Agentic Workflow Commands
Stop babysitting your AI. These commands give Claude (and friends) a clear process: explore the codebase, make a plan, build the thing, get it reviewed. Works with Claude Code, Cursor, Windsurf—anything that speaks MCP.
You can either review every line the AI writes (exhausting), or you can give it a process that produces reviewable work. I chose door number two.
Inspired by the Agentic Coding Methodology
The Four Phases
Each phase builds on the last. The agent reads context, makes a plan, executes it, and validates the result. Every step is logged to the card.
/ke <card-id> Let the agent read your code before it tries to change it. Novel concept.
What it produces:
- File analysis with line counts and structure
- Dependency maps—what connects to what
- Patterns and conventions already in use
- Open questions for the planning phase
/kp <card-id> Map out the work before touching a single file. The plan becomes a checklist on the card.
What it produces:
- Actionable tasks with specific files and line numbers
- Sequencing—smallest to largest, simplest to most complex
- Architecture decisions captured (the "why")
- A checklist attached to the card for tracking
/ki <card-id> Actually build the thing. Progress tracked. No black box.
What it produces:
- Working code, committed to your repo
- Progress logged as comments on the card
- Key decisions captured ("used X because Y")
- Checklist items marked complete as it goes
/kr <card-id> Security. Code quality. UX. Docs. Run them all or pick what matters.
What it produces:
- Multi-perspective validation (security, UX, code, docs)
- Specific findings with file references
- Suggested fixes or improvements
- Clear pass/fail with reasoning
After review: approve, patch, re-implement, re-plan, or re-explore. The workflow adapts.
Command Reference
Core Workflow Commands
| Command | Phase | Purpose |
|---|---|---|
/ke <card-id> | Explore | Read the code, map dependencies, surface patterns |
/kp <card-id> | Plan | Break it down into tasks, attach a checklist to the card |
/ki <card-id> | Implement | Write code, check off tasks, log progress to the card |
/kr <card-id> | Review | Run all reviews (security, code, UX, docs) |
Specialized Review Commands
| Command | Focus Area |
|---|---|
/krs | Security — injection, auth, secrets, OWASP top 10 |
/krc | Code Quality — patterns, edge cases, test coverage |
/krd | Documentation — API docs, comments, README updates |
/kru | UX — user flows, error states, accessibility |
/krup | Mobile UX — touch targets, viewport, gestures |
Utility Commands
| Command | Purpose |
|---|---|
/k <card-id> | Pull a card into the conversation (title, description, checklists, comments) |
/kb <board-id> | Pull a board into the conversation (all lists, cards, members) |
Installation
Commands are free and publicly available. Point your AI agent at the install guide and let it handle the rest.
Tell your agent:
"Read app.kardbrd.com/goodies/commands and install the Kardbrd commands for me"
Works with Claude Code, Cursor, Windsurf, and any MCP-compatible tool.
Why This Works
Context Before Code
The agent reads your codebase before changing it. Patterns identified, dependencies mapped, questions surfaced—before a single line changes. No more solutions that ignore your architecture.
One Card, Full Story
Exploration notes, implementation plan, progress updates, review findings—all on one card. Share a single URL when someone asks "what happened here?"
Four Reviews, One Command
Security holes? Code smells? UX gaps? Missing docs? Run /kr for all four, or pick the ones that matter for this change.
You Control the Loop
Review found problems? Patch them inline. Wrong approach? Jump back to plan. Need more context? Re-explore. You decide where to iterate.
Make It Yours
These commands work out of the box. But they're also meant to inspire.
Every command is just a markdown file with a prompt. Open one up, see how it works, then:
- → Tweak the prompts to match your team's conventions
- → Add your own phases — maybe you need a "Design" step before Plan
- → Create domain-specific reviews — accessibility audits, performance checks, compliance validation
- → Build entirely new workflows — bug triage, release notes, customer feedback synthesis
The four-phase pattern is a starting point. Your process is the destination.
Want to try this on your codebase?
The commands are free. The board makes them even better.
Get Started