gg_dna 4.1.2

The DNA engine - resolves DNA packages declared as dev-dependencies, merges their dna folders and instantiates configs, docs, scripts and agent skills into the target repo.

Changelog

4.1.2 - 2026-08-12

Removed

• BREAKING: File-naming conversion at instantiation. DNA files now instantiate under exactly the name they carry in the layer — no kebab/snake/camel conversion, and no rewriting of references to renamed files. The fileNaming config key is gone; a _dna.json still setting it now warns as an unknown key. Layers that relied on conversion must rename their DNA files to the name the target should see.
• BREAKING: The dot_ dot escape. dot- is the only accepted form; the dot_ variant only existed for snake_case layers, which the removal of file-naming conversion makes obsolete. The placed DNA test now rejects a dot_-escaped path in dna/ before instantiating and reports the rename to make (Rename dna/dot_vscode to dna/dot-vscode.) instead of silently instantiating a literal dot_vscode/ folder.
• Remove auto renaming of dna files. Did break links.

4.1.1 - 2026-08-12

Changed

• A dna/ folder without dna/_dna.json is a hard error instead of silently falling back to "role": "project" — that fallback let the engine treat a hand-authored DNA as generated content.
• Ensure correct naming, Require dna/_dna.json

Fixed

• The snake_case dot escape dot_ is decoded like dot-, so dna/dot_vscode/settings.json also instantiates to .vscode/settings.json.

4.1.0 - 2026-08-09

Changed

• Increase version
• Publishing dna

4.0.1 - 2026-08-08

Added

• Dotfiles in DNA content are escaped with a dot- prefix: dna/dot-vscode/settings.json instantiates to .vscode/settings.json. Without it dart pub publish silently drops them.
• Layer resolution goes through the manifests and their lock files (pubspec.lock, pnpm-lock.yaml): the lock supplies the ecosystem, the resolved version and the installed-name index, while .dart_tool/package_config.json and node_modules/ supply the folder.
• The npm scope is dropped when folding a package name to its layer identity, so a DNA published to both registries (@tssuite/base-dna and base_dna) is one layer, not two. A warning fires when the two copies carry different dna/ trees.

Changed

• Distribute DNA using npm and dart package inheritance
• Set version back to 1

Changed — breaking

• DNA configuration moved from .gg/dna.json to dna/_dna.json. dart pub publish drops every path with a leading dot, so a config below .gg/ never reached a pub-installed consumer — the layer silently fell back to defaults. dna/ is the one folder both ecosystems publish.
• dna/_dna.json is hand-authored and the engine only ever reads it. Everything the engine writes moved to dna/_generated.json, which also absorbed dna/_instances.json.
• Layers are declared explicitly in layers and referenced by the package name they are declared under in pubspec.yaml / package.json. The old order key is gone, and so is the inference from dependency declaration order — a dependency that is not listed is not a layer.
• Path overrides (dependencies: { x: { path: … } }) were removed. Local checkouts are wired up by gg_localize_refs, which writes pubspec_overrides.yaml / pnpm-workspace.yaml and thereby repoints .dart_tool/package_config.json and the node_modules symlink.
• A package is a DNA layer only if its dna/_dna.json declares "role": "dna". A dna/ folder alone no longer qualifies.
• config.claude.claude_md.include flattened to claude.claudeMdInclude.
• "version": 1 is required in both files.
• pnpm is the only supported npm package manager; package-lock.json and yarn.lock are not read.
• No migration path: .gg/dna.json, dna.yaml and dna: blocks in pubspec.yaml / package.json are simply no longer read, and format version 5 files are rejected.

Fixed

• resolvedVersion read package.json even for pub-resolved packages. A pub tarball ships both manifests, so a pub layer reported the npm version. The version now comes from the lock file of the ecosystem that provided the copy, and from the on-disk manifest for localized checkouts.
• A dna/_generated.json that exists but cannot be read now throws instead of being treated as "never instantiated" — the latter made every instance count as unowned and overwrote hand edits without the hand-modified check ever firing.
• Fix versions

4.0.0 - 2026-08-07

Added

• Instances: every public file of the merged dna/ replica (path segments not starting with _) is copied to its project location, with ownership tracking, adoption of existing files, and removal of files the DNA no longer produces — including the folders they leave empty
• Inheritance tree: DNAs are declared as dev-dependencies (npm/pub); parent DNAs are regular dependencies of their child DNA; gg_dna resolves the tree recursively from node_modules/ and .dart_tool/package_config.json (diamond dedup, cycle detection)
• JSON overrides: sidecar X.overrides.json merges into X.json — objects deep-merge, null deletes, "key!" replaces outright, "key+" joins arrays (deduplicated); JSONC (comments, trailing commas) is tolerated
• Variables: dna/_vars.json (deep-merged across layers, overridable via vars in .gg/dna.json); references dnaMyVar, DnaMyVar, dna_my_var, DNA_MY_VAR, dna-my-var are replaced case-adaptively; non-identifier values verbatim
• File-naming conversion: DNA files are canonical kebab-case and are converted per target (pubspec.yaml → snake_case, package.json → camelCase, configurable via fileNaming); references to renamed files are rewritten in text instances
• gg_dna init: places the DNA wrapper test (Dart and/or vitest), a .gg/dna.json skeleton and the !.gg/dna.json gitignore exception
• Public engine API: instantiateDna() and runDnaTest() (package:gg_dna/gg_dna.dart); the engine core is free of dart:io and compilable to WebAssembly (host access via injectable DnaHost)
• Per-file guard: an existing file that carries uncommitted work is never overwritten or deleted — the run names those files and writes nothing; unrelated dirty files do not block it
• Everything the DNA generates is committed right away as #gg: generated DNA (path-limited, so unrelated changes stay in the working tree); without a repository or git identity the files are kept for a manual commit
• Provenance in every failure report: generated files are reported with the DNA source they are produced from (edit instead: base_dna/dna/doc/develop.md), so hand edits go into the DNA instead of the generated copy (DnaInstantiationResult.sources)
• global.overrides.md in the src root of any layer: its string blocks rewrite {{@tag:…}} placeholders in every merged .md file; file-specific overrides of the same layer win, heading-form blocks warn
• Warnings (file + line) for leftover pre-4.0 tag notation

Changed

• BREAKING: dna/src is gone — a DNA's dna/ folder itself mirrors the project root (public/private via the _ prefix convention)
• BREAKING: DNA configuration lives only in .gg/dna.json; dna: blocks in pubspec.yaml/package.json and dna.yaml are migration errors
• BREAKING: gg_dna sync was replaced by gg_dna init — the instantiation runs inside the placed test on every test run (hand-modified instances fail, DNA updates are written and committed)
• BREAKING: git layers were removed — declare DNAs as dev-dependencies; path: overrides in .gg/dna.json remain for local development
• BREAKING: the bookkeeping is split in two and renamed — dna/_dna.json (format v5: layers with package/via, hashes) and dna/_instances.json (the files the DNA owns); leftover .dna.json / .instances.json files are read once, so ownership survives, and then removed
• BREAKING: the effective variables are no longer duplicated into the bookkeeping — dna/_vars.json is their only home
• BREAKING: config: claude: skills: was removed — skills ship as normal instances at dna/.claude/skills/<name>; only the managed CLAUDE.md block (claude_md: include:) remains special
• Base DNA restructured: human documentation at dna/doc/** (conventions, guides — English), skills at dna/.claude/skills/**
• BREAKING: every DNA source ships its mergeable DNA under dna/src — the gg_dna base DNA, git layers, and path layers; sources without dna/src are an error with a migration hint
• BREAKING: <target>/dna/src is the implicit last layer — it survives the sync verbatim, is never listed in the config (layer name src is reserved), and replaces dna/_override; path layers pointing into <target>/dna are an error
• BREAKING: override files are named <file>.overrides.md — the old .tag.md suffix is a hard error with a rename hint
• BREAKING: new tag notation ## [@tag] … and {{@tag:default}} (comment markers <!-- @tag -->); the old notation without @ is no longer recognized
• BREAKING: .dna.json manifest format v4 (records the implicit src layer; --check reports older manifests as outdated — run gg_dna sync)

Removed

• gg_dna sync (including --check, --source, --target), git tag resolution, the skills mirroring and the staging/backup folders (.gg_dna_staging, .gg_dna_backup are no longer created)

3.1.1 - 2026-08-07

Added

• Add .prettierignore and .prettierrc to dna

Changed

• BREAKING: TypeScript projects instantiate with kebab-case file names — package.json no longer defaults to camelCase (the value stays selectable via fileNaming); Dart projects keep snake_case
• Rework DNA repos
• Provide first DNA, i.e. installation and .vscode settings
• Define dna repos

3.1.0 - 2026-08-04

Note: released out of order — this version shipped after 4.0.0 and contains the rename below on top of the 4.0.0 changes.

Changed

• Rename .master to .ocean with automatic migration at next start

3.0.0 - 2026-07-24

Added

• dna: config: claude: section: claude_md: include: maintains a managed @-import block in the target's CLAUDE.md (files stay as-is, folders expand to their .md files); skills: include: mirrors skill folders into .claude/skills — only skills gg_dna installed are overwritten or removed, hand-installed skills are never touched
• gg_dna sync creates CLAUDE.md when missing and removes leftover pre-3.0 gg_dna:conventions blocks
• gg_dna sync --check also verifies the CLAUDE.md block and the installed skills

Changed

• BREAKING: layers now live under dna: dependencies: — the pre-3.0 syntax (layer maps directly under dna:) is rejected with a migration hint
• BREAKING: gg_dna sync is fully non-interactive — the install-skills and apply-conventions subcommands, all prompts, and the --no-install flag were removed; everything is driven by the dna: config: block
• BREAKING: .dna.json manifest format v3 (adds the claude section; --check reports older manifests as outdated — run gg_dna sync once)

2.1.1 - 2026-07-24

2.1.0 - 2026-07-15

Changed

• Support dna config via dna.yaml and package.json for non-Dart repos
• Harden dna config discovery: require dna block in dna.yaml, tolerate foreign package.json dna fields and undecodable sibling files

2.0.0 - 2026-07-15

Added

• Add init-architecture skill that creates/updates README.architecture.md
• Add dna sync manifest with hash/check support and overlay tracking
• Multi-layer DNA: layers are configured in the target pubspec.yaml dna: block — git layers (optionally pinned via a semver version: constraint resolved against the repo tags), local path layers, and repo-local layers like dna/_override that survive the sync verbatim
• Markdown tag overrides: ### [tag] sections and {{tag|default}} strings can be overridden per layer via X.tag.md files; all markers are rendered away in the synced output

Changed

• BREAKING: gg_dna sync no longer accepts a positional overlay argument — configure layers via the dna: block in the target pubspec.yaml
• BREAKING: the overlay stored in .dna.json is no longer auto-reused
• BREAKING: .dna.json manifest format v2 (ordered layer list; --check reports pre-2.0 manifests as outdated)
• Release 2.0.0 docs: changelog, readme, dna config example
• Sync builds the new dna tree in a staging folder and swaps it in via renames — errors can no longer destroy the existing dna/ (in particular in-dna override layers like dna/_override); interrupted swaps are recovered from the backup folder on the next run
• Layers are resolved in parallel (clones, ls-remotes) — also in --check
• Prefer stable tags over prereleases in semver resolution
• Doc-comment density rule (max 3 lines before classes, 1 per member) added to the shipped code conventions
• Use MIT license and shorten package description for pub.dev

Fixed

• Crash (RangeError) when the same section tag appeared on nested headings
• Crash (TypeError) in --check on structurally malformed .dna.json
• Longer code fences now nest shorter ones (CommonMark closing rule), so fenced markdown examples inside fences stay untouched
• Layer hashes ignore .git/ and are line-ending agnostic — no more false --check drift after git activity in a layer folder or CRLF checkouts
• A foreign [tag] in a replacement heading is replaced instead of double-prefixed; mixed line endings normalize to the dominant one
• A configured but missing in-dna layer (e.g. dna/_override on a fresh clone — git does not track empty folders) is skipped as empty instead of failing the sync; --check agrees
• Backslash path: values in the pubspec work on every platform
• Fix negative hex hash rendering and cwd race between parallel test suites

1.1.0 - 2026-05-26

Added

• Initial boilerplate.
• Add install-skills tests and .gitattributes; refactor for testability
• Add convention docs and apply-conventions command
• Add sync command and refactor install-skills/apply-conventions
• Add install.bat script and ignore .aicode in .gitignore
• Add workflow and commit discipline guides for gg-kidney and rljson
• Add init skill guide for generating project-aware CLAUDE.md

Changed

• migrate claude/ configs to agents/, update references throughout
• Move dna files to dna folder