v0.1.3 — contributable, and the versions caught up with reality

The month where this stopped being a repo you read and became one you can contribute to.

New

• Docs site at https://palo-alto-ai-research-lab.github.io/claude-bible/, linked from the README.
• AGENTS.md — how to verify a change when there is nothing to execute. Four structural checks, written because "it is prose, there is nothing to run" is not a verification story, and an agent sending a PR here needs to know what proof looks like.
• Every open roadmap row is now a claimable issue. accepted means scoped, free to take, nobody on it: comment "claiming this" and it is yours for 7 days. The five open ones are the memory index-vs-body discipline (#1), the test-after-build gate (#2), declined-decisions revisit/OVERTURN mechanics (#3), the taste gate (#4) and history-aware CRM drafting (#5).
• The contributor deal, in the open: no CLA, you keep your copyright, an answer within 48 hours including "no, and here is why". It is inherited from one org-wide CONTRIBUTING.md instead of a local copy — we found six repos carrying a byte-identical stale copy that silently shadowed the real terms, and deleted them.
• CITATION.cff, SUPPORT.md, the lab-wide AI-contributor credit policy, and changelog categories for these auto-generated notes.

Versioning, corrected

The README and the changelog both promised "a versioned release with a changelog twice a week (Mon & Thu)". Between 4 July and 4 August not one release was cut, while the commits kept landing. The promise has been replaced with a rule tied to the work rather than the calendar: every noticeable change ships as a new release, and this feed is the record of maturity.

Two versions that existed only in the changelog (v0.1.2) or only as a tag with no notes (v0.1.0, v0.1.1) have been reconciled today — with their real dates in the text, not backdated tags.

What's next

The five open roadmap issues above. A pain leaves the roadmap by becoming its own repo — two already have.

Full Changelog: v0.1.2...v0.1.3

v0.1.2 — pain #8 leaves as agent-leash (written 4 July 2026)

Written 4 July 2026, tagged 4 August 2026. The content is the July commit; the tag was simply never cut at the time, and it is recorded that way rather than backdated.

Roadmap pain #8 — agent security and delegated authority — shipped as its own repo: agent-leash.

Inside it: the LEASH-8 eight-domain control model, a 24-statement scored self-assessment, the plan-vs-authorize architecture pattern, an approval-design checklist, and a reference A2A Agent Card.

Claim discipline carried across with it: controls and coverage, never "secure".

Full Changelog: v0.1.1...v0.1.2

v0.1.1 — pain #5 leaves as claude-consensus (2 July 2026)

Backfilled release note. The v0.1.1 tag has pointed at this commit since 2 July 2026.

Roadmap pain #5 — "multiple machines, one system" — shipped out of order as its own repo: claude-consensus.

That repo carries the consensus protocol, the dual-rail bus, ACK discipline, the leader/follower canon, self-healing sync and a sanitized reference implementation in stdlib-only Python.

The Bible stays the family map; the diplomacy now has its own home. This is what the roadmap is for — a pain leaves as a repo, and the row is marked shipped.

Full Changelog: v0.1.0...v0.1.1

v0.1.0 — the governance skeleton (2 July 2026)

Make Claude Code behave consistently across repos, sessions and humans. Rules-as-files governance, extracted from a live production system.

Backfilled release note. The v0.1.0 tag has pointed at this commit since 2 July 2026; what was missing was the changelog, not the code.

What is in it

The governance skeleton:

• docs/SPEC.md — rule anatomy (trigger + essence + pointer), the frontmatter schema (origin, date_established, status, supersedes, audience), precedence (newer beats older; a rule whose origin is the owner can only be overridden by the owner), the routing tree that decides whether a thing belongs in the codex, the agent config, memory or a hook, the declined-decisions journal, objection sparring, and the intake ritual.
• templates/ — rule, decision memo, declined-decisions journal.
• examples/rule-objection-sparring.md — a real rule, born the same day from a real overturned verdict.
• FOR-ROBOTS.md — the entry point for an AI agent mining this repo.
• devlog/2026-07-02.md — how this release happened: the decision chain and the research inputs.
• docs/the-day-my-ai-said-no.md — the launch story.
• ROADMAP.md — the pain-driven plan of what gets opened next.

The problem it solves

One behavioural codex that humans and agents follow, where a rule has a defined home, a defined precedence, and a defined way to be argued out of existence. Without that, "the rules" live in six files, three chat histories and one person's head, and every session re-litigates them.

Full Changelog: https://github.com/Palo-Alto-AI-Research-Lab/claude-bible/commits/v0.1.0