Update policy_stop_hook.sh to support both v1 and v2 policy formats

The hook now:
- Checks for v2 policies in .deepwork/policies/ first
- Falls back to v1 policies in .deepwork.policy.yml if no v2 found
- Passes JSON input directly to policy_check.py for v2 (via wrapper)
- Maintains existing behavior for v1 evaluate_policies.py

Author: claude

src/deepwork/standard_jobs/deepwork_policy/hooks/policy_stop_hook.sh

@@ -2,16 +2,27 @@
 # policy_stop_hook.sh - Evaluates policies when the agent stops
 #
 # This script is called as a Claude Code Stop hook. It:
-# 1. Evaluates policies from .deepwork.policy.yml
+# 1. Evaluates policies from .deepwork/policies/ (v2) or .deepwork.policy.yml (v1)
 # 2. Computes changed files based on each policy's compare_to setting
 # 3. Checks for <promise> tags in the conversation transcript
 # 4. Returns JSON to block stop if policies need attention
 
 set -e
 
-# Check if policy file exists
-if [ ! -f .deepwork.policy.yml ]; then
-    # No policies defined, nothing to do
+# Determine which policy system to use
+USE_V2=false
+V1_POLICY_FILE=".deepwork.policy.yml"
+V2_POLICY_DIR=".deepwork/policies"
+
+if [ -d "${V2_POLICY_DIR}" ]; then
+    # Check if there are any .md files in the v2 directory
+    if ls "${V2_POLICY_DIR}"/*.md 1>/dev/null 2>&1; then
+        USE_V2=true
+    fi
+fi
+
+# If no v2 policies and no v1 policy file, nothing to do
+if [ "${USE_V2}" = false ] && [ ! -f "${V1_POLICY_FILE}" ]; then
     exit 0
 fi
 
@@ -21,36 +32,38 @@ if [ ! -t 0 ]; then
     HOOK_INPUT=$(cat)
 fi
 
-# Extract transcript_path from the hook input JSON using jq
-# Claude Code passes: {"session_id": "...", "transcript_path": "...", ...}
-TRANSCRIPT_PATH=""
-if [ -n "${HOOK_INPUT}" ]; then
-    TRANSCRIPT_PATH=$(echo "${HOOK_INPUT}" | jq -r '.transcript_path // empty' 2>/dev/null || echo "")
-fi
+if [ "${USE_V2}" = true ]; then
+    # Use v2 policy system via cross-platform wrapper
+    # The wrapper reads JSON input and handles transcript extraction
+    result=$(echo "${HOOK_INPUT}" | DEEPWORK_HOOK_PLATFORM=claude DEEPWORK_HOOK_EVENT=Stop python -m deepwork.hooks.policy_check 2>/dev/null || echo '{}')
+else
+    # Use v1 policy system - extract conversation context for evaluate_policies
 
-# Extract conversation text from the JSONL transcript
-# The transcript is JSONL format - each line is a JSON object
-# We need to extract the text content from assistant messages
-conversation_context=""
-if [ -n "${TRANSCRIPT_PATH}" ] && [ -f "${TRANSCRIPT_PATH}" ]; then
-    # Extract text content from all assistant messages in the transcript
-    # Each line is a JSON object; we extract .message.content[].text for assistant messages
-    conversation_context=$(cat "${TRANSCRIPT_PATH}" | \
-        grep -E '"role"\s*:\s*"assistant"' | \
-        jq -r '.message.content // [] | map(select(.type == "text")) | map(.text) | join("\n")' 2>/dev/null | \
-        tr -d '\0' || echo "")
-fi
+    # Extract transcript_path from the hook input JSON using jq
+    # Claude Code passes: {"session_id": "...", "transcript_path": "...", ...}
+    TRANSCRIPT_PATH=""
+    if [ -n "${HOOK_INPUT}" ]; then
+        TRANSCRIPT_PATH=$(echo "${HOOK_INPUT}" | jq -r '.transcript_path // empty' 2>/dev/null || echo "")
+    fi
+
+    # Extract conversation text from the JSONL transcript
+    # The transcript is JSONL format - each line is a JSON object
+    # We need to extract the text content from assistant messages
+    conversation_context=""
+    if [ -n "${TRANSCRIPT_PATH}" ] && [ -f "${TRANSCRIPT_PATH}" ]; then
+        # Extract text content from all assistant messages in the transcript
+        # Each line is a JSON object; we extract .message.content[].text for assistant messages
+        conversation_context=$(cat "${TRANSCRIPT_PATH}" | \
+            grep -E '"role"\s*:\s*"assistant"' | \
+            jq -r '.message.content // [] | map(select(.type == "text")) | map(.text) | join("\n")' 2>/dev/null | \
+            tr -d '\0' || echo "")
+    fi
 
-# Call the Python evaluator
-# The Python module handles:
-# - Parsing the policy file
-# - Computing changed files based on each policy's compare_to setting
-# - Matching changed files against triggers/safety patterns
-# - Checking for promise tags in the conversation context
-# - Generating appropriate JSON output
-result=$(echo "${conversation_context}" | python -m deepwork.hooks.evaluate_policies \
-    --policy-file .deepwork.policy.yml \
-    2>/dev/null || echo '{}')
+    # Call the Python v1 evaluator
+    result=$(echo "${conversation_context}" | python -m deepwork.hooks.evaluate_policies \
+        --policy-file "${V1_POLICY_FILE}" \
+        2>/dev/null || echo '{}')
+fi
 
 # Output the result (JSON for Claude Code hooks)
 echo "${result}"