Separate cognitive work from mechanical API execution to reduce skill-related token consumption by ~76-83% during sprint workflows. Changes: - Add batch-execution.md skill with 4-phase protocol - Promote mcp-tools-reference and batch-execution to frontmatter for planner and orchestrator agents (auto-injected, zero re-read) - Replace "Skills to Load" with phase-based "Skill Loading Protocol" - Restructure planning-workflow.md Steps 8-10 for batch execution - Update agent matrix in CLAUDE.md and docs/CONFIGURATION.md - Add Phase-Based Skill Loading documentation section - Clean up .gitignore (transient files, dev symlinks) Token impact: - 6-issue sprint planning: ~76% reduction - 10-issue sprint planning: ~80% reduction - 8-issue status updates: ~83% reduction Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
46 KiB
Changelog
All notable changes to the Leo Claude Marketplace will be documented in this file.
The format is based on Keep a Changelog.
[Unreleased]
[6.0.0] - YYYY-MM-DD
Added
Plan-Then-Batch Skill Optimization (projman)
New execution pattern that separates cognitive work from mechanical API operations, reducing skill-related token consumption by ~76-83% during sprint workflows.
-
skills/batch-execution.md— New skill defining the plan-then-batch protocol:- Phase 1: Cognitive work with all skills loaded
- Phase 2: Execution manifest (structured plan of all API operations)
- Phase 3: Batch execute API calls using only frontmatter skills
- Phase 4: Batch report with success/failure summary
- Error handling: continue on individual failures, report at end
-
Frontmatter skill promotion:
- Planner agent:
mcp-tools-referenceandbatch-executionpromoted to frontmatter (auto-injected, zero re-read cost) - Orchestrator agent: same promotion
- Eliminates per-operation skill file re-reads during API execution loops
- Planner agent:
-
Phase-based skill loading:
- Planner: 3 phases (validation → analysis → approval) with explicit "read once" instructions
- Orchestrator: 2 phases (startup → dispatch) with same pattern
- New
## Skill Loading Protocolsection replaces flat## Skills to Loadin agent files
Changed
-
planning-workflow.md— Steps 8-10 restructured:- Step 8: "Draft Issue Specifications" (no API calls — resolve all parameters first)
- Step 8a: "Batch Execute Issue Creation" (tight API loop, frontmatter skills only)
- Step 9: Merged into Step 8a (dependencies created in batch)
- Step 10: Milestone creation moved before batch (must exist for assignment)
-
Agent matrix updated:
- Planner:
body text (14)→frontmatter (2) + body text (12) - Orchestrator:
body text (12)→frontmatter (2) + body text (10)
- Planner:
-
docs/CONFIGURATION.md— New "Phase-Based Skill Loading" subsection documenting the pattern
Token Impact
| Scenario | Before | After | Savings |
|---|---|---|---|
| 6-issue sprint (planning) | ~23,800 lines | ~5,600 lines | ~76% |
| 10-issue sprint (planning) | ~35,000 lines | ~7,000 lines | ~80% |
| 8-issue status updates (orchestrator) | ~9,600 lines | ~1,600 lines | ~83% |
[5.10.0] - 2026-02-03
Added
NetBox MCP Server: Module-Based Tool Filtering
Environment-variable-driven module filtering to reduce token consumption:
- New config option:
NETBOX_ENABLED_MODULESin~/.config/claude/netbox.env - Token savings: ~15,000 tokens (from ~19,810 to ~4,500) with recommended config
- Default behavior: All modules enabled if env var unset (backward compatible)
- Startup logging: Shows enabled modules and tool count on initialization
- Routing guard: Clear error message when calling disabled module's tools
Recommended configuration for cmdb-assistant users:
NETBOX_ENABLED_MODULES=dcim,ipam,virtualization,extras
This enables ~43 tools covering all cmdb-assistant commands while staying well below the 25K token warning threshold.
Fixed
cmdb-assistant Documentation: Incorrect Tool Names
Fixed documentation referencing non-existent virtualization_* tool names:
| File | Wrong | Correct |
|---|---|---|
claude-md-integration.md |
virtualization_list_virtual_machines |
virt_list_vms |
claude-md-integration.md |
virtualization_create_virtual_machine |
virt_create_vm |
cmdb-search.md |
virtualization_list_virtual_machines |
virt_list_vms |
Also fixed NetBox README.md tool name references for virtualization, wireless, and circuits modules.
Gitea MCP Server: Standardized Build Backend
Changed mcp-servers/gitea/pyproject.toml from hatchling to setuptools:
- Matches all other MCP servers (contract-validator, viz-platform, data-platform)
- Updated license format to PEP 639 compliance
- Added pytest configuration for consistency
[5.9.0] - 2026-02-03
Added
Plugin Installation Scripts
New scripts for installing marketplace plugins into consumer projects:
-
scripts/install-plugin.sh— Install a plugin to a consumer project- Adds MCP server entry to target's
.mcp.json(if plugin has MCP server) - Appends integration snippet to target's
CLAUDE.md - Idempotent: safe to run multiple times
- Validates plugin exists and target path is valid
- Adds MCP server entry to target's
-
scripts/uninstall-plugin.sh— Remove a plugin from a consumer project- Removes MCP server entry from
.mcp.json - Removes integration section from
CLAUDE.md
- Removes MCP server entry from
-
scripts/list-installed.sh— Show installed plugins in a project- Lists fully installed, partially installed, and available plugins
- Shows plugin versions and descriptions
Usage:
./scripts/install-plugin.sh data-platform ~/projects/personal-portfolio
./scripts/list-installed.sh ~/projects/personal-portfolio
./scripts/uninstall-plugin.sh data-platform ~/projects/personal-portfolio
Documentation: docs/CONFIGURATION.md updated with "Installing Plugins to Consumer Projects" section.
Fixed
Plugin Installation Scripts — MCP Mapping & Section Markers
MCP Server Mapping:
- Added
mcp_serversfield to plugin.json for plugins that use shared MCP servers projmanandpr-reviewnow correctly installgiteaMCP servercmdb-assistantnow correctly installsnetboxMCP server- Scripts read MCP server names from plugin.json instead of assuming plugin name = server name
CLAUDE.md Section Markers:
- Install script now wraps integration content with HTML comment markers:
<!-- BEGIN marketplace-plugin: {name} -->and<!-- END marketplace-plugin: {name} --> - Uninstall script uses markers for precise section removal (no more code block false positives)
- Backward compatible: falls back to legacy header detection for pre-marker installations
Plugins updated with mcp_servers field:
projman→["gitea"]pr-review→["gitea"]cmdb-assistant→["netbox"]data-platform→["data-platform"]viz-platform→["viz-platform"]contract-validator→["contract-validator"]
Agent Model Selection
Per-agent model selection using Claude Code's now-supported model frontmatter field.
- All 25 marketplace agents assigned appropriate model (
sonnet,haiku, orinherit) - Model assignment based on reasoning depth, tool complexity, and latency requirements
- Documentation added to
CLAUDE.mdanddocs/CONFIGURATION.md
Supported values: sonnet (default), opus, haiku, inherit
Model assignments:
| Model | Agent Types |
|---|---|
| sonnet | Planner, Orchestrator, Executor, Code Reviewer, Coordinator, Security Reviewers, Data Advisor, Design Reviewer, etc. |
| haiku | Maintainability Auditor, Test Validator, Component Check, Theme Setup, Git Assistant, Data Ingestion, Agent Check |
Fixed
Agent Frontmatter Standardization
- Fixed viz-platform and data-platform agents using non-standard
agent:field (nowname:) - Removed non-standard
triggers:field from domain agents (trigger info already in agent body) - Added missing frontmatter to 13 agents across pr-review, viz-platform, contract-validator, clarity-assist, git-flow, doc-guardian, code-sentinel, cmdb-assistant, and data-platform
- All 25 agents now have consistent
name,description, andmodelfields
Changed
Agent Frontmatter Hardening v3
Comprehensive agent-level configuration using Claude Code's supported frontmatter fields.
permissionMode added to all 25 agents:
bypassPermissions(1): Executor — full autonomy with code-sentinel + Code Reviewer safety netsacceptEdits(7): Orchestrator, Data Ingestion, Theme Setup, Refactor Advisor, Doc Analyzer, Git Assistant, Maintainerdefault(7): Planner, Code Reviewer, Data Advisor, Layout Builder, Full Validation, Clarity Coach, CMDB Assistantplan(10): All pr-review agents (5), Data Analysis, Design Reviewer, Component Check, Agent Check, Security Reviewer (code-sentinel)
disallowedTools added to 12 agents:
- All
plan-mode agents (10) + Code Reviewer + Clarity Coach receivedisallowedTools: Write, Edit, MultiEdit - Enforces read-only contracts at platform level (defense-in-depth with
permissionMode)
Model promotions:
- Planner:
sonnet→opus(architectural reasoning benefits from deeper analysis) - Code Reviewer:
sonnet→opus(quality gate benefits from thorough review)
skills frontmatter on 3 agents:
- Executor: 7 safety-critical skills auto-injected (branch-security, runaway-detection, etc.)
- Code Reviewer: 4 review skills auto-injected
- Maintainer: 2 config skills auto-injected
- Body text
## Skills to Loadremoved for these agents to avoid duplication
Documentation:
CLAUDE.mdanddocs/CONFIGURATION.mdupdated with complete agent configuration matrix- New subsections: permissionMode Guide, disallowedTools Guide, skills Frontmatter Guide
[5.8.0] - 2026-02-02
Added
claude-config-maintainer v1.2.0 - Settings Audit Feature
New commands for auditing and optimizing settings.local.json permission configurations:
/config-audit-settings— Auditsettings.local.jsonpermissions with 100-point scoring across redundancy, coverage, safety alignment, and profile fit/config-optimize-settings— Apply permission optimizations with dry-run, named profiles (conservative,reviewed,autonomous), and consolidation modes/config-permissions-map— Generate Mermaid diagram of review layer coverage and permission gapsskills/settings-optimization.md— Comprehensive skill for permission pattern analysis, consolidation rules, review-layer-aware recommendations, and named profiles
Key Features:
- Settings Efficiency Score (100 points) alongside existing CLAUDE.md score
- Review layer verification — agent reads
hooks/hooks.jsonfrom installed plugins before recommending auto-allow patterns - Three named profiles:
conservative(prompts for most writes),reviewed(for projects with ≥2 review layers),autonomous(sandboxed environments) - Pattern consolidation detection: duplicates, subsets, merge candidates, stale entries, conflicts
Projman Hardening Sprint
Targeted improvements to safety gates, command structure, lifecycle tracking, and cross-plugin contracts.
Sprint Lifecycle State Machine:
- New
skills/sprint-lifecycle.md- defines valid states and transitions via milestone metadata - States: idle -> Sprint/Planning -> Sprint/Executing -> Sprint/Reviewing -> idle
- All sprint commands check and set lifecycle state on entry/exit
- Out-of-order calls produce warnings with guidance,
--forceoverride available
Sprint Dispatch Log:
- Orchestrator now maintains a structured dispatch log during execution
- Records task dispatch, completion, failure, gate checks, and resume events
- Enables timeline reconstruction after interrupted sessions
Gate Contract Versioning:
- Gate commands (
/design-gate,/data-gate) declaregate_contract: v1in frontmatter domain-consultation.mdGate Command Reference includes expected contract versionvalidate_workflow_integrationnow checks contract version compatibility- Mismatch produces WARNING, missing contract produces INFO suggestion
Shared Visual Output Skill:
- New
skills/visual-output.md- single source of truth for projman visual headers - All 4 agent files reference the skill instead of inline templates
- Phase Registry maps agents to emoji and phase names
Changed
Sprint Approval Gate Hardened:
- Approval is now a hard block, not a warning (was "recommended", now required)
--forceflag added to bypass in emergencies (logged to milestone)- Consistent language across sprint-approval.md, sprint-start.md, and orchestrator.md
RFC Commands Normalized:
- 5 individual commands (
/rfc-create,/rfc-list,/rfc-review,/rfc-approve,/rfc-reject) consolidated into/rfc create|list|review|approve|reject /clear-cacheabsorbed into/setup --clear-cache- Command count reduced from 17 to 12
/test Command Documentation Expanded:
- Sprint integration section (pre-close verification workflow)
- Concrete usage examples for all modes
- Edge cases table
- DO NOT rules for both modes
Removed
plugins/projman/commands/rfc-create.md(replaced by/rfc create)plugins/projman/commands/rfc-list.md(replaced by/rfc list)plugins/projman/commands/rfc-review.md(replaced by/rfc review)plugins/projman/commands/rfc-approve.md(replaced by/rfc approve)plugins/projman/commands/rfc-reject.md(replaced by/rfc reject)plugins/projman/commands/clear-cache.md(replaced by/setup --clear-cache)
[5.7.1] - 2026-02-02
Added
- contract-validator: New
validate_workflow_integrationMCP tool — validates domain plugins expose required advisory interfaces (gate command, review command, advisory agent) - contract-validator: New
MISSING_INTEGRATIONissue type for workflow integration validation
Fixed
scripts/setup.shbanner version updated from v5.1.0 to v5.7.1
Reverted
- marketplace.json: Removed
integrates_withfield — Claude Code schema does not support custom plugin fields (causes marketplace load failure)
[5.7.0] - 2026-02-02
Added
- data-platform: New
data-advisoragent for data integrity, schema, and dbt compliance validation - data-platform: New
data-integrity-audit.mdskill defining audit rules, severity levels, and scanning strategies - data-platform: New
/data-gatecommand for binary pass/fail data integrity gates (projman integration) - data-platform: New
/data-reviewcommand for comprehensive data integrity audits
Changed
- Domain Advisory Pattern now fully operational for both Viz and Data domains
- projman orchestrator
Domain/Datagates now resolve to live/data-gatecommand (previously fell through to "gate unavailable" warning)
[5.6.0] - 2026-02-01
Added
- Domain Advisory Pattern: Cross-plugin integration enabling projman to consult domain-specific plugins during sprint lifecycle
- projman: New
domain-consultation.mdskill for domain detection and gate protocols - viz-platform: New
design-revieweragent for design system compliance auditing - viz-platform: New
design-system-audit.mdskill defining audit rules and severity levels - viz-platform: New
/design-reviewcommand for detailed design system audits - viz-platform: New
/design-gatecommand for binary pass/fail validation gates - Labels: New
Domain/VizandDomain/Datalabels for domain routing
Changed
- projman planner: Now loads domain-consultation skill and performs domain detection during planning
- projman orchestrator: Now runs domain gates before marking Domain/* labeled issues as complete
[5.5.0] - 2026-02-01
Added
RFC System for Feature Tracking
Wiki-based Request for Comments (RFC) system for capturing, reviewing, and tracking feature ideas through their lifecycle.
New Commands (projman):
/rfc-create- Create new RFC from conversation or clarified specification/rfc-list- List all RFCs grouped by status (Draft, Review, Approved, Implementing, Implemented, Rejected, Stale)/rfc-review- Submit Draft RFC for maintainer review/rfc-approve- Approve RFC, making it available for sprint planning/rfc-reject- Reject RFC with documented reason
RFC Lifecycle:
- Draft → Review → Approved → Implementing → Implemented
- Terminal states: Rejected, Superseded
- Stale: Drafts with no activity >90 days
Sprint Integration:
/sprint-plannow detects approved RFCs and offers selection/sprint-closeupdates RFC status to Implemented on completion- RFC-Index wiki page auto-maintained with status sections
Clarity-Assist Integration:
- Vagueness hook now detects feature request patterns
- Suggests
/rfc-createfor feature ideas /clarifyoffers RFC creation after delivering clarified spec
New MCP Tool:
allocate_rfc_number- Allocates next sequential RFC number
New Skills:
skills/rfc-workflow.md- RFC lifecycle and state transitionsskills/rfc-templates.md- RFC page template specifications
Changed
Sprint 8: Hook Efficiency Quick Wins
Performance optimizations for plugin hooks to reduce overhead on every command.
Changes:
- viz-platform: Remove SessionStart hook that only echoed "loaded" (zero value)
- git-flow: Add early exit to
branch-check.shfor non-git commands (skip JSON parsing) - git-flow: Add early exit to
commit-msg-check.shfor non-git commands (skip Python spawn) - project-hygiene: Add 60-second cooldown to
cleanup.sh(reduce find operations)
Impact: Hooks now exit immediately for 90%+ of Bash commands that don't need processing.
Issues: #321, #322, #323, #324 PR: #334
[5.4.1] - 2026-01-30
Removed
Multi-Model Agent Support (REVERTED)
Reason: Claude Code does not support defaultModel in plugin.json or model in agent frontmatter. The schema validation rejects these as "Unrecognized key".
Removed:
defaultModelfield from all plugin.json files (6 plugins)modelfield references from agent frontmatterdocs/MODEL-RECOMMENDATIONS.md- Deleted entirely- Model configuration sections from
docs/CONFIGURATION.mdandCLAUDE.md
Lesson: Do not implement features without verifying they are supported by Claude Code's plugin schema.
[5.4.0] - 2026-01-28 [REVERTED]
Added (NOW REMOVED - See 5.4.1)
Sprint 7: Multi-Model Agent Support
Configurable model selection for agents with inheritance chain.
This feature was reverted in 5.4.1 - Claude Code does not support these fields.
Original sprint work:
- Issues: #302, #303, #304, #305, #306
- PRs: #307, #308
[5.3.0] - 2026-01-28
Added
Sprint 6: Visual Branding Overhaul
Consistent visual headers and progress tracking across all plugins.
Visual Output Headers (109 files):
- Projman: Double-line headers (╔═╗) with phase indicators (🎯 PLANNING, ⚡ EXECUTION, 🏁 CLOSING)
- Other Plugins: Single-line headers (┌─┐) with plugin icons
- All 23 agents updated with Visual Output Requirements section
- All 86 commands updated with Visual Output section and header templates
Plugin Icon Registry:
| Plugin | Icon |
|---|---|
| projman | 📋 |
| code-sentinel | 🔒 |
| doc-guardian | 📝 |
| pr-review | 🔍 |
| clarity-assist | 💬 |
| git-flow | 🔀 |
| cmdb-assistant | 🖥️ |
| data-platform | 📊 |
| viz-platform | 🎨 |
| contract-validator | ✅ |
| claude-config-maintainer | ⚙️ |
Wiki Branding Specification (4 pages):
- branding/visual-spec - Central specification
- branding/plugin-registry - Icons and styles
- branding/header-templates - Copy-paste templates
- branding/progress-templates - Sprint progress blocks
Fixed
- Docs: Version sync - CLAUDE.md, marketplace.json, README.md now consistent
- Docs: Added 18 missing commands from Sprint 4 & 5 to README.md and COMMANDS-CHEATSHEET.md
- MCP: Registered
/sprint-diagramas invokable skill
Sprint Completed:
- Milestone: Sprint 6 - Visual Branding Overhaul (closed 2026-01-28)
- Issues: #272, #273, #274, #275, #276, #277, #278
- PRs: #284, #285
- Wiki: Sprint 6 Lessons
[5.2.0] - 2026-01-28
Added
Sprint 5: Documentation (V5.2.0 Plugin Enhancements)
Documentation and guides for the plugin enhancements initiative.
git-flow v1.2.0:
- Branching Strategy Guide (
docs/BRANCHING-STRATEGY.md) - Complete documentation ofdevelopment -> staging -> mainpromotion flow with Mermaid diagrams
clarity-assist v1.2.0:
- ND Support Guide (
docs/ND-SUPPORT.md) - Documentation of neurodivergent accommodations, features, and usage examples
Gitea MCP Server:
update_issuemilestone parameter - Can now assign/change milestones programmatically
Sprint Completed:
- Milestone: Sprint 5 - Documentation (closed 2026-01-28)
- Issues: #266, #267, #268, #269
- Wiki: Change V5.2.0: Plugin Enhancements (Sprint 5 Documentation)
Sprint 4: Commands (V5.2.0 Plugin Enhancements)
Implementation of 18 new user-facing commands across 8 plugins.
projman v3.3.0:
/sprint-diagram- Generate Mermaid diagram of sprint issues with dependencies and status
pr-review v1.1.0:
/pr-diff- Formatted diff with inline review comments and annotations- Confidence threshold config -
PR_REVIEW_CONFIDENCE_THRESHOLDenv var (default: 0.7)
data-platform v1.2.0:
/data-quality- DataFrame quality checks (nulls, duplicates, types, outliers) with pass/warn/fail scoring/lineage-viz- dbt lineage visualization as Mermaid diagrams/dbt-test- Formatted dbt test runner with summary and failure details
viz-platform v1.1.0:
/chart-export- Export charts to PNG, SVG, PDF via kaleido/accessibility-check- Color blind validation (WCAG contrast ratios)/breakpoints- Responsive layout breakpoint configuration- New MCP tools:
chart_export,accessibility_validate_colors,accessibility_validate_theme,accessibility_suggest_alternative,layout_set_breakpoints - New dependency: kaleido>=0.2.1 for chart rendering
contract-validator v1.2.0:
/dependency-graph- Mermaid visualization of plugin dependencies with data flow
doc-guardian v1.1.0:
/changelog-gen- Generate changelog from conventional commits/doc-coverage- Documentation coverage metrics by function/class/stale-docs- Flag documentation behind code changes
claude-config-maintainer v1.1.0:
/config-diff- Track CLAUDE.md changes over time with behavioral impact analysis/config-lint- 31 lint rules for CLAUDE.md (security, structure, content, format, best practices)
cmdb-assistant v1.2.0:
/cmdb-topology- Infrastructure topology diagrams (rack, network, site views)/change-audit- NetBox audit trail queries with filtering/ip-conflicts- Detect IP conflicts and overlapping prefixes
Sprint Completed:
- Milestone: Sprint 4 - Commands (closed 2026-01-28)
- Issues: #241-#258 (18/18 closed)
- Wiki: Change V5.2.0: Plugin Enhancements (Sprint 4 Commands)
- Lessons: Sprint 4 - Plugin Commands Implementation
Fixed
- MCP: Project directory detection - all run.sh scripts now capture
CLAUDE_PROJECT_DIRfrom PWD before changing directories - Docs: Added Gitea auto-close behavior and MCP session restart notes to DEBUGGING-CHECKLIST.md
Sprint 3: Hooks (V5.2.0 Plugin Enhancements)
Implementation of 6 foundational hooks across 4 plugins.
git-flow v1.1.0:
- Commit message enforcement hook - PreToolUse hook validates conventional commit format on all
git commitcommands (not just/commit). Blocks invalid commits with format guidance. - Branch name validation hook - PreToolUse hook validates branch naming on
git checkout -bandgit switch -c. Enforcestype/descriptionformat, lowercase, max 50 chars.
clarity-assist v1.1.0:
- Vagueness detection hook - UserPromptSubmit hook detects vague prompts and suggests
/clarifywhen ambiguity, missing context, or unclear scope detected.
data-platform v1.1.0:
- Schema diff detection hook - PostToolUse hook monitors edits to schema files (dbt models, SQL migrations). Warns on breaking changes (column removal, type narrowing, constraint addition).
contract-validator v1.1.0:
- SessionStart auto-validate hook - Smart validation that only runs when plugin files changed since last check. Detects interface compatibility issues at session start.
- Breaking change detection hook - PostToolUse hook monitors plugin interface files (README.md, plugin.json). Warns when changes would break consumers.
Sprint Completed:
- Milestone: Sprint 3 - Hooks (closed 2026-01-28)
- Issues: #225, #226, #227, #228, #229, #230
- Wiki: Change V5.2.0: Plugin Enhancements Proposal
- Lessons: Background agent permissions, agent runaway detection, MCP branch detection bug
Known Issues
- MCP Bug #231: Branch detection in Gitea MCP runs from installed plugin directory, not user's project directory. Workaround: close issues via Gitea web UI.
Gitea MCP Server - create_pull_request Tool
create_pull_request: Create new pull requests via MCP- Parameters: title, body, head (source branch), base (target branch), labels
- Branch-aware security: only allowed on development/feature branches
- Completes the PR lifecycle (was previously missing - only had list/get/review/comment)
cmdb-assistant v1.1.0 - Data Quality Validation
- SessionStart Hook: Tests NetBox API connectivity at session start
- Warns if VMs exist without site assignment
- Warns if devices exist without platform
- Non-blocking: displays warning, doesn't prevent work
- PreToolUse Hook: Validates input parameters before VM/device operations
- Warns about missing site, tenant, platform
- Non-blocking: suggests best practices without blocking
/cmdb-auditCommand: Comprehensive data quality analysis- Scopes: all, vms, devices, naming, roles
- Identifies Critical/High/Medium/Low issues
- Provides prioritized remediation recommendations
/cmdb-registerCommand: Register current machine into NetBox- Discovers system info: hostname, platform, hardware, network interfaces
- Discovers running apps: Docker containers, systemd services
- Creates device with interfaces, IPs, and sets primary IP
- Creates cluster and VMs for Docker containers
/cmdb-syncCommand: Sync machine state with NetBox- Compares current state with NetBox record
- Shows diff of changes (interfaces, IPs, containers)
- Updates with user confirmation
- Supports --full and --dry-run flags
- NetBox Best Practices Skill: Reference documentation
- Dependency order for object creation
- Naming conventions (
{role}-{site}-{number},{env}-{app}-{number}) - Role consolidation guidance
- Site/tenant/platform assignment requirements
- Agent Enhancement: Updated cmdb-assistant agent with validation requirements
- Proactive suggestions for missing fields
- Naming convention checks
- Dependency order enforcement
- Duplicate prevention
[5.0.0] - 2026-01-26
Added
Sprint 1: viz-platform Plugin ✅ Completed
- viz-platform v1.0.0 - Visualization tools with Dash Mantine Components validation and theming
- DMC Tools (3 tools):
list_components,get_component_props,validate_component- Version-locked component registry prevents Claude from hallucinating invalid props
- Static JSON registry approach for fast, deterministic validation
- Chart Tools (2 tools):
chart_create,chart_configure_interaction- Plotly-based visualization with theme token support
- Layout Tools (5 tools):
layout_create,layout_add_filter,layout_set_grid,layout_get,layout_add_section- Dashboard composition with responsive grid system
- Theme Tools (6 tools):
theme_create,theme_extend,theme_validate,theme_export_css,theme_list,theme_activate- Design token-based theming system
- Dual storage: user-level (
~/.config/claude/themes/) and project-level
- Page Tools (5 tools):
page_create,page_add_navbar,page_set_auth,page_list,page_get_app_config- Multi-page Dash app structure generation
- Commands:
/chart,/dashboard,/theme,/theme-new,/theme-css,/component,/initial-setup - Agents:
theme-setup,layout-builder,component-check - SessionStart Hook: DMC version check (non-blocking)
- Tests: 94 tests passing
- config.py: 82% coverage
- component_registry.py: 92% coverage
- dmc_tools.py: 88% coverage
- chart_tools.py: 68% coverage
- theme_tools.py: 99% coverage
- DMC Tools (3 tools):
Sprint Completed:
- Milestone: Sprint 1 - viz-platform Plugin (closed 2026-01-26)
- Issues: #170-#182 (13/13 closed)
- Wiki: Sprint-1-viz-platform-Implementation-Plan
- Lessons: sprint-1---viz-platform-plugin-implementation
- Reference:
docs/changes/CHANGE_V04_0_0_PROPOSAL_ORIGINAL.md(Phases 4-5)
[4.1.0] - 2026-01-26
Added
- projman: Wiki-based planning workflow enhancement (V04.1.0)
- Flexible input source detection in
/sprint-plan(file, wiki, or conversation) - Wiki proposal and implementation page creation during sprint planning
- Wiki reference linking in created issues
- Wiki status updates in
/sprint-close(Implemented/Partial/Failed) - Metadata section in lessons learned with implementation link for traceability
- New
/proposal-statuscommand for viewing proposal/implementation tree
- Flexible input source detection in
- projman:
/suggest-versioncommand - Analyzes CHANGELOG and recommends semantic version bump - projman: SessionStart hook now suggests sprint planning when open issues exist without milestone
- projman: SessionStart hook now warns about unreleased CHANGELOG entries
Changed
- doc-guardian: Hook now tracks documentation dependencies and queues specific files needing updates
- Outputs which specific docs need updating (e.g., "commands changed → update needed: docs/COMMANDS-CHEATSHEET.md README.md")
- Maintains queue file (
.doc-guardian-queue) for batch processing
- docs: COMMANDS-CHEATSHEET.md updated with data-platform plugin (7 commands + hook)
Fixed
- Documentation drift: COMMANDS-CHEATSHEET.md was missing data-platform plugin added in v4.0.0
- Proactive sprint planning: projman now suggests
/sprint-planat session start when unplanned issues exist
Known Issues
- MCP Bug #160:
update_wiki_pagetool renames pages to "unnamed" when page_name contains URL-encoded characters (:→%3A). Workaround: usecreate_wiki_pageto overwrite instead.
[4.0.0] - 2026-01-25
Added
New Plugin: data-platform v1.0.0
-
pandas MCP Tools (14 tools): DataFrame operations with Arrow IPC data_ref persistence
read_csv,read_parquet,read_json- Load data with chunking supportto_csv,to_parquet- Export to various formatsdescribe,head,tail- Data explorationfilter,select,groupby,join- Data transformationlist_data,drop_data- Memory management
-
PostgreSQL MCP Tools (10 tools): Database operations with asyncpg connection pooling
pg_connect,pg_query,pg_execute- Core database operationspg_tables,pg_columns,pg_schemas- Schema explorationst_tables,st_geometry_type,st_srid,st_extent- PostGIS spatial support
-
dbt MCP Tools (8 tools): Build tool wrapper with pre-execution validation
dbt_parse- Pre-flight validation (catches dbt 1.9+ deprecations)dbt_run,dbt_test,dbt_build- Execution with auto-validationdbt_compile,dbt_ls,dbt_docs_generate,dbt_lineage- Analysis tools
-
Commands:
/ingest,/profile,/schema,/explain,/lineage,/run -
Agents:
data-ingestion(loading/transformation),data-analysis(exploration/profiling) -
SessionStart Hook: Graceful PostgreSQL connection check (non-blocking warning)
-
Key Features:
- data_ref system for DataFrame persistence across tool calls
- 100k row limit with chunking support for large datasets
- Hybrid configuration (system:
~/.config/claude/postgres.env, project:.env) - Auto-detection of dbt projects
- Arrow IPC format for efficient memory management
[3.2.0] - 2026-01-24
Added
- git-flow:
/commitnow detects protected branches before committing- Warns when on protected branch (main, master, development, staging, production)
- Offers to create feature branch automatically instead of committing directly
- Configurable via
GIT_PROTECTED_BRANCHESenvironment variable
- netbox: Platform and primary_ip parameters added to device update tools
- claude-config-maintainer: Auto-enforce mandatory behavior rules via SessionStart hook
- scripts:
release.sh- Versioning workflow script for consistent releases - scripts:
verify-hooks.sh- Verify all hooks are command type
Changed
- doc-guardian: Hook switched from
prompttype tocommandtype- Prompt hooks unreliable - Claude ignores explicit instructions
- New
notify.shbash script guarantees exact output behavior - Only notifies for config file changes (commands/, agents/, skills/, hooks/)
- Silent exit for all other files - no blocking possible
- All hooks: Converted to command type with stricter plugin prefix enforcement
- All hooks now mandate
[plugin-name]prefix with "NO EXCEPTIONS" rule - Simplified output formats with word limits
- Consistent structure across projman, pr-review, code-sentinel, doc-guardian
- All hooks now mandate
- CLAUDE.md: Replaced destructive "ALWAYS CLEAR CACHE" rule with "VERIFY AND RESTART"
- Cache clearing mid-session breaks MCP tools
- Added guidance for proper plugin development workflow
Fixed
- cmdb-assistant: Complete MCP tool schemas for update operations (#138)
- netbox: Shorten tool names to meet 64-char API limit (#134)
- cmdb-assistant: Correct NetBox API URL format in setup wizard (#132)
- gitea/projman: Type safety for
create_label_smart, curl-based debug-report (#124) - netbox: Add diagnostic logging for JSON parse errors (#121)
- labels: Add duplicate check before creating labels (#116)
- hooks: Convert ALL hooks to command type with proper prefixes (#114)
- Protected branch workflow: Claude no longer commits directly to protected branches (fixes #109)
- doc-guardian hook no longer blocks workflow (fixes #110)
[3.1.1] - 2026-01-22
Added
- git-flow:
/commit-syncnow prunes stale remote-tracking branches withgit fetch --prune - git-flow:
/commit-syncdetects and reports local branches with deleted upstreams - git-flow:
/branch-cleanupnow handles stale branches (upstream gone) separately from merged branches - git-flow: New
GIT_CLEANUP_STALEenvironment variable for stale branch cleanup control
Changed
- All hooks: Added
[plugin-name]prefix to all hook messages for better identification[projman],[pr-review],[code-sentinel],[doc-guardian]prefixes
- doc-guardian: Hook now notification-only (no file reads or blocking operations)
- Suggests running
/doc-syncinstead of performing inline checks - Significantly reduces workflow interruption
- Suggests running
Fixed
- doc-guardian hook no longer stalls workflow with deep file analysis
[3.1.0] - 2026-01-21
Added
Debug Workflow Commands (projman)
-
/debug-report- Run diagnostics in test projects, create structured issues in marketplace- Runs 5 diagnostic MCP tool tests with explicit repo parameter
- Captures full project context (git remote, cwd, branch)
- Generates structured issue with hypothesis and investigation steps
- Creates issue in configured marketplace repository automatically
-
/debug-review- Investigate diagnostic issues with human approval gates- Lists open diagnostic issues for triage
- Maps errors to relevant code files using error-to-file mapping
- MANDATORY: Reads relevant files before proposing any fix
- Three approval gates: investigation summary, fix approach, PR creation
- Creates feature branch, commits, and PR with proper linking
MCP Server Improvements
- Dynamic label format detection in
suggest_labels- Supports slash format (
Type/Bug) and colon-space format (Type: Bug) - Fetches actual labels from repo and matches suggestions to real format
- Handles Effort/Efforts singular/plural normalization
- Supports slash format (
Changed
/labels-synccompletely rewritten with explicit execution steps- Step 1 now explicitly requires running
git remote get-url originvia Bash - All MCP tool calls show required
repoparameter - Added "DO NOT" section preventing common mistakes
- Removed confusing "Label Reference" section that caused file creation prompts
- Step 1 now explicitly requires running
Fixed
- MCP tools no longer fail with "Use 'owner/repo' format" error
- Root cause: MCP server is sandboxed and cannot auto-detect project directory
- Solution: Command documentation now instructs Claude to detect repo via Bash first
[3.0.1] - 2026-01-21
Added
/project-initcommand for quick project setup when system is already configured/project-synccommand to sync .env with git remote after repository move/rename- SessionStart hooks for automatic mismatch detection between git remote and .env
- Interactive setup wizard (
/initial-setup) redesigned to use Claude tools instead of bash script
Changed
GITEA_ORGmoved from system-level to project-level configuration (different projects may belong to different organizations)- Environment variables renamed to match MCP server expectations:
GITEA_URL→GITEA_API_URL(must include/api/v1)GITEA_TOKEN→GITEA_API_TOKENNETBOX_URL→NETBOX_API_URL(must include/api)NETBOX_TOKEN→NETBOX_API_TOKEN
- Setup commands now validate repository via Gitea API before saving configuration
- README.md simplified to show only wizard setup path (manual setup moved to CONFIGURATION.md)
Fixed
- API URL paths in curl commands (removed redundant
/api/v1since it's now in the URL variable) - Documentation now correctly references environment variable names
[3.0.0] - 2026-01-20
Added
New Plugins
-
clarity-assist v1.0.0 - Prompt optimization with ND accommodations
/clarifycommand for full 4-D methodology optimization/quick-clarifycommand for rapid single-pass clarification- clarity-coach agent with ND-friendly questioning patterns
- prompt-patterns skill with optimization rules
-
git-flow v1.0.0 - Git workflow automation
/commitcommand with smart conventional commit messages/commit-push,/commit-merge,/commit-syncworkflow commands/branch-start,/branch-cleanupbranch management commands/git-statusenhanced status with recommendations/git-configinteractive configuration- git-assistant agent for complex operations
- workflow-patterns skill with branching strategies
-
pr-review v1.0.0 - Multi-agent pull request review
/pr-reviewcommand for comprehensive multi-agent review/pr-summarycommand for quick PR overview/pr-findingscommand for filtering review findings- coordinator agent for orchestrating reviews
- security-reviewer, performance-analyst, maintainability-auditor, test-validator agents
- review-patterns skill with confidence scoring rules
Gitea MCP Server Enhancements
- 6 new Pull Request tools:
list_pull_requests- List PRs with filtersget_pull_request- Get PR detailsget_pr_diff- Get PR diffget_pr_comments- Get PR commentscreate_pr_review- Create review (approve, request changes, comment)add_pr_comment- Add comment to PR
Documentation
docs/CONFIGURATION.md- Centralized configuration guide for all plugins
Changed
- BREAKING: Marketplace renamed from
claude-code-marketplacetoleo-claude-mktplace - BREAKING: MCP servers moved from plugin directories to shared
mcp-servers/at repository root - All plugins now have
category,tags, andlicensefields in marketplace.json - Plugin MCP dependencies now use symlinks to shared servers
- projman version bumped to 3.0.0 (includes PR tools integration)
- projman CONFIGURATION.md slimmed down, links to central docs
Removed
- Standalone MCP server directories inside plugins (replaced with symlinks)
[2.3.0] - 2026-01-20
Added
New Plugins
-
doc-guardian v1.0.0 - Documentation lifecycle management
/doc-auditcommand for full project documentation drift analysis/doc-synccommand to batch apply pending documentation updates- PostToolUse hook for automatic drift detection
- Stop hook reminder for pending updates
- doc-analyzer agent for cross-reference analysis
- doc-patterns skill for documentation structure knowledge
-
code-sentinel v1.0.0 - Security scanning and refactoring
/security-scancommand for comprehensive security audit/refactorcommand to apply refactoring patterns/refactor-drycommand to preview refactoring opportunities- PreToolUse hook for real-time security scanning
- security-reviewer agent for vulnerability analysis
- refactor-advisor agent for code structure improvements
- security-patterns skill for vulnerability detection rules
projman Enhancements
/test-gencommand - Generate unit, integration, and e2e tests for specified code
Changed
- Marketplace version bumped to 2.3.0
- projman version bumped to 2.3.0
[2.2.0] - 2026-01-20
Added
/reviewcommand for pre-sprint-close code quality checks (projman)/test-checkcommand for test verification before sprint close (projman)code-revieweragent for structured code review workflow (projman)- Validation script (
scripts/validate-marketplace.sh) for marketplace compliance homepageandrepositoryfields to all plugin entries in marketplace.jsonmetadatawrapper for description/version in marketplace.json- Keywords to all plugin manifests for better discoverability
commandsandagentsdirectory references to plugin manifests- Versioning rule: version displayed only in main README.md title
Changed
- Updated marketplace.json with required fields per Claude Code spec
- Fixed installation documentation to use official Claude Code methods
- Prioritized public HTTPS URL over Tailscale SSH URL in documentation
- Updated all plugin manifests with author, homepage, repository, license fields
- Consolidated version display to main README.md title only
- Removed version numbers from plugin documentation titles
Fixed
- Plugin manifests now include all required fields per Claude Code spec
- Installation section uses
extraKnownMarketplacesinstead of undocumentedpluginMarketplace
Removed
docs/references/directory (obsolete planning documents)- Version numbers from individual plugin README titles
- Version section from plugins/projman/README.md
[2.1.0] - 2026-01-15
Added
docs/CANONICAL-PATHS.md- Single source of truth for all file paths- Path verification rules in CLAUDE.md (mandatory pre-flight check)
- Recovery protocol for path issues
- Installation script (
scripts/setup.sh) for new users - Post-update script (
scripts/post-update.sh) for updates - Update documentation (
docs/UPDATING.md) /initial-setupslash command- File creation governance rules in CLAUDE.md
.scratch/directory for transient workscripts/directory for setup automation
Changed
- Replaced
docs/CORRECT-ARCHITECTURE.mdreference withdocs/CANONICAL-PATHS.md - Added mandatory path verification section to CLAUDE.md
- Updated CLAUDE.md with file creation governance
Fixed
- Removed dead reference to non-existent
docs/CORRECT-ARCHITECTURE.md
Removed
- Organization/workspace GID variable (no longer needed)
- Development output files (test scripts, status reports)
- IDE-specific workspace files
- Stray files from project root
[2.0.0] - 2026-01-06
Added
- Full Gitea integration with wiki, milestones, dependencies
- Parallel execution batching via dependency graph
- Wiki tools for lessons learned (
create_lesson,search_lessons) - Milestone tools (
list_milestones,create_milestone,update_milestone) - Dependency tools (
list_issue_dependencies,create_issue_dependency,get_execution_order) - Validation tools (
validate_repo_org,get_branch_protection) - MCP servers bundled inside plugins (not shared at root)
Changed
- MCP server architecture: bundled in plugins instead of shared at root
- Configuration uses
${CLAUDE_PLUGIN_ROOT}/mcp-servers/paths
[1.0.0] - 2025-12-15
Added
- projman plugin with basic sprint commands
/sprint-plan,/sprint-start,/sprint-status,/sprint-closecommands/labels-synccommand for label taxonomy synchronization- Three-agent model (planner, orchestrator, executor)
- Gitea MCP server with issue and label tools
- 43-label taxonomy system
- Hybrid configuration system (system + project level)
- Branch-aware security model
[0.1.0] - 2025-12-01
Added
- Initial repository structure
- projman plugin structure (planned)
- projman-pmo plugin structure (planned)
- project-hygiene plugin for cleanup automation
- claude-config-maintainer plugin structure
- cmdb-assistant plugin structure
- Basic marketplace manifest