2.7.1 (2026-03-18)
2.7.0 (2026-03-17)
2.6.0 (2026-03-17)
- add Samverk MCP routing + fix KG#152 scope + [MCP-TOOLS] checklist (#396) (62aeace)
- add samverk-dispatch skill for dispatcher work protocol (#398) (d2732ff)
2.5.0 (2026-03-17)
- add archive recovery workflow and Synapset conformance check (#379) (96b5956)
- add synapset skill with structured retrieval guide (#356) (1c3be95)
- autolearn: ingest #330 + update status.md (#332) (daade3c)
- autolearn: ingest #343 (SQLite PRAGMA + Gitea assign gotchas) (#347) (ac4c5f8)
- autolearn: ingest #348 + #349 (Ollama gotchas) (#351) (19d2915)
- autolearn: ingest #354 (trivy cleanup + Gitea merge API) (#355) (25c8e66)
- autolearn: ingest 15 issues as rules entries (#309-#328) (#329) (dab9c64)
- autolearn: sync batch-ingested entries to Synapset pool (#333) (#339) (5c784e3)
- hooks: add commit/push verification reminder hook (#369) (e05e2c7)
- hooks: add pre-commit and commit-msg git hook templates (#366) (919fe75)
- hooks: add SessionStop hook for session-end hygiene (#365) (16e7115)
- hooks: add SubagentStop verification hook (#368) (1b75c04)
- metrics: add conformance score persistence and autolearn velocity (#336) (#341) (18d6eb4)
- metrics: add effectiveness measurement skill and data model (#334) (#338) (5b99d1e)
- metrics: add visual HTML dashboard generator (#346) (3a058f7)
- metrics: add weekly PR metrics collection GH Action (#337) (#342) (1b51616)
- metrics: enhance dashboard with unified theme (#350) (#352) (ad80efa)
- metrics: session instrumentation and pattern tracking (#340) (c5981dd)
- metrics: track Synapset pattern applications across all projects (#345) (2554388)
- Synapset-backed compaction architecture (#376) (ae2948a)
- collect-pr-metrics.sh Python fallback for MSYS/Windows (#344) (39e32f2)
- docs: correct pattern counts and skill list drift (#353) (5888070)
- docs: correct pattern counts and status drift (#359) (b39018a)
- rules: correct KG#135 -- OAuth not required for Custom Connectors (#357) (113e5a2)
- sync release-please manifest to v2.4.0 (#382) (29520df)
Autonomous autolearn -- removes manual menu and runs learnings capture automatically.
- Autolearn autonomous routing (PR #322): Removed blocking 5-option intake menu from
/autolearn. Skill now auto-selects the appropriate workflow (quick-reflect, session-review, update-knowledge, skill-improvement, audit-rules) based on session context analysis. Supports explicit args:/autolearn review,/autolearn audit, etc. - Autolearn runs autonomously: CLAUDE.md updated from "suggest the user run /autolearn" to "run /autolearn autonomously at task completion boundaries." No user intervention required.
- Hook text updated: UserPromptSubmit hook instructs autonomous execution instead of passive suggestions. Both template and live settings updated.
- Stale doc references fixed:
docs/hook-expansion-research.mdupdated to reflect autonomous model
Rules compaction, Synapset integration, and MCP research.
- Rules compaction (PR #295): AP 42k->35k, KG 50k->30k -- archived 29 stale entries, consolidated 9 clusters, trimmed 27 entries
- Synapset integration (PR #313): Semantic vector memory backend for autolearn --
store_memory/search_memoryin dual-store workflows, [SYNAPSET] checklist block, pool strategy documentation - Tool selection guide (PR #300): New rules file
claude/rules/tool-selection-guide.md-- proactive decision tree for GitHub, web research, file ops, Go dev, shell scripting, and knowledge tools - MCP lazy-loading research (PR #305): Feasibility assessment for custom MCP gateway -- verdict: wait (ToolSearch provides ~85% context reduction). Feature request filed: anthropics/claude-code#34471
- Cross-reference verification (PR #311): Step 8 in
/rules-compactskill catches stale AP/KG cross-references after archiving - Autolearn entries (PRs #296, #308): 16 new patterns and gotchas from 11 issues -- worktree isolation (AP#127), go:embed cache (KG#125), Tailscale Funnel (KG#126), Cloudflare Email Routing (AP#126), sqlite-vec dimensions (KG#127), and more
- Release standardization templates (ADR-0015): release-please config, manifest, workflow, and git-cliff templates
- Rules reconciliation (PR #140): Imported 44 orphaned entries from local copies back to devkit
- Rules drift detection (PR #141):
devkit_drift_check()in SessionStart.sh - Project settings enforcement (PR #145): Three-layer mechanism for
.claude/settings.json - Gitea forge support (PR #285):
-Giteaswitch in new-project.ps1, forge-aware scaffolding
- Stale doc counts (PR #286): README and CLAUDE.md pattern counts updated
- credentials.ps1 StrictMode (PR #283): Removed Set-StrictMode from dot-sourced lib (KG#114)
- sync.ps1 StrictMode crash (PR #142):
PSObject.Properties.Match()for safe property checks - sync.ps1 Read-Host null (PR #144): Null guard for non-interactive mode
- Rules files: 10 -> 11 (added tool-selection-guide.md)
- Pattern counts: 170+ -> 150 (post-compaction: AP 74, KG 76)
- Autolearn workflows now dual-store to MCP Memory + Synapset
- 71 stale remote branches cleaned up
/devkit-sync verifychecks.claude/settings.jsonpresence
Rule lifecycle management for the autolearn system.
- Rule lifecycle metadata format (PR #126): ADR-0014 defines per-entry metadata (
**Added:**,**Source:**,**Status:**,**Last relevant:**,**See also:**), deprecation states, and archive strategy - Archive directory (PR #126):
claude/rules/archive/for deprecated entries (not loaded into sessions, frees context tokens) - Frontmatter extensions (PR #126):
entry_countandlast_updatedfields in Tier 2 rules file frontmatter - Metadata PoC (PR #127): 10 proof-of-concept entries annotated with lifecycle metadata (5 AP, 5 KG)
- Duplicate resolution (PR #127): AP#27 superseded by KG#17, archived; swagger cluster cross-referenced (5 entries)
- Rules audit workflow (PR #128):
/reflectoption 5 for health check -- parses entries, generates report, identifies stale/duplicate entries, proposes actions - Last-relevant tracking (PR #128):
/reflectquick and session workflows update**Last relevant:**timestamps on entries applied during the session
- Autolearn-patterns entry count: 76 -> 75 (AP#27 archived)
Governance and quality gates for the autolearn system.
- Tiered rule governance (PR #117):
core-principles.md(Tier 0, immutable) anderror-policy.md(Tier 1, governed) rules files with YAML frontmatter tier metadata - SessionStart health checks (PR #118): symlink integrity verification and CLAUDE.md detection at session start
- Pre-commit verification (PR #119): build/test/lint gates before commit in workflow preferences
- Fix-forward workflow (PR #120): error-policy.md with zero-tolerance fix-forward, replaces "pre-existing" classification
- Template quality gates (PR #121): CI scaffolding templates include lint and test verification steps
- Autolearn scope-aware routing (PR #122): in DevKit, write Tier 2 rules directly; in projects, write to MCP Memory and create DevKit issues for universal learnings
- Rule validation pipeline (PR #123): five-stage gate for proposed rules (dangerous pattern scan, core principles check, conflict check, risk classification, storage decision) in
references/validation-pipeline.md - Propagation verification (PR #124):
/devkit-sync verifychecks all active projects for DevKit update propagation via symlink health; SessionStart reports rule file changes after pull /devkit-sync promotesubcommand for graduating local patterns to universal rules/devkit-sync updatesubcommand for version checking and upgrading
- Autolearn workflows (
quick-reflect.md,session-review.md) now include validation and scope assessment steps update-knowledge.mdrequires DevKit context (context guard rejects non-DevKit sessions)- Rules file count increased from 8 to 10
v2.0 represents a major architectural shift from a bash-centric toolkit to a cross-platform, multi-tier system with formal versioning.
- Three-tier settings architecture (ADR-0012): User > DevSpace > Project cascade
- Dual-language scripting strategy (ADR-0013): PowerShell primary, bash legacy
- Cross-platform path resolution via
~/.devkit-config.json - Forge abstraction layer for GitHub/GitLab operations
- Project registry for multi-project coordination
- Local overrides via
.local.mdpattern (gitignored, machine-specific) - SessionStart hardening: rate limiting (1/hour), lock-file awareness, pull logging
- Auto-push prompt after
/reflectsessions - Version-tagged releases with
VERSIONfile as single source of truth - SessionStart version check: notifies when a newer DevKit release is available
devkit updatecommand via/devkit-syncskill (check version, upgrade to tag or latest)devkit promotecommand for graduating local patterns to universal rules
- MAJOR: Breaking changes to rules, skill interfaces, or sync protocol
- MINOR: New skills, agents, or non-breaking rule additions
- PATCH: Bug fixes, documentation, pattern additions
- claude/CLAUDE.md: replaced
YOUR_PLATFORMplaceholder with actual value and substitution note - METHODOLOGY.md: added Windows warning for BMAD/Spec Kit tools (see known-gotchas #42-44)
- known-gotchas.md: renumbered contiguously 1-46 (was non-contiguous with gaps at 8-9, 13, 40, 42, 44 and #47 out of order)
- Cross-references to gotcha numbers updated in autolearn-patterns.md and workflow-preferences.md
- AGENT-WORKFLOW-GUIDE.md: replaced Python pseudo-code agent examples with correct
.claude/agents/*.mdformat
- CI: README skill count accuracy check (fails if count doesn't match
claude/skills/directories) - CI: verify.sh skill list accuracy check (fails on missing directories, warns on unlisted skills)
profiles/directory for Kit 2 stack profilesproject-templates/directory for Kit 3 scaffoldingdocs/andmachine/directories for guides and snapshots- PowerShell stub files: setup.ps1, bootstrap.ps1, stack.ps1, new-project.ps1, verify.ps1, lib/*.ps1
setup/lib/ui.ps1: console output library (Write-Section/Step/OK/Warn/Fail, Write-VerifyTable, Read-Required/Confirm/Menu, Invoke-ManualChecklist)setup/lib/checks.ps1: prerequisite and tool detection library (Test-Tool, Test-HyperV, Test-WSL2, Test-Virtualization, Test-DeveloperMode, Test-WindowsVersion, Test-ClaudeAuth/Skill/MCP, Test-DockerRunning/WSLBackend, Test-Credential, Get-PreflightStatus)setup/lib/install.ps1: winget and manual install wrappers (Install-WingetPackage/Packages, Install-VSCodeExtension/Extensions, Invoke-ManualInstall, Export-WingetManifest, Export-VSCodeExtensions)setup/lib/credentials.ps1: Windows Credential Manager integration (Set/Get/Test/Remove-DevkitCredential, Invoke-CredentialCollection with validation and secure input)setup/setup.ps1: main menu entry point with -Kit parameter for direct dispatch, version display, quick status checkmachine/winget.json: curated winget export (25 dev packages, personal apps removed)machine/git-config.template: gitconfig template with YOUR_NAME/YOUR_EMAIL placeholdersmachine/manual-requirements.md: reference for non-automatable setup steps (Hyper-V, WSL2, Docker config, Dev Mode)setup/backup.ps1: refresh machine snapshot files from current state with diff summarysetup/bootstrap.ps1phases 1-2: pre-flight checks (Windows version, Hyper-V, WSL2, virtualization, Developer Mode) and core tool installs from machine/winget.json + vscode-extensions.txtsetup/bootstrap.ps1phases 3-4: git config from template, devspace directory setup with ~/.devkit-config.json, PowerShell profile alias, credentials collection via Windows Credential Managersetup/bootstrap.ps1phases 5-6: AI layer deploy (Claude Code npm install, skills/rules/agents/hooks with hash-based overwrite, CLAUDE.md placeholder substitution) and full verification table with next stepsdocs/BOOTSTRAP.md: step-by-step new machine setup guide with troubleshooting sectionsetup/lib/profiles.ps1: profile format parser with YAML frontmatter, dependency resolution, and cycle detectionproject-templates/concept-brief.md: project vision capture template for Kit 3 scaffoldingproject-templates/claude-md-template.md: fallback CLAUDE.md template for new projectsproject-templates/github-labels.json: standard label set (13 labels) for new GitHub reposprofiles/go-cli.md: base Go profile (winget, linters, VS Code extensions, cross-compilation)profiles/go-web.md: extended Go+Web profile (buf, gRPC, REST client, Docker deployment)profiles/iot-embedded.md: ESP32/ESPHome profile (uv-managed tools, OTA/serial flashing, sensor patterns)setup/stack.ps1: Kit 2 profile selection and installer (-List, -ShowProfile, -Install, -Force flags)setup/new-project.ps1steps 1-2: concept collection (interactive + file), project scaffolding (git, GitHub, directories, labels, workspace)setup/new-project.ps1steps 3-4: Claude-generated CLAUDE.md (with template fallback), Phase 0 issue creation, workspace open
- Bash setup scripts moved to
setup/legacy/with deprecation headers (PowerShell primary) - SKILLS-ECOSYSTEM.md: expanded Chat skill installation guide with rationale and sync process
devspace/CLAUDE.mdmoved toproject-templates/workspace-claude-md-template.md
- go-development SKILL.md: removed network_security_patterns section (SubNetree-specific content)
- coordination-sync, research-mode, dashboard, pm-view, dev-mode skills (moved to SubNetree project -- too project-specific for a general toolkit)
- settings.template.json: JSON syntax errors (missing commas on lines 17, 75)
- settings.template.json: removed hardcoded Windows drive path (
Read(//d/**)) - AUTOMATION-SETUP.md: fixed
Claude.md->CLAUDE.mdcapitalization (3 instances) - SessionStart.sh: removed stale
D:/DevSpace/.coordination/references - verify.sh: now checks all 9 remaining skills (was only checking 5 of 14)
- devspace/CLAUDE.md converted to template with generic placeholders
- memory-seeds.md converted to example with generic entity names
- claude/CLAUDE.md OS line made generic (was hardcoded to Windows MSYS_NT)
- settings.template.json: removed redundant granular permissions, trimmed plugin list to core set
- README.md: added Quick Start, What's Included section, updated skill inventory
- METHODOLOGY.md: updated tool references for removed skills
- SessionStart.sh: simplified to generic CLAUDE.md detection only
- GitHub Actions CI workflow: markdown lint + skill routing validation
- CHANGELOG.md
Initial release. 101 files: 5 rules (70+ patterns, 47+ gotchas), 14 skills, 6 agent templates, hooks, setup scripts, MCP config, project templates, and development methodology.