v0.2.1

Patch release of megabrain.md — three conformance-checker fixes, all false positives found while upgrading a heavily customized 0.1.0 instance. No conformance requirement changes; no migration pack, and none needed.

What's in this release

The megabrain specification, v0.2.0 (unchanged — attached: SPEC.md)

The fixes are entirely in the managed checker (scripts/megabrain.py):

• Block scalars in frontmatter — the bundled minimal YAML parser now consumes literal (|) and folded (>) block scalars instead of reporting unexpected indentation on valid YAML. A manifest using note: >- for a long declaration no longer fails [M-4] — and no longer masks every downstream check behind an unparseable manifest
• Non-Markdown entity paths — a declared entity path that is not a .md file (an Obsidian .base derived view, say) is no longer treated as a note. The spec defines a note as a Markdown file with YAML frontmatter, so the frontmatter and filename checks ([S-1], [S-6], [S-16]) were unsatisfiable for such files
• Directory coverage in [E-1] — undeclared-directories now flags the specific notes no declaration covers, rather than any directory that isn't itself a declared path. A directory whose notes are all individually declared — a file-path entity next to a declared subdirectory — is no longer reported, and the finding names the offending files

Upgrading

Instances at 0.2.0 get the fixed checker on their next upgrade through ordinary managed-file replacement — there is nothing to do now. Instances mid-bootstrap from 0.1.0 should fetch this release rather than v0.2.0: the pre-migration conformance check (step 5) runs from the release, and 0.2.0's checker can block a conforming instance.

Full Changelog: v0.2.0...v0.2.1

v0.2.0

Second release of megabrain.md — the specification now defines how instances are distributed and how they upgrade as the standard evolves.

What's in this release

The megabrain specification, v0.2.0 (attached: SPEC.md)

One new normative section — §15 Distribution and upgrades — replacing the fork-based distribution model with a mechanism that works on heavily customized, private instances:

• The ownership boundary — every file shipped by the standard is either managed (replaced wholesale on upgrade) or instance-owned (touched only where a migration step directs it). AGENTS.md and megabrain.md are managed in structure, instance-owned in content. No merge machinery, no patch programs
• The lock file — .megabrain/lock.json records the installed version, release, and hash of every managed file as installed. It is the authoritative upgrade anchor, makes drift detection a local comparison, and collapses contract-portability checking into a hash equality test
• Migration packs — every conformance-changing release ships a MIGRATING.md index plus scripts. Steps are programmatic (idempotent POSIX bash) or agentic (prose executed by the agent), so semantic changes that no patch could express — a renamed status value across all your notes — are still shippable
• The upgrade procedure — agent-executed by design: clean-tree gate, rollback tag, sequentially applied packs, and wholesale managed-file replacement with drift warnings. Git being mandatory is what makes "replace wholesale" non-destructive
• The conformance checker — a managed, version-aware script that validates an instance against the spec_version it actually satisfies, reporting violations by requirement identifier. It gates every upgrade before and after, and bisects failed chains
• Release policy — migration packs are tested against the two preceding minor versions; older chains remain applicable in sequence but are labeled experimental

Upgrading from 0.1.0

The migration pack for 0.1.0 → 0.2.0 — which adds the lock file, the conformance checker, and the upgrade procedure to existing instances — ships with the distribution tooling tracked in the project plan. Until then, 0.2.0 is informational for existing instances: nothing in 0.1.0 content or schema changes.

Full Changelog: v0.1.0...v0.2.0

v0.1.0

First release of megabrain.md — an open standard for keeping a second brain as a git repository of plain Markdown files, read and written by AI agents as first-class clients.

What's in this release

The megabrain specification, v0.1.0 (attached: SPEC.md)

The normative document defining what makes a repository a megabrain:

• Core invariants — plain Markdown + YAML under git, files as the source of truth, procedures that live in the repository, human-in-the-loop intake, fail-loudly agents
• Archetypes — a closed set of seven structural kinds (ephemeral work item, durable entity, background context, append-only log, dated series, captured external, derived view) that make any instance legible to an unfamiliar agent
• Schema — two universal required keys (domain, date_added), reserved keys with fixed meanings, declared types for instance-defined fields
• The manifest — a single megabrain.md file in which every instance declares itself, so an agent orients in one read
• Agent runtime contract — the AGENTS.md skeleton, intent-based dispatch to procedure files, and the in-turn re-read rule that makes behavior reproducible across harnesses, models, and sessions
• Integration, lifecycle, and version-control contracts — live-queried (never mirrored) integrations, a completion protocol that logs before deleting, and a content/contract split that keeps the standard human-owned