keel
daneb.github.io- Category
- Developer Tools
- Rank
- No. 2096Tools index
Previous survey · No. 2057 ·
- Type
- TOOL
- Use case
- Software Testing & Quality
- Interfaces
- CLI
- GitHub
- 2 stars
- Latest release
- v0.14.0
- Date
About
Keel is a command-line tool that puts AI coding agents (Claude Code, Codex, Copilot, Kiro) on a defined leash: it requires a written spec of what 'done' means, blocks work from shipping until builds, tests, and a review pass are green, and requires human sign-off at each stage. It records every approval and gate verdict in a tamper-evident hash chain that can be exported and verified offline without access to the underlying repo, and it ships a GitHub Actions integration that blocks pull requests lacking a passing, matching evidence record.
At a glance
keel wraps coding-agent work in approved plans, verification gates and an exportable record. It is worth evaluating when you need to show what was checked and who approved the change.
Good fit
Teams with explicit acceptance checks that want agent runs to produce a reviewable evidence bundle.
Consider something else when
Use a simpler checklist if your change is small and you have no need to retain a separate run record.
Start here
Define one testable requirement in a throwaway repository. Force a failed check, then fix it and verify the resulting bundle separately.
Cost and requirements
Rust CLI documented for macOS and Linux. It runs with a supported coding agent and the project’s configured build and test tools.
Limits to know
An intact record shows what the configured gates observed. It cannot establish that the gates cover every failure. Bundles can include the exact diff, so inspect them before sharing.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Solves a real trust gap in agentic coding by forcing agents to prove work against a spec instead of just claiming completion, with a verifiable, offline-checkable evidence bundle and CI-enforced PR gating.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.