Releases: Data-Wise/craft
v2.21.0 — Orchestrate Pipeline & Insights Lifecycle
Highlights
/craft:orchestrate:planCommand — New pipeline from spec → ORCHESTRATE → worktree. Discovers specs, parses phases, generates session-ready ORCHESTRATE files with friction prevention, and creates worktrees for isolated development./craft:insightsCommand — Generates usage pattern reports from session facets data. Aggregates friction patterns, goal categories, and CLAUDE.md suggestions. Supports--format html|terminal|jsonand--since <days>filtering.- Brainstorm → Orchestrate Pipeline — New Step 6 offers ORCHESTRATE + worktree creation after spec capture, connecting the full workflow: brainstorm → spec → ORCHESTRATE → worktree → implement → PR.
- Brainstorm Insights Integration — New Step 1.8 checks session insights for relevant past friction patterns before brainstorming.
- AskUserQuestion Enforcement — Brainstorm command now uses exact JSON tool call syntax instead of prose references.
New Features
- Worktree Types Taxonomy: 4-type system (Manual, Pipeline, Swarm, Cross-Repo) documented across REFCARD, orchestrate command, worktree reference, and tutorials
- Swarm Clarity: Enhanced
--swarmdocumentation with practical dry-run examples and decision guidance - Cross-Command Integration: Insights wired into
/craft:dorouting,/craft:hubdiscovery, and/craft:check --contextmode - Insights Lifecycle: Full documented flow from sessions → facets → insights → CLAUDE.md rules
New Files
| File | Purpose |
|---|---|
commands/orchestrate/plan.md |
Spec → ORCHESTRATE → worktree pipeline |
commands/workflow/insights.md |
Session insights report command |
docs/guide/pipeline-orchestrate-guide.md |
Pipeline guide (237 lines) |
Stats
- 111 commands | 25 skills | 8 agents | ~1575 tests passing
- 42 files changed
Full Changelog: v2.20.0...v2.21.0
v2.20.0 — Homebrew Refactor: Security, Subcommands & Release Hardening
Highlights
- Workflow Security Hardening — env indirection prevents script injection,
sha256sum+--retry 3+ 64-char SHA guard inhomebrew-release.yml - Subcommand Consolidation — 8 → 6 subcommands:
validate→audit(with--build),tokenfolded into setup,release-batchdropped,depsexpanded with inter-formula graph - Formula Name Mapping —
.craft/homebrew.jsonconfig with 3-priority lookup chain (config → git remote → basename) - Release Skill Hardening — Step 8.5 uses
shasum -a 256(macOS-portable),ruby -csyntax validation, SHA guard - New Documentation — Homebrew Automation Guide (310 lines), Setup Tutorial, Quick Reference Card
Changes
Security
homebrew-release.yml: env indirection forgithub.event_nameandgithub.event.inputs.version- SHA256 calculation:
sha256sumon CI (Ubuntu),shasum -a 256in release skill (macOS) curl --retry 3 --retry-delay 2for resilient downloads- 64-char hex guard catches empty/corrupt SHA before workflow proceeds
ruby -cvalidates formula syntax aftersedupdates
Commands
dist:homebrew: 8 → 6 subcommands (formula, workflow, audit, setup, update-resources, deps)audit: replacesvalidate, adds--buildflag forbrew install --build-from-sourcedeps: inter-formula dependency graph + system deps matrix +--dotGraphviz outputtoken: folded into setup wizard Step 4
Configuration
.craft/homebrew.json: formula name + tap + source type config- Release skill Step 8.5: 3-priority lookup (config → git remote → basename)
Documentation
- New:
docs/guide/homebrew-automation.md(310 lines) - New:
docs/tutorials/TUTORIAL-homebrew-setup.md - New:
docs/reference/REFCARD-HOMEBREW.md - Updated: dist commands, REFCARD, homebrew installation guide
Tests
- 50 new dogfood tests (
test_homebrew_refactor_dogfood.py) - 47 new e2e tests (
test_homebrew_refactor_e2e.sh)
Stats
109 commands | 25 skills | 8 agents | ~1575 tests passing
Full Changelog: v2.19.0...v2.20.0
v2.19.0 — Marketplace & Insights-Driven Improvements
Highlights
- Marketplace Distribution — First-class
dist:marketplacecommand for publishing to the Claude Code marketplace, withmarketplace.jsonschema and GitHub source integration - Insights-Driven Improvements — Guard audit skill (5-step pipeline), insights apply skill, autonomous releases, context-only checks, worktree validation, swarm orchestration
- Guard Audit Skill — Analyzes
branch-guard.shrules for false positives and proposes JSON config changes (read-only, never modifies guard script) - Autonomous Releases —
--autonomousflag for unattended release pipeline with--adminsafety warnings and audit trail - PreToolUse Hook — Python hook that warns when file operations target paths outside the current worktree
Changes
Features
- Add
dist:marketplacecommand for marketplace distribution channel - Add
guard-auditskill — 5-step branch guard audit pipeline - Add
insights-applyskill — extracts and applies CLAUDE.md suggestions from insights reports - Add
--autonomousflag to release skill for unattended pipelines - Add
--contextflag to check command for context-only phase detection - Add
validateaction to worktree command for health checks - Add
--swarmflag to orchestrate command for unlimited parallel agents - Add PreToolUse Python hook for worktree path validation
Fixes
- Fix VERSION-HISTORY v2.17.0 status (was "In Development", now "Released")
- Fix phase detection gap in check command (pr-prep now checks both PR exists AND clean tree)
- Add safety warning for
--autonomous --adminoverride in release skill, tutorial, and guide - Reconcile marketplace schema with spec, fix stale counts
- Use GitHub source object in marketplace.json
Documentation
- Marketplace distribution guide, tutorial, and getting-started integration
- Insights-driven improvements guide (7 Mermaid diagrams) and tutorial (9 steps)
- Hub, REFCARD, and QUICK-START updated with new features
Tests
- 57 marketplace distribution tests (dogfood + e2e)
- 89 insights improvements tests (59 dogfood + 30 e2e)
Stats
- 109 commands | 25 skills | 8 agents | ~1575 tests passing
- 33 commits since v2.18.0
Full Changelog: v2.18.0...v2.19.0
v2.18.0 — Release Pipeline Enhancements
Highlights
- CI mirror pre-flight —
/craft:check --for releasenow runs full test suite, lint, and security as Step 2a in the release pipeline, catching failures before push - README/index version checks —
pre-release-check.shnow validates version badges in README.md and docs/index.md (check 4/5), preventing stale versions in releases - Release command delegation —
/craft:code:releasedetects Craft plugins and delegates to the skill's 10-step pipeline automatically - CLAUDE.md trimmed 74% — 587 → 155 lines via lean pointer architecture
Changes
Features
- Add
/releaseskill for end-to-end release pipeline automation (10-step pipeline with dry-run) - Add release skill tests (e2e + dogfooding), dry-run mode, refcard, and tutorial
- Add
/craft:check --for releaseas CI mirror pre-flight (Step 2a) - Add README.md and docs/index.md version checks to
pre-release-check.sh(check 4/5)
Documentation
- Trim CLAUDE.md from 587 to 155 lines (74% reduction)
- Sync release pipeline docs across all tutorials, refcards, guides, and workflows
- Update
/craft:code:releaseto delegate to release skill for Craft projects - Correct skill count: 21 → 23 across all docs
Fixes
- Update orch-flag test to check VERSION-HISTORY.md instead of CLAUDE.md (avoids content-coupling anti-pattern)
Stats
- 108 commands | 23 skills | 8 agents | ~1504 tests passing
- 9 commits since v2.17.0
Full Changelog: v2.17.0...v2.18.0
v2.17.0 — Branch Guard v2: Smart Protection Mode
Branch Guard v2: Smart Protection Mode
Highlights
- 3-tier risk classification (LOW/MEDIUM/HIGH) replaces binary block/allow
- Fade-to-brief learning — verbosity decreases over your session
- One-shot approvals with 5-min TTL — stale markers auto-expire
- Destructive command detection — blocks dangerous git operations and rm variants targeting .git
- Critical file protection — blocks writes to .env, .pem, .key, branch-guard.json on dev
- Bash write-through detection — catches redirect/tee/cp creating new code files on dev
Code Review Fixes
- Race condition: flock-based atomic append to session counter
- TOCTOU: file existence re-check after age-out
- Broadened rm .git regex to catch all flag ordering variants
Documentation
- Smart mode guide (
docs/guide/branch-guard-smart-mode.md) - Setup tutorial (
docs/tutorials/TUTORIAL-branch-guard-setup.md) - Quick reference card (
docs/reference/REFCARD-BRANCH-GUARD.md) - Updated mkdocs.yml navigation and index page
Tests
- 1449 total tests (up from 1432)
- 155 branch guard tests: 94 unit + 31 e2e + 17 review fixes + 6 integration + 7 dogfood
Full Changelog: v2.16.0...v2.17.0
v2.16.0 — Teaching Ecosystem + Branch Protection
What's New
Branch Protection Hooks
scripts/branch-guard.sh— PreToolUse hook that prevents edits on protected branches- Protection levels:
block-all(main) blocks all writes;block-new-code(dev) blocks new code files but allows docs/specs/existing edits - Bypass:
/craft:git:unprotectcreates a marker file;/craft:git:protectremoves it - Dry-run mode:
.claude/branch-guard-dryrunmarker logs blocks but allows through - New commands:
/craft:git:protectand/craft:git:unprotect - 138 tests: 49 unit + 31 e2e + 6 integration + 52 dogfooding
Teaching Ecosystem Coordination
- Config normalizer:
_normalize_config()maps flow-cli schema to Craft-native format - Single-day break support: Break validation now accepts
start == endfor holidays - 8 normalization tests in teaching workflow integration suite
- Dedicated Teaching tab in docs site navigation
Housekeeping
- Archived 23 completed specs to
docs/specs/_archive/ - Fixed 12+ broken links from spec archival
- Test count: 1294 → 1432 (+138)
- Command count: 106 → 108 (+protect, +unprotect)
Full Changelog: v2.15.0...v2.16.0
v2.15.0 — Brainstorm v2.5.0: Spec Simplification + Smart Questions
Brainstorm v2.5.0: Spec Simplification + Smart Questions
Spec Simplification (84% reduction)
- Brainstorm command spec: 1,919 → 312 lines (84% reduction)
- Extracted to 4 reference documents:
docs/specs/SPEC-brainstorm-question-bank.md— Full question bank with all Q&A textdocs/tutorials/TUTORIAL-brainstorm-power-user.md— Power user guide with advanced patternsdocs/reference/REFCARD-BRAINSTORM.md— Quick reference card with flowchartdocs/VERSION-HISTORY.md— Brainstorm evolution table (v2.1→v2.5)
Context-Aware Smart Questions
- New utility:
utils/brainstorm_context.py(~280 lines)- Scans
.STATUS,docs/specs/, git log,CLAUDE.mdfor project context - Pre-fills answers from existing project state
- Detects prior brainstorms, matching specs, failing tests
- Scans
- Project-type question extensions: 6 types (R, Python, Node.js, Quarto, Claude Plugin, Teaching) with 12 new questions
- Dynamic questions: Matching spec detection, prior brainstorm resume, failing test alerts
Tests
- 38 new tests in
tests/test_brainstorm_context.py(all passing) - 53 + 24 existing brainstorm tests verified
- 1286 total tests across the project
Documentation
- 3 new nav entries in mkdocs.yml
- Version refs synced across README, docs/index.md, CLAUDE.md, mkdocs.yml
- All CI checks passing
Full Changelog: v2.14.0...v2.15.0
v2.14.0 — Unified Formatting Library
What's New
Unified Formatting Library (scripts/formatting.sh)
Shared bash library (~180 lines) providing box-drawing (double/single line), FMT_ prefixed color constants, ANSI-aware padding, table formatting, and source guard. All boxes standardized to 63 visible characters.
API: box_header, box_single, box_row, box_separator, box_footer, box_empty_row, box_table, fmt_set_width, fmt_divider, _fmt_strip_ansi
Migrations
- 8 box-drawing scripts migrated (install.sh, migrate-from-workflow.sh, convert-cast.sh, health-check.sh, consent-prompt.sh, dependency-installer.sh, dependency-manager.sh)
- 15 color-only scripts migrated (validate-counts, pre-release-check, batch-convert, repair-tools, 3 installers, tool-detector, version-check, sync-version, verify-phase1/2, install-hooks, test-fix-flag, pre-commit-markdownlint)
Tests
- 74 new tests (28 unit + 30 integration + 16 edge cases)
- 1248 total tests passing
Documentation
- Formatting library guide, migration tutorial, reference card
Full Changelog: v2.13.1...v2.14.0
v2.13.1 — Fix plugin.json strict schema compatibility
What's Changed
Fixed
- Plugin loading: Moved
claude_md_budgetfromplugin.jsonto.claude-plugin/config.json— Claude Code's strict schema rejects unrecognized keys, silently breaking plugin loading (anthropics/claude-code#20415) - Budget utilities: Updated
claude_md_optimizer.py,claude_md_sync.py, andclaude-md-budget-check.shto read budget fromconfig.json - Homebrew formula: Dual-protection JSON cleanup in
post_install— Ruby allowlist strips unrecognized keys + Python fallback incraft-installscript
Changed
- Budget config fallback chain:
.claude-plugin/config.json→package.json(claudeMd.budget) → default 150 - Updated sync command docs, conventions, spec, changelog, and CLAUDE.md troubleshooting
Files Changed
10 files changed (+111/-59)
Full Changelog: v2.13.0...v2.13.1
v2.13.0 — Documentation Gap-Fill
What's New
30 new documentation pages filling gaps across the entire docs site, plus CLAUDE.md v3 command refactoring.
Documentation Gap-Fill
- 5 beginner cookbook recipes (find-command, update-doc, code-checks, git-branch, view-architecture)
- 8 troubleshooting guides (timeouts, wrong output, CI fails, pre-commit, worktree, broken links, CLAUDE.md sync, mkdocs)
- 1 beginner tutorial — "Your First 10 Minutes with Craft"
- 13 individual command pages (arch, code, test, git, docs)
- 4 category decision guides with "When to Use What" tables
- 7 broken links fixed, documentation status 98% → 99%
CLAUDE.md v3 Command Refactoring
- Command consolidation:
scaffold→init,update/audit/fix→sync,editwith--globalflag - New utilities:
claude_md_optimizer.py(1,030 lines),claude_md_sync.py(1,465 lines) - Pointer templates with
->syntax for lean CLAUDE.md files - Budget enforcement via
scripts/claude-md-budget-check.sh - 71 new tests (51 unit + 9 integration + 11 audit)
Stats
- 106 commands · 21 skills · 8 agents
- 1,174 tests passing · 99% documentation coverage
- Site: https://data-wise.github.io/craft/
Full Changelog: v2.12.0...v2.13.0