fix: guide Claude to use goals instead of plans and auto-continue steps

- Add behavioral guidance to goal context: use goal system for multi-step
  tasks instead of EnterPlanMode, and continue to next step after completing one
- Stop-reorient hook now explicitly tells Claude to continue with active goal
- PostToolUse hook nudges continuation when a step is completed

Co-developed-by: Claude Code v2.1.50 (claude-opus-4-6)

Author: robtaylor

hooks/goals-post-tool.sh

@@ -6,11 +6,22 @@ set -euo pipefail
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../scripts" && pwd)"
 PROJECT_ROOT="${PWD}"
 
+# Read hook input from stdin to detect which tool was called
+INPUT=$(cat)
+TOOL_NAME=$(echo "$INPUT" | python3 -c "import sys,json; print(json.load(sys.stdin).get('tool_name',''))" 2>/dev/null || true)
+TOOL_OUTPUT=$(echo "$INPUT" | python3 -c "import sys,json; print(json.load(sys.stdin).get('tool_result',''))" 2>/dev/null || true)
+
 # Get updated status
 GOAL_STATUS=$(bash "${SCRIPT_DIR}/goal-context-helper.sh" --status "${PROJECT_ROOT}" 2>/dev/null || true)
 
 if [[ -n "${GOAL_STATUS}" ]]; then
-    STATUS_ESCAPED=$(echo -n "${GOAL_STATUS}" | python3 -c "import sys,json; print(json.dumps(sys.stdin.read()))" | sed 's/^"//;s/"$//')
+    # Add continuation nudge when a step was just completed
+    NUDGE=""
+    if [[ "${TOOL_NAME}" == *"goal_update_step"* ]] && echo "${TOOL_OUTPUT}" | grep -q "completed"; then
+        NUDGE=" | Continue to the next step."
+    fi
+    MSG="${GOAL_STATUS}${NUDGE}"
+    STATUS_ESCAPED=$(echo -n "${MSG}" | python3 -c "import sys,json; print(json.dumps(sys.stdin.read()))" | sed 's/^"//;s/"$//')
     cat << EOF
 {
   "systemMessage": "${STATUS_ESCAPED}"

scripts/goal-context-helper.sh

@@ -197,5 +197,12 @@ if learnings:
 lines.append(f"   Full goal: ~/.claude/goals/{goal_uuid}.md")
 lines.append("   Commands: /context-daddy:goal (manage) | /context-daddy:goal-done (complete step) | /context-daddy:goal-focus (change step)")
 
+# Behavioral guidance
+lines.append("")
+lines.append("   **IMPORTANT - Goal-driven workflow:**")
+lines.append("   - When the user asks for a large multi-step task, use the goal system (goal_create + goal_add_step) instead of EnterPlanMode.")
+lines.append("   - After completing a step, use goal_update_step to mark it done, then CONTINUE to the next step without waiting to be asked.")
+lines.append("   - Only stop between steps if you need user input or clarification.")
+
 print("\n".join(lines))
 ' 2>/dev/null || true

scripts/stop-reorient.sh

@@ -95,8 +95,7 @@ INSTRUCTIONS="${INSTRUCTIONS}
 1. **Read ${CLAUDE_DIR}/CLAUDE.md** (if exists) - Project rules
 2. **Read ${CLAUDE_DIR}/learnings.md** (if exists) - Recent discoveries
 3. **Update narrative** (if significant learning): Run \`/context-daddy:refresh\`
-
-Then continue with the current task."
+4. **Check the active goal** above - identify the current step and continue working on it. Do NOT wait to be asked."
 
 # Escape for JSON
 INSTRUCTIONS_ESCAPED=$(echo -n "$INSTRUCTIONS" | python3 -c "import sys,json; print(json.dumps(sys.stdin.read()))" | sed 's/^"//;s/"$//')