Ready, Spec, Ship ยท built with Kiro

Health, where you already code.

DevWell is a Kiro plugin that lives inside your coding session, not a separate terminal. It tracks hydration, movement, nutrition and mindfulness, scores burnout risk from real signals and adapts to how you respond. Fully local, no telemetry. Also runs in Claude Code and Codex.

0
MCP tools
0
Health domains
0
Tests passing
0
Local, no cloud
devwell status
$ devwell water 2
๐Ÿ’ง Water Logged: 2 glasses
[โ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘] 25%
๐Ÿ“Š Today: 2/8 glasses
๐Ÿ’ก 6 more glasses to reach your goal.

$ devwell status
๐Ÿƒ Movement   sitting 42 min  โ†’ time to move
๐Ÿง˜ Mindfulness   stress: low
honest by design
# analytics with no history yet
> analytics_dashboard
๐Ÿ”ฌ Correlations:
   Not enough data yet. Keep logging to unlock them.
# it never fakes a number
The problem

Long sessions quietly cost you.

Developers sit for hours and skip water, movement, meals and mental breaks. Focus and code quality slip. The toll compounds over years. Standalone wellness apps live outside the editor, so they get ignored.

How it works

It rides the AI you already use.

DevWell is an MCP server, not a background daemon. It evaluates your health context on each AI interaction and through Kiro hooks, then lets the assistant coach you in the moment. Honest about what it is.

1. Add it to Kiro

DevWell's .kiro config loads it as an MCP server with proactive agent hooks, so it is a health companion in your Kiro sessions. In Claude Code it is a one-step /plugin install; Codex uses the same MCP server. No account, no key.

2. Evaluated on every interaction

There is no timer running in the background. On each AI turn or when a Kiro hook fires, DevWell reads your live health snapshot and decides whether anything is worth raising.

3. Coaches inside your flow

It injects a coaching context so the assistant reminds you naturally. Productivity-phase detection protects deep flow and only interrupts for something critical.

4. Learns from real responses

It records what you follow and when, then adapts which hours and intervention types it uses. Analytics and learned patterns are computed from your logged history and say so plainly when there is not enough yet.

What it tracks

Four domains, one companion.

Each domain ships evidence-based content and feeds one health score, a burnout model and the adaptive engine.

๐Ÿ’ง

Hydration

Water goal and reminders tuned to time since your last drink, with progress bars and research-backed tips.

๐Ÿƒ

Movement

18 desk-friendly exercises across stretching, posture, eye care and light cardio, escalating the longer you sit.

๐Ÿฅ—

Nutrition

18 brain-food suggestions by time of day, meal and caffeine logging with a daily limit and a cutoff hour.

๐Ÿง˜

Mindfulness

15 guided techniques from breathing to mindful coding, matched to your stress level and work style.

๐Ÿ”ฅ

Burnout scoring

A weighted risk score from session length, break deficiency, tracked stress and health trend. Real math, not a label.

๐Ÿ“Š

Analytics that earn it

Correlations and trends computed from your logged history. When there is too little data it says so, rather than inventing a number.

Honest by design

Every number on screen is earned.

DevWell only shows a correlation once it has computed one from your real logged days. It tracks state that actually persists across sessions. Where there is not enough data, it says "not enough data yet" rather than printing a placeholder that looks like a finding.

The autonomy is real but modest: it evaluates on every AI interaction and through Kiro hooks. There is no background daemon, which the README states plainly.

adaptive_insights
> adaptive_insights
๐Ÿง  Sessions tracked: 0
   Still learning. Not enough data yet.

# after real sessions:
๐Ÿง  Sessions tracked: 3
   Peak responsiveness: ~15:00 (from history)
Built with Kiro

Spec-driven from the first commit.

The .kiro directory ships in the repo. It is the plan the code was built against, not decoration.

6

Feature specs

The MCP server plus the hydration, movement, nutrition and mindfulness modules and the autonomous engine, each specified before it was built.

3

Steering documents

Project conventions, MCP protocol guidelines and a health-science reference that keeps the recommendations evidence-based.

5

Agent hooks

Build, lint and test on change, plus a health-monitoring hook, portable across any clone via the project directory.

.kiro/
.kiro/
โ”œโ”€โ”€ specs/
โ”‚   โ”œโ”€โ”€ devwell-health-mcp.md
โ”‚   โ”œโ”€โ”€ autonomous-engine.md
โ”‚   โ”œโ”€โ”€ hydration-module.md
โ”‚   โ”œโ”€โ”€ movement-module.md
โ”‚   โ”œโ”€โ”€ nutrition-module.md
โ”‚   โ””โ”€โ”€ mindfulness-module.md
โ”œโ”€โ”€ steering/        # 3 knowledge files
โ”œโ”€โ”€ hooks/           # 5 agent hooks
โ”œโ”€โ”€ agents/          # wellness-coach
โ””โ”€โ”€ settings/        # mcp.json
  • โœ“Specs led the code. Each module and the autonomous engine started as a Kiro spec, then the implementation followed.
  • โœ“Steering kept it evidence-based. The health-science file anchors every tip to a public source.
  • โœ“Hooks kept it green. Build, lint and test run on change; the health hook wires the autonomous evaluation.
  • โœ“A wellness-coach agent defines the AI personality that delivers the coaching.
Run it yourself

Lives in your Kiro session.

DevWell's .kiro config loads it as an MCP server with proactive agent hooks. Also a one-command plugin in Claude Code and an MCP server for Codex. Fully local, no build.

.kiro/settings/mcp.json
# Kiro loads DevWell as an MCP server
{
  "mcpServers": {
    "devwell": { "command": "node", "args": ["./plugin/devwell-mcp.cjs"] }
  }
}
# + .kiro/hooks/ and .kiro/specs/  proactive, spec-driven