feat: add skill-creator skill with eval-driven validation and blind A/B testing

.claude/settings.json

@@ -96,81 +96,109 @@
     ],
     "PreToolUse": [
       {
+        "matcher": "Bash|Write|Edit",
         "hooks": [
           {
             "type": "command",
             "command": "python3 \"$HOME/.claude/hooks/pretool-unified-gate.py\"",
             "description": "Unified gate: gitignore-bypass, git-submission, dangerous-command, creation-gate, sensitive-file (ADR-068)",
             "timeout": 3000
-          },
+          }
+        ]
+      },
+      {
+        "matcher": "Bash",
+        "hooks": [
           {
             "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/pretool-synthesis-gate.py\"",
-            "description": "Consultation synthesis gate: blocks implementation when ADR consultation is incomplete",
+            "command": "python3 \"$HOME/.claude/hooks/pretool-branch-safety.py\"",
+            "description": "Branch safety: blocks git commit on main/master, forces feature branches",
             "timeout": 3000
           },
           {
             "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/pretool-branch-safety.py\"",
-            "description": "Branch safety: blocks git commit on main/master, forces feature branches",
+            "command": "python3 \"$HOME/.claude/hooks/ci-merge-gate.py\"",
+            "description": "Gate: block merge to main/master when CI checks are red",
             "timeout": 3000
-          },
+          }
+        ]
+      },
+      {
+        "matcher": "Bash|Edit",
+        "hooks": [
           {
             "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/pretool-plan-gate.py\"",
-            "description": "Plan gate: blocks implementation code without task_plan.md",
+            "command": "python3 \"$HOME/.claude/hooks/pretool-learning-injector.py\"",
+            "description": "Inject known error patterns before Bash/Edit tools run",
             "timeout": 3000
-          },
+          }
+        ]
+      },
+      {
+        "matcher": "Write|Edit",
+        "hooks": [
           {
             "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/pretool-adr-creation-gate.py\"",
-            "description": "ADR creation gate: blocks new components without an ADR in adr/",
+            "command": "python3 \"$HOME/.claude/hooks/pretool-synthesis-gate.py\"",
+            "description": "Consultation synthesis gate: blocks implementation when ADR consultation is incomplete",
             "timeout": 3000
           },
           {
             "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/pretool-learning-injector.py\"",
-            "description": "Inject known error patterns before Bash/Edit tools run",
+            "command": "python3 \"$HOME/.claude/hooks/pretool-plan-gate.py\"",
+            "description": "Plan gate: blocks implementation code without task_plan.md",
             "timeout": 3000
           },
           {
             "type": "command",
             "command": "python3 \"$HOME/.claude/hooks/pretool-prompt-injection-scanner.py\"",
             "description": "Advisory scan for prompt injection patterns in agent context files (ADR-070)",
             "timeout": 3000
-          },
+          }
+        ]
+      },
+      {
+        "matcher": "Write",
+        "hooks": [
           {
             "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/pretool-subagent-warmstart.py\"",
-            "description": "Inject parent session context into subagent prompts (ADR-088)",
-            "timeout": 5000
-          },
+            "command": "python3 \"$HOME/.claude/hooks/pretool-adr-creation-gate.py\"",
+            "description": "ADR creation gate: blocks new components without an ADR in adr/",
+            "timeout": 3000
+          }
+        ]
+      },
+      {
+        "matcher": "Edit",
+        "hooks": [
           {
             "type": "command",
             "command": "python3 \"$HOME/.claude/hooks/pretool-file-backup.py\"",
             "description": "Backup files before Edit tool modifies them",
             "timeout": 3000
-          },
+          }
+        ]
+      },
+      {
+        "matcher": "Agent",
+        "hooks": [
           {
             "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/ci-merge-gate.py\"",
-            "description": "Gate: block merge to main/master when CI checks are red",
-            "timeout": 3000
+            "command": "python3 \"$HOME/.claude/hooks/pretool-subagent-warmstart.py\"",
+            "description": "Inject parent session context into subagent prompts (ADR-088)",
+            "timeout": 5000
           }
         ]
       }
     ],
     "PostToolUse": [
       {
+        "matcher": "Write|Edit",
         "hooks": [
           {
             "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/post-tool-lint-hint.py\""
-          },
-          {
-            "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/error-learner.py\"",
-            "description": "Learn from tool errors and suggest solutions"
+            "command": "python3 \"$HOME/.claude/hooks/post-tool-lint-hint.py\"",
+            "description": "Gentle lint reminder after file modifications"
           },
           {
             "type": "command",
@@ -185,48 +213,82 @@
           },
           {
             "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/routing-gap-recorder.py\"",
-            "description": "Record /do routing gaps to learning DB for pattern tracking",
-            "timeout": 2000
-          },
+            "command": "python3 \"$HOME/.claude/hooks/posttool-security-scan.py\"",
+            "description": "Advisory scan for credentials and SQL injection in Write/Edit output",
+            "timeout": 3000
+          }
+        ]
+      },
+      {
+        "matcher": "Bash",
+        "hooks": [
           {
             "type": "command",
             "command": "python3 \"$HOME/.claude/hooks/retro-graduation-gate.py\"",
             "description": "Warn about ungraduated retro entries when creating PRs in toolkit repo",
             "timeout": 3000
-          },
+          }
+        ]
+      },
+      {
+        "matcher": "Edit|Write|Bash",
+        "hooks": [
           {
             "type": "command",
             "command": "python3 \"$HOME/.claude/hooks/record-activation.py\"",
             "description": "Record session activation stats for ROI tracking (ADR-032)"
-          },
-          {
-            "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/record-waste.py\"",
-            "description": "Record wasted tokens from tool failures for ROI tracking (ADR-032)"
-          },
+          }
+        ]
+      },
+      {
+        "matcher": "Read",
+        "hooks": [
           {
             "type": "command",
             "command": "python3 \"$HOME/.claude/hooks/posttool-session-reads.py\"",
             "description": "Track files read this session for subagent warmstart (ADR-088)"
-          },
-          {
-            "type": "command",
-            "command": "python3 \"$HOME/.claude/hooks/posttool-security-scan.py\"",
-            "description": "Advisory scan for credentials and SQL injection in Write/Edit output",
-            "timeout": 3000
-          },
+          }
+        ]
+      },
+      {
+        "matcher": "Skill|Agent",
+        "hooks": [
           {
             "type": "command",
             "command": "python3 \"$HOME/.claude/hooks/usage-tracker.py\"",
             "description": "Record Skill and Agent invocation analytics",
             "timeout": 3000
-          },
+          }
+        ]
+      },
+      {
+        "matcher": "Agent",
+        "hooks": [
           {
             "type": "command",
             "command": "python3 \"$HOME/.claude/hooks/review-capture.py\"",
             "description": "Capture CRITICAL/HIGH review findings to learning DB",
             "timeout": 3000
+          }
+        ]
+      },
+      {
+        "hooks": [
+          {
+            "type": "command",
+            "command": "python3 \"$HOME/.claude/hooks/error-learner.py\"",
+            "description": "Learn from tool errors and suggest solutions"
+          },
+          {
+            "type": "command",
+            "command": "python3 \"$HOME/.claude/hooks/routing-gap-recorder.py\"",
+            "description": "Record /do routing gaps to learning DB for pattern tracking",
+            "timeout": 2000
+          },
+          {
+            "type": "command",
+            "command": "python3 \"$HOME/.claude/hooks/record-waste.py\"",
+            "description": "Record wasted tokens from tool failures for ROI tracking (ADR-032)"
           },
           {
             "type": "command",

.gitignore

@@ -94,3 +94,11 @@ draft-*.md
 
 # Scratch notes (session working files, not committed)
 scratch/
+
+# Eval workspaces (A/B/C test outputs, generated code, grading artifacts)
+# These are ephemeral experiment data — not committed
+*-workspace/
+evals/
+
+# Feature state (ephemeral, per-session feature lifecycle)
+.feature/

agents/INDEX.json

@@ -4,7 +4,7 @@
   "agents": {
     "agent-creator-engineer": {
       "file": "agent-creator-engineer.md",
-      "short_description": "**DEPRECATED**: Use skill-creator-engineer agent instead",
+      "short_description": "**DEPRECATED**: Use skill-creator skill instead",
       "triggers": [
         "create agent",
         "new agent",
@@ -14,7 +14,7 @@
         "legacy agent creation"
       ],
       "pairs_with": [
-        "skill-creator-engineer",
+        "skill-creator",
         "agent-evaluation"
       ],
       "complexity": "Simple",
@@ -1066,26 +1066,6 @@
       "complexity": "Simple",
       "category": "meta"
     },
-    "skill-creator-engineer": {
-      "file": "skill-creator-engineer.md",
-      "short_description": "Use this agent when creating new Claude Code skills, designing workflow automation,\nor improving existing skill architecture",
-      "triggers": [
-        "create skill",
-        "new skill",
-        "skill template",
-        "skill design",
-        "workflow automation",
-        "skill improvement",
-        "refactor skill"
-      ],
-      "pairs_with": [
-        "agent-evaluation",
-        "verification-before-completion",
-        "workflow-orchestrator"
-      ],
-      "complexity": "Medium-Complex",
-      "category": "meta"
-    },
     "sqlite-peewee-engineer": {
       "file": "sqlite-peewee-engineer.md",
       "short_description": "Use this agent when you need expert assistance with SQLite database development using the Peewee ORM in Python",

agents/README.md

@@ -96,12 +96,12 @@ Each agent is defined in `agents/*.md` with YAML frontmatter specifying model, v
 
 | Agent | Description |
 |-------|-------------|
-| `skill-creator-engineer` | Create Claude Code skills: progressive disclosure, SKILL.md structure, complexity tier selection |
+| `skill-creator` | Create Claude Code skills: progressive disclosure, SKILL.md structure, complexity tier selection |
 | `hook-development-engineer` | Python hooks: PostToolUse/PreToolUse/SessionStart handlers, sub-50ms performance, learning DB |
 | `pipeline-orchestrator-engineer` | Build pipelines: multi-component scaffolding, fan-out/fan-in patterns, routing integration |
 | `system-upgrade-engineer` | Ecosystem upgrades: 6-phase pipeline for adapting to Claude Code releases or goal shifts |
 | `toolkit-governance-engineer` | Toolkit internal architecture: SKILL.md edits, routing tables, ADR lifecycle, INDEX.json, hook compliance |
-| `agent-creator-engineer` | **DEPRECATED** — use `skill-creator-engineer` instead |
+| `agent-creator-engineer` | **DEPRECATED** — use `skill-creator` instead |
 
 ---