fix: remove prompt hooks that blocked operations, export CLAUDE_PLUGIN_ROOT

Prompt hooks (UserPromptSubmit, PostToolUse, PreCompact) caused Claude to
treat workflow guidance as hard gates, blocking normal user messages. Replaced
with command-only hook architecture: hooks provide dynamic state data,
CLAUDE.md provides static workflow instructions.

Also fixed hook-dispatch.sh not exporting CLAUDE_PLUGIN_ROOT, which caused
Go binaries to fail with "plugin root not found".

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: rand

.claude-plugin/marketplace.json

@@ -10,7 +10,7 @@
       "name": "rlm-claude-code",
       "source": "./",
       "description": "Recursive Language Model integration for Claude Code - Go binary hooks (~5ms startup), cross-plugin event coordination (DP/RLM), version-aware config migration, and intelligent multi-provider routing",
-      "version": "0.6.0"
+      "version": "0.6.1"
     }
   ]
 }

.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "rlm-claude-code",
   "description": "Recursive Language Model integration for Claude Code - intelligent multi-provider routing and unbounded context handling",
-  "version": "0.6.0",
+  "version": "0.6.1",
   "author": {
     "name": "Rand",
     "url": "https://github.com/rand"

CHANGELOG.md

@@ -1,5 +1,19 @@
 # Changelog
 
+## [0.6.1] - 2026-02-02
+
+### Fixed
+- **hook-dispatch.sh**: Export `CLAUDE_PLUGIN_ROOT` env var so Go binaries can find plugin root (fixes "plugin root not found" error)
+
+### Removed
+- `UserPromptSubmit` prompt hook — caused Claude to block normal messages; RLM guidance now lives in CLAUDE.md
+- `PostToolUse` catch-all prompt hook — fired on every tool use adding cost/latency; redundant with PreToolUse complexity-check
+- `PreCompact` prompt hook — low value; trajectory-save on Stop handles persistence
+
+### Changed
+- All hooks are now command-type only (no prompt hooks remain) — eliminates invisible gating behavior
+- Hook architecture: command hooks provide dynamic state data, CLAUDE.md provides static workflow instructions
+
 ## [0.6.0] - 2026-02-02
 
 ### Added

hooks/hooks.json

@@ -13,18 +13,6 @@
         ]
       }
     ],
-    "UserPromptSubmit": [
-      {
-        "matcher": ".*",
-        "hooks": [
-          {
-            "type": "prompt",
-            "prompt": "You have access to RLM (Recursive Language Model) for complex tasks. If the user's request involves multi-file analysis, debugging complex issues, architectural decisions, or exhaustive searches, consider activating RLM with `/rlm mode balanced` or `/rlm mode thorough`. For simple tasks, RLM micro mode runs automatically. Do not mention RLM unless the task complexity warrants it.",
-            "description": "RLM activation guidance based on task complexity"
-          }
-        ]
-      }
-    ],
     "PreToolUse": [
       {
         "matcher": ".*",
@@ -38,30 +26,6 @@
         ]
       }
     ],
-    "PostToolUse": [
-      {
-        "matcher": ".*",
-        "hooks": [
-          {
-            "type": "prompt",
-            "prompt": "If the tool output reveals unexpected complexity (large error traces, many files affected, cross-module dependencies), consider suggesting RLM escalation to the user. RLM mode should align with the current DP phase: spec/review phases warrant thorough mode, orient/decide/test/implement phases warrant balanced mode.",
-            "description": "Post-tool RLM escalation check with DP phase awareness"
-          }
-        ]
-      }
-    ],
-    "PreCompact": [
-      {
-        "matcher": ".*",
-        "hooks": [
-          {
-            "type": "prompt",
-            "prompt": "Before compaction, ensure any active RLM context (trajectory, memory state, orchestration state) has been externalized. If RLM is active, save trajectory with `/rlm save`.",
-            "description": "Externalize RLM context before compaction"
-          }
-        ]
-      }
-    ],
     "Stop": [
       {
         "matcher": ".*",

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "rlm-claude-code"
-version = "0.6.0"
+version = "0.6.1"
 description = "Recursive Language Model integration for Claude Code"
 readme = "README.md"
 requires-python = ">=3.12"

scripts/hook-dispatch.sh

@@ -10,6 +10,7 @@ set -e
 
 # Determine plugin root
 PLUGIN_ROOT="${CLAUDE_PLUGIN_ROOT:-$(cd "$(dirname "$0")/.." && pwd)}"
+export CLAUDE_PLUGIN_ROOT="$PLUGIN_ROOT"
 HOOK_NAME="$1"
 
 if [ -z "$HOOK_NAME" ]; then