kiro-goal-mode

Autonomous goal-driven workflow for Kiro IDE — inspired by Claude Code's /goal command.

Set a goal in your chat message, and Kiro will keep working toward it automatically until the condition is met.

How it works

1. You state your goal in the first message (e.g., "Goal: all tests in src/auth pass and lint is clean")
2. Kiro works toward it
3. After each agent turn, the agentStop hook fires and evaluates whether your goal has been achieved
4. If NOT achieved → agent automatically continues working
5. If achieved → agent announces completion and stops

Installation

Copy the hook and steering files into your workspace's .kiro/ directory:

# Clone this repo
git clone https://github.com/YOUR_USERNAME/kiro-goal-mode.git

# Copy files to your workspace
cp kiro-goal-mode/.kiro/hooks/goal-evaluator.kiro.hook YOUR_PROJECT/.kiro/hooks/
cp kiro-goal-mode/.kiro/steering/goal-mode.md YOUR_PROJECT/.kiro/steering/

Or manually copy the files from this repo's .kiro/ directory.

Usage

In Kiro chat, start your message with a clear goal statement:

Goal: all unit tests pass and there are no TypeScript errors

The auth module needs refactoring to use the new token validation library.

Kiro will:

• Parse your goal from the conversation
• Work on the task
• Auto-evaluate after each turn
• Continue until the goal condition is satisfied

Goal writing tips

Write goals that are verifiable from conversation output:

Good	Bad
"all tests pass (npm test exits 0)"	"code is clean"
"no TypeScript errors in src/"	"everything works"
"API responds 200 for all endpoints in test suite"	"API is fast"
"file count in src/modules/ is under 10"	"code is organized"

Include:

• A measurable end state (test result, build exit code, file count)
• How to verify it (which command to run)
• Constraints (what must NOT change)

Disabling

To temporarily disable goal mode:

• Delete or rename goal-evaluator.kiro.hook
• Or set "enabled": false in the hook file

Limitations

• No separate evaluator model — the same agent evaluates its own progress
• No built-in turn counter or token spend tracking
• Goal must be stated in conversation (not set via command)
• No /goal clear equivalent — start a new conversation to reset
• Evaluation quality depends on context window size

How it compares to Claude Code's /goal

Feature	Claude Code	kiro-goal-mode
Set goal dynamically	/goal condition	State in chat message
Separate evaluator	Yes (Haiku)	No (same agent)
Status indicator	◎ with timer	No
Clear goal	/goal clear	New conversation
Non-interactive mode	claude -p "/goal ..."	Not supported
Turn/token tracking	Yes	No
Auto-continue	Yes	Yes (via agentStop hook)

License

MIT