Add rate limiting to API gateway
Implement token bucket rate limiting middleware for the public REST API.
AI Agents
Autonomous Coding Loop Board
Give your coding agent a self-managed plan-code-evaluate-fix loop
Autonomous coding agents need more than a to-do list — they need a structured loop that mirrors real software development: plan the approach, write the code, evaluate the result, and fix what breaks. This Kanban Template gives your agent six columns (Backlog, Planning, Coding, Evaluating, Fix Needed, Done) mapped directly to MCP tool calls like move_task and update_task, so the agent can drive its own workflow while you retain full visibility on the board. It works on the free plan out of the box and scales with your team as you add more automated or human-reviewed tasks.
kanboard.io/templates/autonomous-coding-loop
Template preview
Autonomous Coding Loop Board
6 columns
Backlog
3Holds coding tasks the agent has not yet started, queued by priority.
Fix flaky integration test suite
Investigate and stabilize the CI integration tests that fail intermittently.
Refactor auth token refresh logic
Simplify the token refresh flow to remove duplicate network calls.
Planning
2Where the agent drafts an implementation approach before touching code.
Plan schema migration for orders table
Outline migration steps and rollback plan for adding a status column to orders.
Design retry strategy for webhook delivery
Draft exponential backoff and dead-letter queue approach for failed webhooks.
Coding
3Active implementation work the agent is currently writing and committing.
Implement rate limiting middleware
Write and unit test the token bucket middleware per the approved plan.
Build webhook retry queue
Code the dead-letter queue and backoff logic for webhook delivery.
Apply orders table migration
Write and run the migration script with rollback support.
Evaluating
2Tasks under automated test, lint, or review checks after implementation.
Run test suite on rate limiter branch
Execute unit and load tests against the new middleware and record results.
Review webhook retry PR diff
Static-analyze the diff for edge cases in backoff timing and queue overflow.
Fix Needed
2Tasks that failed evaluation and require the agent to patch and resubmit.
Fix failing rate limiter load test
Load test showed request drops above 500rps; adjust bucket refill rate.
Resolve migration rollback error
Rollback script throws foreign key violation; add cascading delete handling.
Done
2Completed tasks that passed evaluation and are merged or deployed.
Merge auth token refresh refactor
Refactored token refresh code passed all checks and was merged to main.
Deploy webhook retry queue
Retry queue passed evaluation and was deployed to production.
How to run this board
Step 1
Agent pulls the highest-priority task from Backlog and moves it into Planning, drafting an implementation approach in the task description.
Step 2
Agent moves the task to Coding and implements the change, committing work incrementally via update_task notes.
Step 3
Once implementation is ready, the agent moves the task to Evaluating and triggers tests, lint, or review checks.
Step 4
If checks fail, the agent moves the task to Fix Needed, patches the issue, and sends it back to Evaluating; if checks pass, it moves to Done.
Step 5
Agent repeats the loop by pulling the next Backlog task, keeping the board as the single source of truth for autonomous progress.
AI agent usage
Run it with AI agents
This playbook lets a coding agent manage its own autonomous work loop directly on the Kanboard board via MCP tools, moving tasks through Backlog, Planning, Coding, Evaluating, and Fix Needed as it plans features, writes code, runs evaluations, and repairs failures without human intervention for routine steps.
# Autonomous Coding Loop Board
Columns: Backlog, Planning, Coding, Evaluating, Fix Needed, Done.
Agent rules:
1. Pull the top task from Backlog and move it to Planning using move_task before writing any code.
2. While in Planning, write an implementation plan in the task description via update_task, then move the task to Coding.
3. In Coding, implement the change, commit, and move the task to Evaluating when work is ready for tests/review.
4. In Evaluating, run tests/lint via CI. If checks fail, move the task to Fix Needed with failure details in the comment field; if checks pass, move to Done.
5. In Fix Needed, resolve the issue, then move the task back to Evaluating for re-check. Never mark Done without a passing Evaluating pass.
Use get_task before update_task to confirm current state and avoid overwriting concurrent edits.Pull the next task from Backlog, draft a plan, and move it into Planning.
Take the task in Fix Needed, describe the failing test, fix the bug, and move it back to Evaluating.
List every task currently sitting in Coding and summarize their status.
Learn more about the board-centric workflow in Kanban for AI Agents or open the MCP guide.
Frequently asked questions
How many tasks can I track on the free plan?
The free plan supports 1 project with up to 25 tasks total, which comfortably covers this six-column autonomous coding loop template with its 18 sample tasks.
Can multiple agents work the same board at once?
Yes, any number of agents or humans can call the MCP tools (list_tasks, get_task, update_task, move_task) against the same project, since Kanboard's board state is shared and updated in real time.
Does the agent need to use the macOS app to update tasks?
No, agents typically interact through the MCP server or REST API for automated updates, while the macOS app or web kanban board is available for human oversight and manual edits.
What happens if a task fails evaluation repeatedly?
The task simply cycles between Evaluating and Fix Needed via move_task calls until it passes; there's no limit on retries within your plan's task cap.
Do I need the pro plan to run this template?
No, the free plan (1 project, 25 tasks) is sufficient for this template's 6 columns and 18 sample tasks; upgrade to pro ($12/mo or $96/yr) only if you need more projects or a higher task count.
Can I customize the column names for my own coding workflow?
Yes, you can rename columns as needed, but be sure to update your AGENTS.md snippet so the agent's move_task calls reference the correct column names.
Related templates
Ready to create this board?
Sign in and use the template to create a project with columns and sample tasks.