DAAF-Contribution-Community · brhkim · Apr 5, 2026 · Apr 5, 2026 · Apr 5, 2026 · Apr 5, 2026
@@ -96,7 +96,7 @@ When processing "Incorporate Learnings" work, each action item from a LEARNINGS.
 
 ### 4. Cross-File Consistency
 
-When updating a count ("seven engagement modes" → "eight"), search for ALL occurrences across the codebase, not just the one you know about. When adding a table row, verify the row matches the column schema of existing rows. When adding an escalation path, verify both the "from" and "to" modes acknowledge the path.
+When updating a count ("eight engagement modes" → "nine"), search for ALL occurrences across the codebase, not just the one you know about. When adding a table row, verify the row matches the column schema of existing rows. When adding an escalation path, verify both the "from" and "to" modes acknowledge the path.
 
 Use `Grep` proactively to find all mentions of related terms before declaring a change complete.
 
@@ -295,7 +295,7 @@ Return findings in this structure:
 
 You MAY deviate without asking for:
 - **RULE 1:** Fixing obvious typos, broken links, or formatting inconsistencies in files you're already editing
-- **RULE 2:** Updating count words ("seven" → "eight") discovered via consistency checks, even if not explicitly scoped
+- **RULE 2:** Updating count words ("eight" → "nine") discovered via consistency checks, even if not explicitly scoped
 - **RULE 3:** Adding missing cross-references when a registration point clearly requires one
 
 You MUST ask before:
@@ -368,7 +368,7 @@ Awaiting guidance before proceeding.
 **This engineering task is INCOMPLETE if:**
 - Any mandatory integration checklist item is unaddressed
 - A new artifact is missing required template sections
-- Count words are stale (e.g., "seven modes" when eight exist)
+- Count words are stale (e.g., "eight modes" when nine exist)
 - Cross-references point to non-existent files or sections
 - The output report omits files that were created or modified
 

@@ -0,0 +1,162 @@
+#!/bin/bash
+# recover-session-logs.sh -- SessionStart hook: activity logging + crash recovery
+#
+# Replaces the previous inline SessionStart command. Performs:
+#   1. Original work: mkdir -p + activity.log append
+#   2. Background recovery: reconciles Claude Code's raw transcript storage
+#      against DAAF's archive directory, archiving any missed sessions
+#
+# Recovery approach:
+#   - Derives the Claude Code transcript directory from the current session's
+#     transcript_path (a common field available in all hook events)
+#   - Builds an index of already-archived session IDs with their file sizes
+#   - Scans for raw transcripts newer than the last recovery timestamp
+#   - For each unarchived (or grown) transcript, pipes a synthesized JSON
+#     payload to archive-session.sh, reusing all existing archiving logic
+#   - archive-session.sh's idempotency guard handles concurrent sessions:
+#     if a session is still running and its transcript grows, the next
+#     archive invocation (SessionEnd or recovery) will re-archive it
+#
+# Also processes stale pending log collection markers left by
+# collect_session_logs.sh if archive-session.sh didn't process them
+# (e.g., the session that wrote the marker crashed before SessionEnd).
+#
+# Performance:
+#   - Timestamp-gated: only processes transcripts modified since last recovery
+#   - Index-based matching: O(n+m) via bash associative array, not O(n*m) globs
+#   - Background execution: recovery runs in a detached subprocess so session
+#     startup is never blocked (foreground work completes in <1s)
+#
+# Exit codes:
+#   0 = always (observability hook, must never block session start)
+#
+# Hook event: SessionStart (matcher: "")
+# Registered in: .claude/settings.json
+
+# Fail OPEN: session start is observability, not a security gate
+trap '' ERR
+
+INPUT=$(cat)
+
+# Parse session metadata from JSON input
+mapfile -t _meta < <(
+    printf '%s' "$INPUT" | jq -r '
+        (.session_id // "unknown"),
+        (.transcript_path // ""),
+        (.cwd // "unknown")
+    ' 2>/dev/null
+)
+SESSION_ID="${_meta[0]:-unknown}"
+TRANSCRIPT_PATH="${_meta[1]:-}"
+CWD="${_meta[2]:-unknown}"
+
+PROJECT_DIR="${CLAUDE_PROJECT_DIR:-$(pwd)}"
+LOG_DIR="$PROJECT_DIR/.claude/logs"
+ARCHIVE_DIR="$LOG_DIR/sessions"
+
+# --- Section 1: Original SessionStart work ---
+mkdir -p "$LOG_DIR"
+mkdir -p "$ARCHIVE_DIR"
+DAAF_VERSION=$(git -C "$PROJECT_DIR" describe --always --dirty 2>/dev/null || echo "unknown")
+echo "Session started: $(date '+%Y-%m-%d %H:%M:%S') | DAAF: $DAAF_VERSION | Session: ${SESSION_ID:0:8}" >> "$LOG_DIR/activity.log"
+
+# --- Section 2: Background recovery ---
+# Only attempt recovery if we have a transcript_path to derive the source directory
+if [ -n "$TRANSCRIPT_PATH" ]; then
+    CURRENT_SHORT="${SESSION_ID:0:8}"
+
+    (
+        TRANSCRIPT_DIR=$(dirname "$TRANSCRIPT_PATH")
+
+        # Bail if transcript directory doesn't exist
+        [ -d "$TRANSCRIPT_DIR" ] || exit 0
+
+        # Build index of already-archived session shorts with their file sizes
+        # This is a single directory read -- O(m) where m = archived sessions
+        declare -A ARCHIVED_SIZE
+        for f in "$ARCHIVE_DIR"/*_orchestrator.jsonl; do
+            [ -f "$f" ] || continue
+            short=$(basename "$f" | sed 's/.*_\([a-f0-9]\{8\}\)_orchestrator\.jsonl/\1/')
+            [ -n "$short" ] || continue
+            ARCHIVED_SIZE["$short"]=$(stat -c%s "$f" 2>/dev/null || echo 0)
+        done
+
+        # Timestamp-gated scan: only process transcripts newer than last recovery
+        LAST_RECOVERY="$LOG_DIR/.last_recovery"
+        FIND_ARGS=(-maxdepth 1 -name "*.jsonl")
+        [ -f "$LAST_RECOVERY" ] && FIND_ARGS+=(-newer "$LAST_RECOVERY")
+
+        RECOVERED=0
+        SKIPPED=0
+
+        while IFS= read -r raw; do
+            [ -f "$raw" ] || continue
+            uuid=$(basename "$raw" .jsonl)
+            short="${uuid:0:8}"
+
+            # Skip the current session -- it just started
+            [ "$short" = "$CURRENT_SHORT" ] && continue
+
+            RAW_SIZE=$(stat -c%s "$raw" 2>/dev/null || echo 0)
+
+            # Skip if already archived with same or larger size
+            if [ -n "${ARCHIVED_SIZE[$short]+x}" ]; then
+                if [ "$RAW_SIZE" -le "${ARCHIVED_SIZE[$short]}" ] 2>/dev/null; then
+                    SKIPPED=$((SKIPPED + 1))
+                    continue
+                fi
+            fi
+
+            # Archive this session by piping synthesized JSON to archive-session.sh
+            # archive-session.sh's own idempotency guard provides a second safety net
+            printf '{"session_id":"%s","transcript_path":"%s","cwd":"%s","reason":"recovered"}' \
+                "$uuid" "$raw" "$PROJECT_DIR" \
+                | "$PROJECT_DIR/.claude/hooks/archive-session.sh" 2>/dev/null
+
+            RECOVERED=$((RECOVERED + 1))
+        done < <(find "$TRANSCRIPT_DIR" "${FIND_ARGS[@]}" 2>/dev/null)
+
+        # Update recovery timestamp
+        touch "$LAST_RECOVERY"
+
+        # Log recovery activity
+        if [ $RECOVERED -gt 0 ]; then
+            echo "$(date '+%Y-%m-%d %H:%M:%S') RECOVERY: archived $RECOVERED session(s), skipped $SKIPPED" >> "$LOG_DIR/activity.log"
+        fi
+
+        # --- Process stale pending log collection markers ---
+        PENDING_FILE="$LOG_DIR/pending_log_collection.jsonl"
+        if [ -f "$PENDING_FILE" ]; then
+            PENDING_TMP="${PENDING_FILE}.recovery.$$"
+            if mv "$PENDING_FILE" "$PENDING_TMP" 2>/dev/null; then
+                while IFS= read -r entry; do
+                    [ -z "$entry" ] && continue
+                    P_BASENAME=$(printf '%s' "$entry" | jq -r '.project_basename // empty' 2>/dev/null)
+                    P_PATH=$(printf '%s' "$entry" | jq -r '.project_path // empty' 2>/dev/null)
+                    [ -z "$P_BASENAME" ] || [ -z "$P_PATH" ] || [ ! -d "$P_PATH" ] && continue
+
+                    DEST="$P_PATH/logs"
+                    mkdir -p "$DEST"
+
+                    # Search all archived orchestrator transcripts for this project
+                    for arc in "$ARCHIVE_DIR"/*_orchestrator.jsonl; do
+                        [ -f "$arc" ] || continue
+                        if grep -q -- "$P_BASENAME" "$arc" 2>/dev/null; then
+                            arc_short=$(basename "$arc" | sed 's/.*_\([a-f0-9]\{8\}\)_orchestrator\.jsonl/\1/')
+                            for src in "$ARCHIVE_DIR"/*_${arc_short}_*.jsonl "$ARCHIVE_DIR"/*_${arc_short}_*.md; do
+                                [ -f "$src" ] || continue
+                                tgt="$DEST/$(basename "$src")"
+                                [ -f "$tgt" ] || cp "$src" "$tgt" 2>/dev/null
+                            done
+                        fi
+                    done
+                done < "$PENDING_TMP"
+                rm -f "$PENDING_TMP" 2>/dev/null
+            fi
+        fi
+
+    ) </dev/null >/dev/null 2>&1 &
+    disown
+fi
+
+exit 0
@@ -196,18 +196,6 @@
         ]
       }
     ],
-    "SubagentStop": [
-      {
-        "matcher": "",
-        "hooks": [
-          {
-            "type": "command",
-            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/subagent-registry.sh",
-            "timeout": 10
-          }
-        ]
-      }
-    ],
     "SessionEnd": [
       {
         "matcher": "",
@@ -226,7 +214,7 @@
         "hooks": [
           {
             "type": "command",
-            "command": "mkdir -p \"$CLAUDE_PROJECT_DIR\"/.claude/logs && echo \"Session started: $(date '+%Y-%m-%d %H:%M:%S') | DAAF: $(git -C \"$CLAUDE_PROJECT_DIR\" describe --always --dirty 2>/dev/null || echo 'unknown')\" >> \"$CLAUDE_PROJECT_DIR\"/.claude/logs/activity.log",
+            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/recover-session-logs.sh",
             "timeout": 5
           }
         ]