Our product · Kiln line · Plugin for Claude Code
Your coding agent is fast. Kiln Guard makes it careful.
Quality gates and guardrails for Claude Code. Block the commands you can’t undo, keep secrets out of the chat, and stop the agent from saying “done” while your tests are red. Installs in 2 minutes.
$19 launch price, then $29. One-time payment · 14-day refund · Not on sale yet
More autonomy pays off, until the day your agent…
-
git reset --hardwipes two hours of uncommitted work
-
it.skip(…)“fixes” a failing test by skipping it
-
cat .envprints your secrets into the transcript
-
package-lock.jsongets hand-edited instead of regenerated
-
“All done!”while the build is still red
Instructions in CLAUDE.md are suggestions. Hooks are enforced. Kiln Guard packages the hooks, skills and subagents that turn “please be careful” into checks that run on every tool call.
What you get
-
3 enforced hooks
A command guard that denies rm -rf ~ and asks before force-pushes, resets, DROP TABLE, curl | sh, publishes and production deploys. A file guard for secrets, lockfiles and .git/. And a stop gate that runs your tests before the agent may finish, then sends it back to fix the cause.
-
9 one-word commands
/kiln-guard:fix-bug (failing test first, then the fix), review (an independent reviewer subagent), precommit, done (an evidence-based completion report), release, secrets, gate, setup and doctor.
-
8 skills that change how the agent works
Definition of done, test-first fixes, safe refactors, diff self-review, dependency vetting, release readiness, secret hygiene and quality gates.
-
4 focused subagents
Gatekeeper, diff-reviewer, test-writer and secret-auditor. The reviewers get no Edit or Write tools, and the test writer is instructed to change test files only.
-
3 CLAUDE.md templates
Node/TypeScript web app, Python service and a generic template, with gates and guardrails already written.
-
Readable, tested, fails closed
Plain bash you can audit in ten minutes, plus a 379-check offline test suite. If jq is missing it falls back to node or python3, and with no parser at all it asks you instead of allowing. On Claude Code 2.1.295 or later, a guard that crashes or times out blocks the call too.
How it works
- Step 1
Install
claude plugin marketplace add ~/kiln/kiln-guard claude plugin install kiln-guard@kiln - Step 2
Set up your project
/kiln-guard:setup - Step 3
Keep working
# guards run on every tool call; # the gate runs when the agent finishes a change
What’s inside
| Type | Count | Items |
|---|---|---|
| Hooks | 3 | guard-bash, protect-files, stop-gate |
| Slash commands | 9 | setup, doctor, gate, done, review, precommit, fix-bug, secrets, release |
| Skills | 8 | quality-gates, definition-of-done, diff-self-review, test-first-fix, safe-refactor, secret-hygiene, dependency-vetting, release-readiness |
| Subagents | 4 | gatekeeper, diff-reviewer, test-writer, secret-auditor |
| CLAUDE.md templates | 3 | Node/TypeScript web app, Python service, generic |
| Scripts | 5 | gate detection, secret scanner, leftovers finder, changed-lines helper, doctor |
Light on context: about 1.2K tokens always on, and hooks cost none. It works with plain Claude Code, Agentkit.best, Superpowers or your own skills. New to the building blocks? Read skills vs subagents vs hooks.
Guarantee
The 14-day “it earned its keep” guarantee
If Kiln Guard doesn’t catch something worth catching in your first two weeks, email us for a full refund. No questions asked.
Frequently asked questions
Does it slow Claude Code down?
The guards take milliseconds. The stop gate only runs when the agent’s turn changed files (edits or commits) that haven’t already passed, so keep your gate command under about 60 seconds (lint, typecheck and unit tests).
Is it a sandbox?
No, it’s a seatbelt. The guards match the common spellings of dangerous commands (quotes, sudo, bash -c, chains, absolute paths), not every possible one, and file rules cover Claude’s file tools rather than every shell write. Keep backups and use git. The README lists the exact limits.
Can the agent get stuck in a loop?
No. The gate gives up after 2 consecutive failures (configurable) and hands control back to you. A warn-only mode is available too.
I already use Agentkit.best or another kit. Do I need this?
Kiln Guard is an enforcement layer, not a workflow kit, so it sits underneath whatever you use. Its commands are namespaced (/kiln-guard:…), so nothing collides.
Which stacks and systems are supported?
Gate detection covers Node (npm, pnpm, yarn, bun), Python (ruff, mypy, pyright, pytest, uv, poetry), Go, Rust and Makefiles; any stack works if you type your gate command. It runs on macOS and Linux, and on Windows through WSL or Git Bash.
Is Kiln Guard made by Anthropic? Can I use it for client work?
No. Kiln Guard is an independent product by Claudekit.ai; Claude and Claude Code are trademarks of Anthropic, PBC. One licence covers one person on unlimited personal, client and commercial projects, and every 1.x update is free.
Give your agent more autonomy, safely.
One-time purchase for one developer. Unlimited projects, every 1.x update included.
$19 launch price, then $29. One-time payment · 14-day refund · Not on sale yet
Kiln Guard is an independent product by Claudekit.ai. Not affiliated with, endorsed by, or sponsored by Anthropic. Claude and Claude Code are trademarks of Anthropic, PBC.