Skip to content

Releases: meaningfy-ws/skillery

v2.9.0 — writing-antipatterns + Neo4j LinkML generators

Choose a tag to compare

@costezki costezki released this 24 Jul 06:40
468f0a0

Added

  • writing-antipatterns skill (meaningfy-core). The register/genre-conditional catalogue of
    how writing fails: a genre map (Tutorial, How-to, Reference, Explanation, Decision, Decision
    record, Contract/specification, Coaching dialogue), a short list of defects that apply regardless
    of genre, and a single-sourced inversion matrix of genre-conditional antipatterns — a rhetorical
    question is craft in an explainer and a defect in an ADR, and the catalogue names exactly where
    each move flips. Per-genre reference files are pointer tables only, never restating an entry.
  • Tested Neo4j-targeting LinkML generators (tools/linkml_neo4j/). A Cypher-constraint
    generator and a neomodel OGM generator, vendored as tested repo tooling — not a skill asset, per
    skill-repo governance. Fixes real bugs in the generators' original form: relationship targets now
    resolve correctly against abstract LinkML classes, existence constraints are emitted for required
    multivalued scalar/enum properties, and an opt-in slot annotation lowers to a generated Neo4j
    index. project-setup projects both generators via a new --neo4j scaffold flag.
  • docs/skill-inventory.md, generated. A skill-by-skill map (purpose, category, dependencies,
    and a related-skills graph), generated with a freshness test guarding drift.

Changed

  • Install docs rewritten as a literal, audience-scoped step-by-step.
  • README skill count corrected: 23 skills across 4 role bundles.

Fixed

  • tools/opencode_gen no longer mirrors __pycache__/.pyc/.pyo into .opencode/.
  • CI dependency gap: click, linkml, neomodel, jinja2 were missing from
    requirements-dev.txt — passed locally only because a dev venv already had them.
  • OpenSpec validator quirk: its SHALL/MUST check only scans a requirement's first physical line.
  • Archived the linkml-neo4j-generators change (completed but never archived); corrected a stale
    spec delta that still named the generators' pre-relocation path.

Full changelog: CHANGELOG.md#290---2026-07-24

v2.8.0 — linkml-engineering + modelling-conventions skills + Cockburn use-cases

Choose a tag to compare

@costezki costezki released this 10 Jul 19:45
cdebb87

Two new meaningfy-architecture skills plus the Cockburn use-case catalogue.

Added

  • modelling-conventions — the shared, representation-agnostic modelling-craft layer (naming, anti-patterns, editor guardrails) reused by the modelling skills the way cosmic-python is reused by the build skills. Carries the reusable-property and URI-everywhere principles and the conceptual-definition rule.
  • linkml-engineering — the operational LinkML craft, downstream of an existing model or spec (never greenfield): derive → author → generate & guard. Reusable slots, URI-as-datatype, implicit class_uri/slot_uri, enums-as-SKOS, schema-level validation, abstract/mixins/is_a, tree_root, slot_usage, subproperty_of, imports vs subsets, complete OWL/SHACL generation, custom generator templates + make-target automation (incl. diagrams), per-module output, and a guided quality-gate process (linkml-lint + codegen-freshness + example round-trip). References LinkML 1.11.x spec + model2owl conventions/checkers.

Changed

  • conceptual-modelling refocused on the concept + model-source decision + terminology; LinkML overhang relocated to linkml-engineering, generic conventions to modelling-conventions.
  • project-setup gains a conditional LinkML branch.
  • Boundary references in architecture and cosmic-python repointed to linkml-engineering.
  • Adds the Cockburn White/Blue use-case catalogue to architecture (+ templates); wires bdd-gherkin to derive scenarios from it.

Full changelog: CHANGELOG.md

v2.7.3 — CLI-symmetric README + CONTRIBUTING

Choose a tag to compare

@costezki costezki released this 01 Jul 07:41
c656383

PATCH — documentation only. See CHANGELOG 2.7.3.

README and CONTRIBUTING put Claude Code and opencode on equal footing; external-deps compatibility stated per-CLI (code-review is the lone Claude-only optional); compatibility matrix completed. No code change.

v2.7.2 — Single-home dual-CLI authoring-rule requirement

Choose a tag to compare

@costezki costezki released this 30 Jun 07:11
b729058

Spec/housekeeping release. See CHANGELOG 2.7.2.

dual-cli-distribution gains the Single-home dual-CLI authoring rule requirement (synced from the archived dual-cli-docs-refactor change). No code or behaviour change.

v2.7.1 — dual-CLI documentation re-seat

Choose a tag to compare

@costezki costezki released this 30 Jun 07:00
53b8bf8

PATCH — documentation only. See CHANGELOG 2.7.1.

opencode support documented in the README; dual-CLI docs re-seated into the install canon with a single-home authoring rule in AGENTS.md. No code changes.

v2.7.0 — dual-CLI distribution + opencode generator

Choose a tag to compare

@costezki costezki released this 29 Jun 17:03
44054f9

See CHANGELOG 2.7.0.

Run the skillery catalogue identically on Claude Code and opencode from one source set, with a generator + drift/parity/version-sync gates. Bumps VERSION 2.6.1 → 2.7.0.

v2.6.1 — parallel lens dispatch + catalogue-complete checklist

Choose a tag to compare

@costezki costezki released this 25 Jun 09:32
616a6d1

Patch fixes for the v2.6.0 PR review:

  • Standalone review fans out five subagents in parallel, one per lens (explicit Dispatch contract; one combined subagent is a defect).
  • Review checklist is now catalogue-complete — a per-lens procedure traversing the whole cosmic-python region a lens owns, not a curated id subset; code-reviewer agent hardened to one lens per run.

CHANGELOG

v2.6.0 — two-mode five-lens PR review

Choose a tag to compare

@costezki costezki released this 25 Jun 09:07
6ddf91e

Two-mode (standalone subagent passes / interactive main-thread), five-lens (L1–L5, one subagent run each, MECE) PR review for meaningfy-code-review, with a recommendations-only L5 fit & refactoring investigation carrying gitnexus blast radius. code-reviewer agent scoped to one lens per run. New pr-review-modes capability spec.

Full notes: CHANGELOG.

v2.5.1

Choose a tag to compare

@costezki costezki released this 23 Jun 08:15
23dccc4

First tagged release since v2.3.0 — brings the accumulated develop work to main.

Added (2.5.0)

  • explanatory-writing skill (meaningfy-core) — the craft for clear Diátaxis Explanation prose: controlling metaphor, concrete-beside-abstract, question pivots, an 8-axis voice schema, positive-default knobs cross-linked to the existing house voice (company-voice.md, inherited not forked), and variant-batch validation.
  • Writing reference discipline — writing knowledge lives only in the writing-skill family; technical-writing / executive-communication / epic-planning reference it. No bundle reorg.

Fixed (2.5.1)

  • Spine validator hardened against three flakes: archive link-depth (auto-fixer + make fix-spec-links), EXPECTED_BUNDLES duplication (membership now read from marketplace.json), and un-groomed spec Purpose (now a blocking check).

Housekeeping

  • Archived voice-extension-and-enhancement + catalogue-ux-and-docs into openspec/specs/.

Full detail in CHANGELOG.md. make validate green (43 tests).