feat: add elegance-pipeline plugin (Codex bundle -> Claude Code native)

.claude-plugin/marketplace.json

@@ -5,13 +5,13 @@
     "url": "https://github.com/ANcpLua"
   },
   "metadata": {
-    "description": "Claude Code plugin marketplace: 10 plugins, 25 commands, 5 skills, 19 agents. Multi-agent orchestration, cognitive amplification, OpenTelemetry docs, .NET build enforcement, and design intelligence."
+    "description": "Claude Code plugin marketplace: 11 plugins, 28 commands, 6 skills, 24 agents. Multi-agent orchestration, cognitive amplification, OpenTelemetry docs, .NET build enforcement, design intelligence, and code elegance pipeline."
   },
   "plugins": [
     {
       "name": "metacognitive-guard",
       "description": "Cognitive amplification stack: epistemic hooks, competitive review, commit integrity checks, CI verification scripts, and deep-thinking agents.",
-      "version": "0.4.5",
+      "version": "0.5.0",
       "source": "./plugins/metacognitive-guard"
     },
     {
@@ -29,7 +29,7 @@
     {
       "name": "hookify",
       "description": "User-configurable hooks from .local.md files. Define blocking and warning rules for Bash commands, file edits, and session events.",
-      "version": "0.2.0",
+      "version": "0.2.1",
       "source": "./plugins/hookify"
     },
     {
@@ -67,6 +67,12 @@
       "description": "Design intelligence studio: creative direction + data-driven recommendations. 50 styles, 97 palettes, 57 font pairings, 99 UX guidelines, 25 chart types, 13 stacks. BM25 search engine with design system generator.",
       "version": "1.0.0",
       "source": "./plugins/design-studio"
+    },
+    {
+      "name": "elegance-pipeline",
+      "description": "Multi-agent code-elegance workflow: 4 scouts (sonnet), 2 judges (opus), 1 planner, 1 verifier, 1 gated implementer. Persistent state with stage gates.",
+      "version": "1.0.0",
+      "source": "./plugins/elegance-pipeline"
     }
   ]
 }

.gitignore

@@ -45,6 +45,10 @@ obj/
 # --- Eight Gates runtime ---
 .eight-gates/
 
+# --- Elegance pipeline runtime state ---
+.claude/elegance_pipeline/
+.codex.backup.*
+
 # --- Metacognitive-guard runtime state ---
 .blackboard/
 

CHANGELOG.md

@@ -8,6 +8,14 @@ and the project follows [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 
 ### Added
 
+- **`elegance-pipeline` plugin (1.0.0)**: Multi-agent code-elegance workflow converted from Codex bundle to native Claude Code plugin. 4 scouts (sonnet, read-only), 2 judges (opus, read-only), 1 planner (opus), 1 verifier (opus), 1 gated implementer (opus, full edit). Persistent state manager with stage gates and implementation signal. 3 commands (`init`, `status`, `run`), 1 skill, 5 agents. All `.codex/` paths rewritten to `${CLAUDE_PLUGIN_ROOT}`, Codex-Spark->sonnet, GPT-5.4->opus, state moved to project-local `.claude/elegance_pipeline/state/`
+
+### Fixed
+
+- **marketplace.json version mismatches**: hookify 0.2.0->0.2.1, metacognitive-guard 0.4.5->0.5.0 (synced with plugin.json)
+
+### Added
+
 - **Codex PR review automation**: Added `.github/workflows/codex-code-review.yml`, `.github/codex/prompts/review.md`, and `.github/codex/schemas/review-output.schema.json`. Codex now reviews pull requests in a read-only sandbox, returns structured verdicts, publishes formal GitHub reviews, and skips PRs that only modify Codex review automation
 - **metacognitive-guard `InstructionsLoaded` hook**: Truth beacon now fires on both SessionStart AND InstructionsLoaded — ground truth re-injected when CLAUDE.md/rules are loaded, ensuring authoritative facts arrive after instructions context
 - **metacognitive-guard `agent_type` filtering**: Struggle detector and Ralph Loop now skip subagents via `agent_type` field in hook events — prevents wasted haiku calls and false positives from subagent responses

plugins/elegance-pipeline/.claude-plugin/plugin.json

@@ -0,0 +1,27 @@
+{
+  "name": "elegance-pipeline",
+  "version": "1.0.0",
+  "description": "Multi-agent code-elegance workflow: 4 scouts, 2 judges, 1 planner, 1 verifier, 1 gated implementer. Persistent state manager with stage gates and implementation signal.",
+  "author": {
+    "name": "ANcpLua"
+  },
+  "repository": "https://github.com/ANcpLua/ancplua-claude-plugins",
+  "license": "MIT",
+  "keywords": [
+    "elegance",
+    "code-quality",
+    "multi-agent",
+    "refactoring",
+    "pipeline",
+    "scout",
+    "judge"
+  ],
+  "commands": "./commands",
+  "agents": [
+    "./agents/elegance-scout.md",
+    "./agents/elegance-judge.md",
+    "./agents/elegance-planner.md",
+    "./agents/elegance-verifier.md",
+    "./agents/elegance-implementer.md"
+  ]
+}

plugins/elegance-pipeline/README.md

@@ -0,0 +1,43 @@
+# elegance-pipeline
+
+Multi-agent code-elegance workflow for Claude Code. Evaluates source files across a repository for elegance (difficulty handled / solution complexity) and optionally converts justified weaknesses into narrowly scoped refactor work.
+
+## Pipeline
+
+```text
+4 Scouts (parallel, sonnet) -> 2 Judges (parallel, opus) -> 1 Planner (opus)
+  -> 1 Verifier (opus) -> 1 Implementer (gated, opus)
+```
+
+- **Scouts**: Read-only. Each inspects an assigned scope for elegant code candidates.
+- **Judges**: Read-only. Verify scout findings against the codebase, produce final top-5 ranking.
+- **Planner**: Read-only. Converts judge verdicts into max 3 actionable refactor tasks.
+- **Verifier**: Read-only. Validates the plan is justified and narrow. Controls the implementation gate.
+- **Implementer**: Full edit access. Only runs when the verifier approves. Implements the verified plan.
+
+## Setup
+
+```bash
+python plugins/elegance-pipeline/elegance_pipeline/pipeline.py init \
+  --project-anchor CLAUDE.md \
+  --scope plugins/exodia \
+  --scope plugins/metacognitive-guard \
+  --scope plugins/hookify \
+  --scope plugins/feature-dev
+```
+
+## Usage
+
+```text
+/elegance-pipeline:status     # Check pipeline state
+/elegance-pipeline:run        # Run next ready stage
+/elegance-pipeline:run scouts # Run only scout phase
+```
+
+## State
+
+State is project-local at `.claude/elegance_pipeline/state/`. Not committed to git.
+
+## Origin
+
+Converted from the Codex `elegance-pipeline-bundle.zip`. All `.codex/` references rewritten to Claude Code native plugin structure. Codex-Spark -> sonnet, GPT-5.4 -> opus. Manual thread creation replaced with Claude Code subagent orchestration.

plugins/elegance-pipeline/agents/elegance-implementer.md

@@ -0,0 +1,14 @@
+---
+name: elegance-implementer
+description: Implementation agent. Executes the verifier-approved refactor plan with minimal correct changes. Only runs when the implementation signal is READY. Use when running the elegance pipeline implementer phase.
+tools: Read, Write, Edit, Grep, Glob, Bash
+model: opus
+---
+
+You are the implementation agent in the elegance pipeline.
+
+Your job is to correctly implement the verifier-approved plan with the smallest change set that fully satisfies the decision. You understand scope before changing code, prefer the project's existing abstractions, and verify your work.
+
+You have full edit access. Follow the 4-phase protocol: understand scope, plan, implement, verify.
+
+When you receive your task prompt (rendered by the pipeline state manager), follow it exactly. Submit your results through the pipeline state manager command provided in the prompt.

plugins/elegance-pipeline/agents/elegance-judge.md

@@ -0,0 +1,14 @@
+---
+name: elegance-judge
+description: Code elegance judge. Verifies scout findings against the actual codebase and produces the final top-5 ranking. Use when running the elegance pipeline judge phase.
+tools: Read, Grep, Glob, Bash
+model: opus
+---
+
+You are a code elegance judge in the elegance pipeline.
+
+Your job is to take the shortlist from all 4 scouts, verify finalists directly in the codebase, and produce the definitive top-5 most elegant source files. You score by difficulty times cleanliness.
+
+You are read-only. Do not edit, create, or delete any files.
+
+When you receive your task prompt (rendered by the pipeline state manager), follow it exactly. Submit your results through the pipeline state manager command provided in the prompt.

plugins/elegance-pipeline/agents/elegance-planner.md

@@ -0,0 +1,14 @@
+---
+name: elegance-planner
+description: Refactor planner. Converts judge verdicts into actionable engineering tasks only when evidence justifies real change. Use when running the elegance pipeline planner phase.
+tools: Read, Grep, Glob, Bash
+model: opus
+---
+
+You are the refactor planner in the elegance pipeline.
+
+Your job is to convert judge verdicts into at most 3 narrow, high-confidence implementation tasks. You do not invent work. If the judges found no actionable weaknesses, you say "No implementation warranted."
+
+You are read-only. Do not edit, create, or delete any files.
+
+When you receive your task prompt (rendered by the pipeline state manager), follow it exactly. Submit your results through the pipeline state manager command provided in the prompt.

plugins/elegance-pipeline/agents/elegance-scout.md

@@ -0,0 +1,14 @@
+---
+name: elegance-scout
+description: Read-only code elegance scout. Inspects an assigned scope for the most elegant source files. Use when running the elegance pipeline scout phase.
+tools: Read, Grep, Glob, Bash
+model: sonnet
+---
+
+You are a code elegance scout in the elegance pipeline.
+
+Your job is to inspect source files in your assigned scope and identify the strongest candidates for elegance. You evaluate files by the ratio of problem complexity to solution complexity.
+
+You are read-only. Do not edit, create, or delete any files. Do not run formatting, refactoring, or build commands.
+
+When you receive your task prompt (rendered by the pipeline state manager), follow it exactly. Submit your results through the pipeline state manager command provided in the prompt.

plugins/elegance-pipeline/agents/elegance-verifier.md

@@ -0,0 +1,14 @@
+---
+name: elegance-verifier
+description: Plan verifier judge. Validates that the planner extracted the right work from judge evidence. Controls the implementation gate signal. Use when running the elegance pipeline verifier phase.
+tools: Read, Grep, Glob, Bash
+model: opus
+---
+
+You are the plan verifier judge in the elegance pipeline.
+
+Your job is to verify whether the refactor planner extracted the right work from the judge evidence. You reject fabricated work, under-scoped plans, and over-broad cleanup sprawl. Your verdict controls the implementation signal gate.
+
+You are read-only. Do not edit, create, or delete any files.
+
+When you receive your task prompt (rendered by the pipeline state manager), follow it exactly. Submit your results through the pipeline state manager command provided in the prompt.

plugins/elegance-pipeline/commands/init.md

@@ -0,0 +1,22 @@
+---
+description: Initialize the elegance pipeline for this repository
+disable-model-invocation: true
+---
+
+Initialize the elegance pipeline state manager.
+
+Run:
+
+```bash
+python ${CLAUDE_PLUGIN_ROOT}/elegance_pipeline/pipeline.py init \
+  --project-anchor $ARGUMENTS \
+  --scope <scope-1> \
+  --scope <scope-2> \
+  --scope <scope-3> \
+  --scope <scope-4>
+```
+
+If `$ARGUMENTS` is empty, ask the user for the project anchor file and 4 scout scopes.
+
+The project anchor is any meaningful root file (e.g., `CLAUDE.md`, `package.json`, `*.sln`).
+Scout scopes are directories that each scout will analyze independently.

plugins/elegance-pipeline/commands/run.md

@@ -0,0 +1,54 @@
+---
+description: Run the next ready stage of the elegance pipeline
+disable-model-invocation: true
+allowed-tools: Read, Grep, Glob, Bash, Agent
+---
+
+Run the next ready stage of the elegance pipeline.
+
+## Procedure
+
+1. Check status:
+
+```bash
+python ${CLAUDE_PLUGIN_ROOT}/elegance_pipeline/pipeline.py status
+```
+
+2. For each ready slot, render the prompt:
+
+```bash
+python ${CLAUDE_PLUGIN_ROOT}/elegance_pipeline/pipeline.py prompt --role <role> --slot <slot>
+```
+
+3. Spawn the matching subagent with the rendered prompt as the task:
+   - `scout-*` slots -> spawn `elegance-scout` agent (run all 4 in parallel as background agents)
+   - `judge-*` slots -> spawn `elegance-judge` agent (run both in parallel as background agents)
+   - `planner-1` -> spawn `elegance-planner` agent
+   - `verifier-1` -> spawn `elegance-verifier` agent
+   - `implementer-1` -> spawn `elegance-implementer` agent
+
+4. After each agent completes, pipe its result to the state manager:
+
+```bash
+echo "<agent output>" | python ${CLAUDE_PLUGIN_ROOT}/elegance_pipeline/pipeline.py submit \
+  --role <role> --slot <slot> --stdin
+```
+
+5. Check status again and repeat for newly unlocked slots.
+
+6. Stop when no more slots are ready or the pipeline is complete.
+
+## Arguments
+
+Pass `$ARGUMENTS` to control which stage to run:
+- `scouts` — run only scout phase
+- `judges` — run only judge phase
+- `all` — run all phases sequentially (default)
+- A specific slot like `scout-1` or `judge-2`
+
+## Rules
+
+- Never bypass stage gates
+- Run scouts and judges in parallel (background agents)
+- Run planner, verifier, implementer sequentially (foreground)
+- If the verifier says no implementation is warranted, report that and stop

plugins/elegance-pipeline/commands/status.md

@@ -0,0 +1,14 @@
+---
+description: Show elegance pipeline workflow status
+disable-model-invocation: true
+---
+
+Show the current elegance pipeline status.
+
+Run:
+
+```bash
+python ${CLAUDE_PLUGIN_ROOT}/elegance_pipeline/pipeline.py status
+```
+
+Report the output to the user, highlighting which agents are ready to run next.