All notable changes to Keel are recorded here. The format follows Keep a Changelog, and Keel aims to follow Semantic Versioning — for a docs kit, that means:
- major — a breaking change to file names, ledger formats, or role contracts adopters depend on;
- minor — a new rule, chapter, template, or file;
- patch — corrections, wording, and fixes that change nothing structural.
- The enforcement hook is now actually shipped:
.claude/hooks/block-agent-commits.sh(denies agentgit commit/git pushvia thehookSpecificOutput/permissionDecisioninterface current Claude Code hook docs recommend) plussettings.example.jsonto register it. Previously the hook was only described in prose. CI now proves the hook's behavior on every push (.github/scripts/test_hook.sh) — the evidence rule, applied to Keel's own claims. templates/AGENTS.md— the blank starter template. Keel's rootAGENTS.mdis now filled in for this repo itself (dogfooding); README and adoption guide point adopters at the template.- Worked examples under
examples/— a greenfield (linkli) and a brownfield (billing) scenario with filled-inAGENTS.mdand ledgers, so the model is easy to grasp from real content. - ELI5 explainer — a "five-year-old version" of the pitch (robots, LEGO, two notebooks) in the README, guide chapter 1, and the web guide.
- Ledger chapter: a "Who owns an entry: proposed vs. ratified" section codifying that agent-drafted entries are proposed claims ratified at commit, and that a rejected rationale is superseded explicitly.
- Standardized the project name to Keel across the guide, README, web guide, and templates.
- Refreshed Anthropic references (verified 2026-07): the
CLAUDE.md→@AGENTS.mdimport stub is now the interop pattern officially documented in Anthropic's memory docs, not just a community workaround — noted in the README, adoption chapter, research analysis, and web guide. Native AGENTS.md support (anthropics/claude-code#6235) remains open.
0.1.0 — 2026-07-15
First public release.
- The minimal kernel:
AGENTS.md(gates + project map), theCLAUDE.mdimport stub, and the two ledgers —docs/decisions.mdanddocs/falsified.md— with ADR/MADR-informed formats. - The three role prompts:
roles/coder.md,roles/reviewer.md(with the over-flagging guard), androles/auditor.md(dead-list check first). - The eight-chapter guide under
docs/guide/, plus a self-contained web version atdocs/site/index.html(deployed to https://keel.byaru.com). - Templates (
templates/), an eval-harness scaffold (evals/), and an optional enforcement-hook example (.claude/hooks/). - The research analysis comparing Keel to spec-kit, Kiro, AGENTS.md, and ADR practice
(
docs/research/playbook-analysis.md). - Project scaffolding: CI (internal-link check + a 200-line budget on
AGENTS.md), GitHub Pages auto-deploy, issue/PR templates,CONTRIBUTING.md,CODE_OF_CONDUCT.md, and this changelog.