refactor: consolidate skill catalog from 60 to 44 skills (v0.9.0)

The catalog had grown structurally redundant: every workflow skill had a
knowledge twin covering the same ground (verify/verification-loop,
code-review/code-review-workflow/80-20-review, scaffold/scaffolding, ...),
four overlapping learning mechanisms, three instinct stubs that were really
argument modes, and skills shadowing native Claude Code features. Since all
44 descriptions load into context every session, the duplication cost ~25%
of the always-on token overhead and caused the routing collisions patched
in 0.8.0.

- Merge each workflow/knowledge twin into one skill carrying its
  methodology inline; scaffold's per-architecture templates move to a
  references/ file loaded on demand
- Unify instinct-system, self-correction-loop, learning-log, and the three
  instinct stubs into one learning skill with status/export/import modes
- Fold context-discipline into workflow-mastery; retire model-selection
  (rules already cover it) and split-memory (native memory supersedes it)
- Surviving descriptions absorb absorbed trigger phrases so existing
  invocation phrases still route
- Codify "one skill per concern" and the references/ pattern in
  CLAUDE.md and CONTRIBUTING.md; update all cross-references and counts

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: iammukeshm

.claude-plugin/marketplace.json

@@ -11,8 +11,8 @@
     {
       "name": "dotnet-claude-kit",
       "source": "./",
-      "description": "60 skills (incl. 16 slash-command workflows), 10 agents, 10 rules, 5 templates, Roslyn MCP server for .NET 10 / C# 14",
-      "version": "0.8.0",
+      "description": "44 skills (incl. 13 slash-command workflows), 10 agents, 10 rules, 5 templates, Roslyn MCP server for .NET 10 / C# 14",
+      "version": "0.9.0",
       "license": "MIT",
       "keywords": [
         "dotnet",

.claude-plugin/plugin.json

@@ -1,8 +1,8 @@
 {
   "name": "dotnet-claude-kit",
   "displayName": "dotnet Claude Kit",
-  "version": "0.8.0",
-  "description": "The definitive Claude Code companion for .NET developers. 60 skills (including 16 slash-command workflows), 10 agents, 10 rules, 5 templates, 15 MCP tools, and automation hooks for modern .NET 10 / C# 14.",
+  "version": "0.9.0",
+  "description": "The definitive Claude Code companion for .NET developers. 44 skills (including 13 slash-command workflows), 10 agents, 10 rules, 5 templates, 15 MCP tools, and automation hooks for modern .NET 10 / C# 14.",
   "author": {
     "name": "Mukesh Murugan",
     "url": "https://codewithmukesh.com"

.codex/AGENTS.md

@@ -22,13 +22,13 @@ This project uses [dotnet-claude-kit](https://github.com/codewithmukesh/dotnet-c
 Skills live in `skills/<skill-name>/SKILL.md` and follow the Agent Skills open standard.
 
 ### .NET Domain Skills
-api-versioning, architecture-advisor, aspire, authentication, caching, ci-cd, clean-architecture, configuration, ddd, dependency-injection, docker, ef-core, error-handling, httpclient-factory, logging, messaging, minimal-api, modern-csharp, openapi, opentelemetry, project-setup, project-structure, resilience, scaffolding, scalar, serilog, testing, vertical-slice, container-publish
+api-versioning, architecture-advisor, aspire, authentication, caching, ci-cd, clean-architecture, configuration, ddd, dependency-injection, docker, ef-core, error-handling, httpclient-factory, logging, messaging, minimal-api, modern-csharp, openapi, opentelemetry, project-setup, project-structure, resilience, scalar, serilog, testing, vertical-slice, container-publish
 
-### Workflow Skills
-code-review-workflow, convention-learner, migration-workflow, verification-loop, workflow-mastery
+### Workflow Skills (each registers a slash command and carries its methodology inline)
+build-fix, checkpoint, code-review, de-sloppify, dotnet-init, health-check, migrate, plan, scaffold, security-scan, tdd, verify, wrap-up
 
-### Meta & Productivity Skills
-80-20-review, context-discipline, learning-log, model-selection, self-correction-loop, split-memory, wrap-up-ritual
+### Workflow & Learning Skills
+convention-learner, workflow-mastery, instinct-system
 
 ## MCP Tools
 

AGENTS.md

@@ -60,15 +60,15 @@ Agents load skills in dependency order. Core skills load first.
 
 | Agent | Skills |
 |-------|--------|
-| dotnet-architect | modern-csharp, architecture-advisor, project-structure, scaffolding, project-setup + conditional: vertical-slice, clean-architecture, ddd |
+| dotnet-architect | modern-csharp, architecture-advisor, project-structure, scaffold, project-setup + conditional: vertical-slice, clean-architecture, ddd |
 | api-designer | modern-csharp, minimal-api, api-versioning, authentication, error-handling |
-| ef-core-specialist | modern-csharp, ef-core, configuration, migration-workflow |
+| ef-core-specialist | modern-csharp, ef-core, configuration, migrate |
 | test-engineer | modern-csharp, testing |
 | security-auditor | modern-csharp, authentication, configuration |
 | performance-analyst | modern-csharp, caching |
 | devops-engineer | modern-csharp, docker, ci-cd, aspire |
-| code-reviewer | modern-csharp, code-review-workflow, convention-learner + contextual (loads relevant skills incl. clean-architecture, ddd based on files under review) |
-| build-error-resolver | modern-csharp, autonomous-loops + contextual: ef-core, dependency-injection |
+| code-reviewer | modern-csharp, code-review, convention-learner + contextual (loads relevant skills incl. clean-architecture, ddd based on files under review) |
+| build-error-resolver | modern-csharp, build-fix + contextual: ef-core, dependency-injection |
 | refactor-cleaner | modern-csharp, de-sloppify + contextual: testing, ef-core |
 
 ## MCP Tool Preferences
@@ -91,58 +91,52 @@ Agents should **prefer Roslyn MCP tools over file scanning** to reduce token con
 
 ## Cross-Agent Meta Skills
 
-These 10 meta and productivity skills are not tied to a specific agent — any agent can load them when the context calls for it:
+These meta and productivity skills are not tied to a specific agent — any agent can load them when the context calls for it:
 
 | Skill | When to Load |
 |-------|-------------|
-| `self-correction-loop` | After ANY user correction — capture the rule in MEMORY.md |
-| `wrap-up-ritual` | User signals end of session — write handoff to `.claude/handoff.md` |
-| `context-discipline` | Context running low, large codebase navigation, planning exploration strategy |
-| `model-selection` | Choosing between Opus/Sonnet/Haiku, assigning subagent models |
-| `80-20-review` | Code review, PR review, deciding what to review in depth |
-| `split-memory` | CLAUDE.md exceeds 300 lines, need to split instructions across files |
-| `learning-log` | Non-obvious discovery during development — log the insight |
-| `instinct-system` | Pattern detection across sessions — observe-hypothesize-confirm cycle for project conventions |
-| `session-management` | Session start/end — load handoff, detect solution, write session summary |
-| `autonomous-loops` | Iterative fix loops — build-fix, test-fix, refactor with bounded iterations |
+| `instinct-system` | After ANY user correction, pattern detection across sessions, logging non-obvious discoveries; includes status/export/import modes |
+| `wrap-up` | Session start (load handoff) and session end (write handoff to `.claude/handoff.md`, capture learnings) |
+| `checkpoint` | Mid-session save before risky changes or task switches — commit + brief handoff |
+| `workflow-mastery` | Context running low, large codebase navigation, parallel workflows, subagent strategy |
+| `convention-learner` | Detect and enforce project-specific conventions in new code |
+
+Model selection guidance lives in the always-loaded `.claude/rules/agents.md` — no skill load needed.
 
 ### Meta Skill Routing
 
 | User Intent Pattern | Skill |
 |---|---|
-| "learn from mistakes", "remember this", "don't do that again" | self-correction-loop |
-| "wrap up", "done for today", "save progress", "handoff" | wrap-up-ritual |
-| "context", "running out of tokens", "too many files" | context-discipline |
-| "which model", "use Opus", "use Sonnet", "switch model" | model-selection |
-| "review this", "what should I review", "blast radius" | 80-20-review |
-| "split CLAUDE.md", "too long", "organize instructions" | split-memory |
-| "log this", "document this finding", "gotcha" | learning-log |
-| "show instincts", "what have you learned", "confidence scores" | instinct-system |
-| "start session", "load handoff", "session start" | session-management |
-| "fix build loop", "keep fixing", "auto-fix" | autonomous-loops |
+| "learn from mistakes", "remember this", "log this", "gotcha", "show instincts", "what have you learned" | instinct-system |
+| "wrap up", "done for today", "handoff", "start session", "load handoff" | wrap-up |
+| "save progress", "checkpoint", "pause here" | checkpoint |
+| "context", "running out of tokens", "too many files" | workflow-mastery |
+| "review this", "what should I review", "blast radius" | code-review |
+| "fix build loop", "keep fixing", "auto-fix" | build-fix |
 
 ## Slash Commands
 
 Commands map to skills and agents. Use these as shortcuts for common workflows.
 
-| Command | Primary Skill | Primary Agent | Purpose |
-|---------|--------------|---------------|---------|
+Each workflow skill registers its own slash command and carries its methodology inline (the kit no longer splits workflows from their knowledge twins).
+
+| Command | Supporting Skills | Primary Agent | Purpose |
+|---------|------------------|---------------|---------|
 | `/dotnet-init` | project-setup | dotnet-architect | Interactive project initialization |
 | `/plan` | architecture-advisor | dotnet-architect | Architecture-aware planning |
-| `/verify` | verification-loop | — | 7-phase verification pipeline |
+| `/verify` | — | — | 7-phase verification pipeline |
 | `/tdd` | testing | test-engineer | Red-green-refactor workflow |
-| `/scaffold` | scaffolding | dotnet-architect | Architecture-aware feature scaffolding |
-| `/code-review` | code-review-workflow | code-reviewer | MCP-powered code review |
-| `/build-fix` | autonomous-loops | build-error-resolver | Iterative build error fixing |
-| `/checkpoint` | wrap-up-ritual | — | Save progress (commit + handoff) |
-| `/security-scan` | security-scan | security-auditor | OWASP + secrets + dependency audit |
-| `/migrate` | migration-workflow | ef-core-specialist | Safe EF Core migration workflow |
-| `/health-check` | health-check | code-reviewer | Graded project health report |
-| `/de-sloppify` | de-sloppify | refactor-cleaner | Systematic code cleanup |
-| `/wrap-up` | wrap-up-ritual | — | Session ending ritual |
-| `/instinct-status` | instinct-system | — | Show learned instincts |
-| `/instinct-export` | instinct-system | — | Export instincts to shareable format |
-| `/instinct-import` | instinct-system | — | Import instincts from another project |
+| `/scaffold` | — | dotnet-architect | Architecture-aware feature scaffolding |
+| `/code-review` | convention-learner | code-reviewer | MCP-powered, blast-radius-prioritized code review |
+| `/build-fix` | — | build-error-resolver | Bounded build-fix and test-fix loops |
+| `/checkpoint` | — | — | Mid-session save (commit + handoff) |
+| `/security-scan` | — | security-auditor | OWASP + secrets + dependency audit |
+| `/migrate` | ef-core | ef-core-specialist | EF Core schema, .NET version, and NuGet migrations |
+| `/health-check` | — | code-reviewer | Graded project health report |
+| `/de-sloppify` | — | refactor-cleaner | Systematic code cleanup |
+| `/wrap-up` | instinct-system | — | Session handoff lifecycle (end + start) |
+
+Instinct operations (status, export, import) are modes of the `instinct-system` skill — say "show instincts", "export instincts", or "import instincts".
 
 ## Conflict Resolution
 
@@ -155,7 +149,7 @@ When two agents could handle a query:
 
 ## Token Budget Guidance
 
-For detailed context management strategies, see the **`context-discipline`** skill.
+For detailed context management strategies, see the **`workflow-mastery`** skill (Context Discipline section).
 
 - **Small queries** (single pattern/fix): Load 1-2 skills, use MCP tools for context
 - **Medium queries** (feature implementation): Load 3-4 skills, use MCP tools to understand existing code

CHANGELOG.md

@@ -5,6 +5,19 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.9.0] — 2026-06-12
+
+### Changed
+- **Skill catalog consolidated: 60 → 44 skills** — Sixteen redundant skills merged or retired to cut the always-loaded description overhead (~25%) and eliminate routing collisions between overlapping skills. Surviving descriptions absorb the absorbed skills' trigger phrases, so all existing invocation phrases still route correctly:
+  - **Workflow/knowledge twins merged** — each workflow now carries its methodology inline: `code-review` ← code-review-workflow + 80-20-review (blast-radius prioritization built in), `verify` ← verification-loop, `scaffold` ← scaffolding (per-architecture templates moved to `skills/scaffold/references/architecture-patterns.md`, loaded on demand), `migrate` ← migration-workflow (now also covers .NET version upgrades and NuGet updates), `build-fix` ← autonomous-loops (loop discipline: bounded iterations, progress detection, fail-safes; test-fix loop is a first-class variant), `wrap-up` ← wrap-up-ritual + session-management (owns the full handoff lifecycle including session start), `checkpoint` ← session-management's mid-session save
+  - **Learning skills unified** — `instinct-system` absorbs instinct-status/export/import (now modes: "show instincts", "export instincts", "import instincts"), self-correction-loop (corrections = confirmed instincts at full confidence → MEMORY.md), and learning-log (discoveries → .claude/learning-log.md)
+  - **`workflow-mastery` absorbs context-discipline** — token budget management and MCP-first navigation are now its Context Discipline section
+  - **Retired** — `model-selection` (the always-loaded `.claude/rules/agents.md` already carries model guidance) and `split-memory` (native Claude Code memory supersedes it)
+- **Slash commands: 16 → 13** — `/instinct-status`, `/instinct-export`, `/instinct-import` replaced by instinct-system modes; all other commands unchanged
+- **New structural rule: one skill per concern** — CLAUDE.md and CONTRIBUTING.md now prohibit workflow/knowledge twin pairs and document the `references/` progressive-disclosure pattern for deep content
+- **Cross-references updated everywhere** — AGENTS.md routing tables, README catalog and counts, docs/shorthand-guide.md, docs/longform-guide.md, .codex/AGENTS.md, all 5 templates, and 4 agent skill maps point at the surviving skills
+- Plugin version bumped to 0.9.0
+
 ## [0.8.0] — 2026-06-11
 
 ### Added

CLAUDE.md

@@ -121,8 +121,9 @@ description: >
 ### Quality Standards
 
 - **Maximum 200 lines** — Orchestrators are not encyclopedias (knowledge skills get 400)
-- **Invoke, don't implement** — Orchestrators reference skills and agents for the actual logic
-- **Clear trigger phrases** — Users should know when to reach for this workflow
+- **One skill per concern** — A workflow skill carries its methodology inline; never create a separate knowledge twin for a workflow (no `verify` + `verification-loop` pairs). Twin skills double the always-loaded description surface and cause routing collisions.
+- **Deep reference material goes in `references/`** — If methodology genuinely exceeds the budget (e.g. per-architecture code templates), put it in `skills/<name>/references/<topic>.md` and point to it from SKILL.md. References load only on demand — this is the token-friendly progressive-disclosure pattern.
+- **Clear trigger phrases** — Users should know when to reach for this workflow. When merging skills, the surviving description must absorb the absorbed skill's trigger phrases.
 
 ## Rule Structure
 

CONTRIBUTING.md

@@ -59,7 +59,8 @@ Claude Code merged slash commands into skills — there is no `commands/` direct
 - YAML frontmatter with `name` and `description` (add `disable-model-invocation: true` for explicit-only invocation)
 - Required sections: What, When, How, Example, Related
 - Maximum 200 lines (knowledge skills get 400)
-- Orchestrators invoke skills/agents — they don't contain the logic themselves
+- One skill per concern — a workflow carries its methodology inline; never create a separate knowledge twin for a workflow (twins double the always-loaded description surface and cause routing collisions)
+- Deep reference material that exceeds the budget goes in `skills/<name>/references/<topic>.md`, loaded on demand from SKILL.md
 
 ### Agents