βfind my high-priority open tasksβ β and the two paths barely rhyme.
It compounds. Every βwhat's next?β, every archive, every board review pays the round-trip tax on GitHub Issues β and models, left to choose, quietly avoid the remote tool and reach for the files instead.
Which should you use?
Reach for the kanban skill when
- You work solo, or with a tight, trusted pair.
- You drive the work through an agent in the terminal.
- You care about moving forward more than a paper trail.
- You want the board in git β offline and portable.
Reach for GitHub Issues when
- You're building in the open and transparency matters.
- Multiple people manipulate the backlog at once.
- You lean on PR/CI links, project boards, and milestones.
- You need outside contributors to file and discuss.
They aren't really competitors. GitHub Issues is the shared system of record; the kanban skill is the agent's fast local scratch-board. If your bottleneck is coordination across people, use GitHub Issues. If it's throughput with an agent, use the kanban skill.
Plenty of solo builders run both β GitHub Issues as the public tracker, the kanban skill as the private surface their agent drives every day.