aw-ecc 1.4.21

package/skills/continuous-learning-v2/hooks/observe.sh

@@ -0,0 +1,412 @@
+#!/bin/bash
+# Continuous Learning v2 - Observation Hook
+#
+# Captures tool use events for pattern analysis.
+# Claude Code passes hook data via stdin as JSON.
+#
+# v2.1: Project-scoped observations — detects current project context
+#       and writes observations to project-specific directory.
+#
+# Registered via plugin hooks/hooks.json (auto-loaded when plugin is enabled).
+# Can also be registered manually in ~/.claude/settings.json.
+
+set -e
+
+# Hook phase from CLI argument: "pre" (PreToolUse) or "post" (PostToolUse)
+HOOK_PHASE="${1:-post}"
+
+# ─────────────────────────────────────────────
+# Read stdin first (before project detection)
+# ─────────────────────────────────────────────
+
+# Read JSON from stdin (Claude Code hook format)
+INPUT_JSON=$(cat)
+
+# Exit if no input
+if [ -z "$INPUT_JSON" ]; then
+  exit 0
+fi
+
+resolve_python_cmd() {
+  if [ -n "${CLV2_PYTHON_CMD:-}" ] && command -v "$CLV2_PYTHON_CMD" >/dev/null 2>&1; then
+    printf '%s\n' "$CLV2_PYTHON_CMD"
+    return 0
+  fi
+
+  if command -v python3 >/dev/null 2>&1; then
+    printf '%s\n' python3
+    return 0
+  fi
+
+  if command -v python >/dev/null 2>&1; then
+    printf '%s\n' python
+    return 0
+  fi
+
+  return 1
+}
+
+PYTHON_CMD="$(resolve_python_cmd 2>/dev/null || true)"
+if [ -z "$PYTHON_CMD" ]; then
+  echo "[observe] No python interpreter found, skipping observation" >&2
+  exit 0
+fi
+
+# ─────────────────────────────────────────────
+# Extract cwd from stdin for project detection
+# ─────────────────────────────────────────────
+
+# Extract cwd from the hook JSON to use for project detection.
+# This avoids spawning a separate git subprocess when cwd is available.
+STDIN_CWD=$(echo "$INPUT_JSON" | "$PYTHON_CMD" -c '
+import json, sys
+try:
+    data = json.load(sys.stdin)
+    cwd = data.get("cwd", "")
+    print(cwd)
+except(KeyError, TypeError, ValueError):
+    print("")
+' 2>/dev/null || echo "")
+
+# If cwd was provided in stdin, use it for project detection
+if [ -n "$STDIN_CWD" ] && [ -d "$STDIN_CWD" ]; then
+  export CLAUDE_PROJECT_DIR="$STDIN_CWD"
+fi
+
+# ─────────────────────────────────────────────
+# Lightweight config and automated session guards
+# ─────────────────────────────────────────────
+#
+# IMPORTANT: keep these guards above detect-project.sh.
+# Sourcing detect-project.sh creates project-scoped directories and updates
+# projects.json, so automated sessions must return before that point.
+
+CONFIG_DIR="${HOME}/.claude/homunculus"
+
+# Skip if disabled (check both default and CLV2_CONFIG-derived locations)
+if [ -f "$CONFIG_DIR/disabled" ]; then
+  exit 0
+fi
+if [ -n "${CLV2_CONFIG:-}" ] && [ -f "$(dirname "$CLV2_CONFIG")/disabled" ]; then
+  exit 0
+fi
+
+# Prevent observe.sh from firing on non-human sessions to avoid:
+#   - ECC observing its own Haiku observer sessions (self-loop)
+#   - ECC observing other tools' automated sessions
+#   - automated sessions creating project-scoped homunculus metadata
+
+# Layer 1: entrypoint. Only interactive terminal sessions should continue.
+# sdk-ts: Agent SDK sessions can be human-interactive (e.g. via Happy).
+# Non-interactive SDK automation is still filtered by Layers 2-5 below
+# (ECC_HOOK_PROFILE=minimal, ECC_SKIP_OBSERVE=1, agent_id, path exclusions).
+case "${CLAUDE_CODE_ENTRYPOINT:-cli}" in
+  cli|sdk-ts) ;;
+  *) exit 0 ;;
+esac
+
+# Layer 2: minimal hook profile suppresses non-essential hooks.
+[ "${ECC_HOOK_PROFILE:-standard}" = "minimal" ] && exit 0
+
+# Layer 3: cooperative skip env var for automated sessions.
+[ "${ECC_SKIP_OBSERVE:-0}" = "1" ] && exit 0
+
+# Layer 4: subagent sessions are automated by definition.
+_ECC_AGENT_ID=$(echo "$INPUT_JSON" | "$PYTHON_CMD" -c "import json,sys; print(json.load(sys.stdin).get('agent_id',''))" 2>/dev/null || true)
+[ -n "$_ECC_AGENT_ID" ] && exit 0
+
+# Layer 5: known observer-session path exclusions.
+_ECC_SKIP_PATHS="${ECC_OBSERVE_SKIP_PATHS:-observer-sessions,.claude-mem}"
+if [ -n "$STDIN_CWD" ]; then
+  IFS=',' read -ra _ECC_SKIP_ARRAY <<< "$_ECC_SKIP_PATHS"
+  for _pattern in "${_ECC_SKIP_ARRAY[@]}"; do
+    _pattern="${_pattern#"${_pattern%%[![:space:]]*}"}"
+    _pattern="${_pattern%"${_pattern##*[![:space:]]}"}"
+    [ -z "$_pattern" ] && continue
+    case "$STDIN_CWD" in *"$_pattern"*) exit 0 ;; esac
+  done
+fi
+
+# ─────────────────────────────────────────────
+# Project detection
+# ─────────────────────────────────────────────
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+SKILL_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
+
+# Source shared project detection helper
+# This sets: PROJECT_ID, PROJECT_NAME, PROJECT_ROOT, PROJECT_DIR
+source "${SKILL_ROOT}/scripts/detect-project.sh"
+PYTHON_CMD="${CLV2_PYTHON_CMD:-$PYTHON_CMD}"
+
+# ─────────────────────────────────────────────
+# Configuration
+# ─────────────────────────────────────────────
+
+OBSERVATIONS_FILE="${PROJECT_DIR}/observations.jsonl"
+MAX_FILE_SIZE_MB=10
+
+# Auto-purge observation files older than 30 days (runs once per session)
+PURGE_MARKER="${PROJECT_DIR}/.last-purge"
+if [ ! -f "$PURGE_MARKER" ] || [ "$(find "$PURGE_MARKER" -mtime +1 2>/dev/null)" ]; then
+  find "${PROJECT_DIR}" -name "observations-*.jsonl" -mtime +30 -delete 2>/dev/null || true
+  touch "$PURGE_MARKER" 2>/dev/null || true
+fi
+
+# Parse using Python via stdin pipe (safe for all JSON payloads)
+# Pass HOOK_PHASE via env var since Claude Code does not include hook type in stdin JSON
+PARSED=$(echo "$INPUT_JSON" | HOOK_PHASE="$HOOK_PHASE" "$PYTHON_CMD" -c '
+import json
+import sys
+import os
+
+try:
+    data = json.load(sys.stdin)
+
+    # Determine event type from CLI argument passed via env var.
+    # Claude Code does NOT include a "hook_type" field in the stdin JSON,
+    # so we rely on the shell argument ("pre" or "post") instead.
+    hook_phase = os.environ.get("HOOK_PHASE", "post")
+    event = "tool_start" if hook_phase == "pre" else "tool_complete"
+
+    # Extract fields - Claude Code hook format
+    tool_name = data.get("tool_name", data.get("tool", "unknown"))
+    tool_input = data.get("tool_input", data.get("input", {}))
+    tool_output = data.get("tool_response")
+    if tool_output is None:
+        tool_output = data.get("tool_output", data.get("output", ""))
+    session_id = data.get("session_id", "unknown")
+    tool_use_id = data.get("tool_use_id", "")
+    cwd = data.get("cwd", "")
+
+    # Truncate large inputs/outputs
+    if isinstance(tool_input, dict):
+        tool_input_str = json.dumps(tool_input)[:5000]
+    else:
+        tool_input_str = str(tool_input)[:5000]
+
+    if isinstance(tool_output, dict):
+        tool_response_str = json.dumps(tool_output)[:5000]
+    else:
+        tool_response_str = str(tool_output)[:5000]
+
+    print(json.dumps({
+        "parsed": True,
+        "event": event,
+        "tool": tool_name,
+        "input": tool_input_str if event == "tool_start" else None,
+        "output": tool_response_str if event == "tool_complete" else None,
+        "session": session_id,
+        "tool_use_id": tool_use_id,
+        "cwd": cwd
+    }))
+except Exception as e:
+    print(json.dumps({"parsed": False, "error": str(e)}))
+')
+
+# Check if parsing succeeded
+PARSED_OK=$(echo "$PARSED" | "$PYTHON_CMD" -c "import json,sys; print(json.load(sys.stdin).get('parsed', False))" 2>/dev/null || echo "False")
+
+if [ "$PARSED_OK" != "True" ]; then
+  # Fallback: log raw input for debugging (scrub secrets before persisting)
+  timestamp=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
+  export TIMESTAMP="$timestamp"
+  echo "$INPUT_JSON" | "$PYTHON_CMD" -c '
+import json, sys, os, re
+
+_SECRET_RE = re.compile(
+    r"(?i)(api[_-]?key|token|secret|password|authorization|credentials?|auth)"
+    r"""(["'"'"'\s:=]+)"""
+    r"([A-Za-z]+\s+)?"
+    r"([A-Za-z0-9_\-/.+=]{8,})"
+)
+
+raw = sys.stdin.read()[:2000]
+raw = _SECRET_RE.sub(lambda m: m.group(1) + m.group(2) + (m.group(3) or "") + "[REDACTED]", raw)
+print(json.dumps({"timestamp": os.environ["TIMESTAMP"], "event": "parse_error", "raw": raw}))
+' >> "$OBSERVATIONS_FILE"
+  exit 0
+fi
+
+# Archive if file too large (atomic: rename with unique suffix to avoid race)
+if [ -f "$OBSERVATIONS_FILE" ]; then
+  file_size_mb=$(du -m "$OBSERVATIONS_FILE" 2>/dev/null | cut -f1)
+  if [ "${file_size_mb:-0}" -ge "$MAX_FILE_SIZE_MB" ]; then
+    archive_dir="${PROJECT_DIR}/observations.archive"
+    mkdir -p "$archive_dir"
+    mv "$OBSERVATIONS_FILE" "$archive_dir/observations-$(date +%Y%m%d-%H%M%S)-$$.jsonl" 2>/dev/null || true
+  fi
+fi
+
+# Build and write observation (now includes project context)
+# Scrub common secret patterns from tool I/O before persisting
+timestamp=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
+
+export PROJECT_ID_ENV="$PROJECT_ID"
+export PROJECT_NAME_ENV="$PROJECT_NAME"
+export TIMESTAMP="$timestamp"
+
+echo "$PARSED" | "$PYTHON_CMD" -c '
+import json, sys, os, re
+
+parsed = json.load(sys.stdin)
+observation = {
+    "timestamp": os.environ["TIMESTAMP"],
+    "event": parsed["event"],
+    "tool": parsed["tool"],
+    "session": parsed["session"],
+    "project_id": os.environ.get("PROJECT_ID_ENV", "global"),
+    "project_name": os.environ.get("PROJECT_NAME_ENV", "global")
+}
+
+# Scrub secrets: match common key=value, key: value, and key"value patterns
+# Includes optional auth scheme (e.g., "Bearer", "Basic") before token
+_SECRET_RE = re.compile(
+    r"(?i)(api[_-]?key|token|secret|password|authorization|credentials?|auth)"
+    r"""(["'"'"'\s:=]+)"""
+    r"([A-Za-z]+\s+)?"
+    r"([A-Za-z0-9_\-/.+=]{8,})"
+)
+
+def scrub(val):
+    if val is None:
+        return None
+    return _SECRET_RE.sub(lambda m: m.group(1) + m.group(2) + (m.group(3) or "") + "[REDACTED]", str(val))
+
+if parsed["input"]:
+    observation["input"] = scrub(parsed["input"])
+if parsed["output"] is not None:
+    observation["output"] = scrub(parsed["output"])
+
+print(json.dumps(observation))
+' >> "$OBSERVATIONS_FILE"
+
+# Lazy-start observer if enabled but not running (first-time setup)
+# Use flock for atomic check-then-act to prevent race conditions
+# Fallback for macOS (no flock): use lockfile or skip
+LAZY_START_LOCK="${PROJECT_DIR}/.observer-start.lock"
+_CHECK_OBSERVER_RUNNING() {
+  local pid_file="$1"
+  if [ -f "$pid_file" ]; then
+    local pid
+    pid=$(cat "$pid_file" 2>/dev/null)
+    # Validate PID is a positive integer (>1) to prevent signaling invalid targets
+    case "$pid" in
+      ''|*[!0-9]*|0|1)
+        rm -f "$pid_file" 2>/dev/null || true
+        return 1
+        ;;
+    esac
+    if kill -0 "$pid" 2>/dev/null; then
+      return 0  # Process is alive
+    fi
+    # Stale PID file - remove it
+    rm -f "$pid_file" 2>/dev/null || true
+  fi
+  return 1  # No PID file or process dead
+}
+
+if [ -f "${CONFIG_DIR}/disabled" ]; then
+  OBSERVER_ENABLED=false
+else
+  OBSERVER_ENABLED=false
+  CONFIG_FILE="${SKILL_ROOT}/config.json"
+  # Allow CLV2_CONFIG override
+  if [ -n "${CLV2_CONFIG:-}" ]; then
+    CONFIG_FILE="$CLV2_CONFIG"
+  fi
+  # Use effective config path for both existence check and reading
+  EFFECTIVE_CONFIG="$CONFIG_FILE"
+  if [ -f "$EFFECTIVE_CONFIG" ] && [ -n "$PYTHON_CMD" ]; then
+    _enabled=$(CLV2_CONFIG_PATH="$EFFECTIVE_CONFIG" "$PYTHON_CMD" -c "
+import json, os
+with open(os.environ['CLV2_CONFIG_PATH']) as f:
+    cfg = json.load(f)
+print(str(cfg.get('observer', {}).get('enabled', False)).lower())
+" 2>/dev/null || echo "false")
+    if [ "$_enabled" = "true" ]; then
+      OBSERVER_ENABLED=true
+    fi
+  fi
+fi
+
+# Check both project-scoped AND global PID files (with stale PID recovery)
+if [ "$OBSERVER_ENABLED" = "true" ]; then
+  # Clean up stale PID files first
+  _CHECK_OBSERVER_RUNNING "${PROJECT_DIR}/.observer.pid" || true
+  _CHECK_OBSERVER_RUNNING "${CONFIG_DIR}/.observer.pid" || true
+
+  # Check if observer is now running after cleanup
+  if [ ! -f "${PROJECT_DIR}/.observer.pid" ] && [ ! -f "${CONFIG_DIR}/.observer.pid" ]; then
+    # Use flock if available (Linux), fallback for macOS
+    if command -v flock >/dev/null 2>&1; then
+      (
+        flock -n 9 || exit 0
+        # Double-check PID files after acquiring lock
+        _CHECK_OBSERVER_RUNNING "${PROJECT_DIR}/.observer.pid" || true
+        _CHECK_OBSERVER_RUNNING "${CONFIG_DIR}/.observer.pid" || true
+        if [ ! -f "${PROJECT_DIR}/.observer.pid" ] && [ ! -f "${CONFIG_DIR}/.observer.pid" ]; then
+          nohup "${SKILL_ROOT}/agents/start-observer.sh" start >/dev/null 2>&1 &
+        fi
+      ) 9>"$LAZY_START_LOCK"
+    else
+      # macOS fallback: use lockfile if available, otherwise skip
+      if command -v lockfile >/dev/null 2>&1; then
+        # Use subshell to isolate exit and add trap for cleanup
+        (
+          trap 'rm -f "$LAZY_START_LOCK" 2>/dev/null || true' EXIT
+          lockfile -r 1 -l 30 "$LAZY_START_LOCK" 2>/dev/null || exit 0
+          _CHECK_OBSERVER_RUNNING "${PROJECT_DIR}/.observer.pid" || true
+          _CHECK_OBSERVER_RUNNING "${CONFIG_DIR}/.observer.pid" || true
+          if [ ! -f "${PROJECT_DIR}/.observer.pid" ] && [ ! -f "${CONFIG_DIR}/.observer.pid" ]; then
+            nohup "${SKILL_ROOT}/agents/start-observer.sh" start >/dev/null 2>&1 &
+          fi
+          rm -f "$LAZY_START_LOCK" 2>/dev/null || true
+        )
+      fi
+    fi
+  fi
+fi
+
+# Throttle SIGUSR1: only signal observer every N observations (#521)
+# This prevents rapid signaling when tool calls fire every second,
+# which caused runaway parallel Claude analysis processes.
+SIGNAL_EVERY_N="${ECC_OBSERVER_SIGNAL_EVERY_N:-20}"
+SIGNAL_COUNTER_FILE="${PROJECT_DIR}/.observer-signal-counter"
+
+should_signal=0
+if [ -f "$SIGNAL_COUNTER_FILE" ]; then
+  counter=$(cat "$SIGNAL_COUNTER_FILE" 2>/dev/null || echo 0)
+  counter=$((counter + 1))
+  if [ "$counter" -ge "$SIGNAL_EVERY_N" ]; then
+    should_signal=1
+    counter=0
+  fi
+  echo "$counter" > "$SIGNAL_COUNTER_FILE"
+else
+  echo "1" > "$SIGNAL_COUNTER_FILE"
+fi
+
+# Signal observer if running and throttle allows (check both project-scoped and global observer, deduplicate)
+if [ "$should_signal" -eq 1 ]; then
+  signaled_pids=" "
+  for pid_file in "${PROJECT_DIR}/.observer.pid" "${CONFIG_DIR}/.observer.pid"; do
+    if [ -f "$pid_file" ]; then
+      observer_pid=$(cat "$pid_file" 2>/dev/null || true)
+      # Validate PID is a positive integer (>1)
+      case "$observer_pid" in
+        ''|*[!0-9]*|0|1) rm -f "$pid_file" 2>/dev/null || true; continue ;;
+      esac
+      # Deduplicate: skip if already signaled this pass
+      case "$signaled_pids" in
+        *" $observer_pid "*) continue ;;
+      esac
+      if kill -0 "$observer_pid" 2>/dev/null; then
+        kill -USR1 "$observer_pid" 2>/dev/null || true
+        signaled_pids="${signaled_pids}${observer_pid} "
+      fi
+    fi
+  done
+fi
+
+exit 0

package/skills/continuous-learning-v2/scripts/detect-project.sh

@@ -0,0 +1,228 @@
+#!/bin/bash
+# Continuous Learning v2 - Project Detection Helper
+#
+# Shared logic for detecting current project context.
+# Sourced by observe.sh and start-observer.sh.
+#
+# Exports:
+#   _CLV2_PROJECT_ID     - Short hash identifying the project (or "global")
+#   _CLV2_PROJECT_NAME   - Human-readable project name
+#   _CLV2_PROJECT_ROOT   - Absolute path to project root
+#   _CLV2_PROJECT_DIR    - Project-scoped storage directory under homunculus
+#
+# Also sets unprefixed convenience aliases:
+#   PROJECT_ID, PROJECT_NAME, PROJECT_ROOT, PROJECT_DIR
+#
+# Detection priority:
+#   1. CLAUDE_PROJECT_DIR env var (if set)
+#   2. git remote URL (hashed for uniqueness across machines)
+#   3. git repo root path (fallback, machine-specific)
+#   4. "global" (no project context detected)
+
+_CLV2_HOMUNCULUS_DIR="${HOME}/.claude/homunculus"
+_CLV2_PROJECTS_DIR="${_CLV2_HOMUNCULUS_DIR}/projects"
+_CLV2_REGISTRY_FILE="${_CLV2_HOMUNCULUS_DIR}/projects.json"
+
+_clv2_resolve_python_cmd() {
+  if [ -n "${CLV2_PYTHON_CMD:-}" ] && command -v "$CLV2_PYTHON_CMD" >/dev/null 2>&1; then
+    printf '%s\n' "$CLV2_PYTHON_CMD"
+    return 0
+  fi
+
+  if command -v python3 >/dev/null 2>&1; then
+    printf '%s\n' python3
+    return 0
+  fi
+
+  if command -v python >/dev/null 2>&1; then
+    printf '%s\n' python
+    return 0
+  fi
+
+  return 1
+}
+
+_CLV2_PYTHON_CMD="$(_clv2_resolve_python_cmd 2>/dev/null || true)"
+CLV2_PYTHON_CMD="$_CLV2_PYTHON_CMD"
+export CLV2_PYTHON_CMD
+
+CLV2_OBSERVER_PROMPT_PATTERN='Can you confirm|requires permission|Awaiting (user confirmation|confirmation|approval|permission)|confirm I should proceed|once granted access|grant.*access'
+export CLV2_OBSERVER_PROMPT_PATTERN
+
+_clv2_detect_project() {
+  local project_root=""
+  local project_name=""
+  local project_id=""
+  local source_hint=""
+
+  # 1. Try CLAUDE_PROJECT_DIR env var
+  if [ -n "$CLAUDE_PROJECT_DIR" ] && [ -d "$CLAUDE_PROJECT_DIR" ]; then
+    project_root="$CLAUDE_PROJECT_DIR"
+    source_hint="env"
+  fi
+
+  # 2. Try git repo root from CWD (only if git is available)
+  if [ -z "$project_root" ] && command -v git &>/dev/null; then
+    project_root=$(git rev-parse --show-toplevel 2>/dev/null || true)
+    if [ -n "$project_root" ]; then
+      source_hint="git"
+    fi
+  fi
+
+  # 3. No project detected — fall back to global
+  if [ -z "$project_root" ]; then
+    _CLV2_PROJECT_ID="global"
+    _CLV2_PROJECT_NAME="global"
+    _CLV2_PROJECT_ROOT=""
+    _CLV2_PROJECT_DIR="${_CLV2_HOMUNCULUS_DIR}"
+    return 0
+  fi
+
+  # Derive project name from directory basename
+  project_name=$(basename "$project_root")
+
+  # Derive project ID: prefer git remote URL hash (portable across machines),
+  # fall back to path hash (machine-specific but still useful)
+  local remote_url=""
+  if command -v git &>/dev/null; then
+    if [ "$source_hint" = "git" ] || [ -e "${project_root}/.git" ]; then
+      remote_url=$(git -C "$project_root" remote get-url origin 2>/dev/null || true)
+    fi
+  fi
+
+  # Compute hash from the original remote URL (legacy, for backward compatibility)
+  local legacy_hash_input="${remote_url:-$project_root}"
+
+  # Strip embedded credentials from remote URL (e.g., https://ghp_xxxx@github.com/...)
+  if [ -n "$remote_url" ]; then
+    remote_url=$(printf '%s' "$remote_url" | sed -E 's|://[^@]+@|://|')
+  fi
+
+  local hash_input="${remote_url:-$project_root}"
+  # Prefer Python for consistent SHA256 behavior across shells/platforms.
+  if [ -n "$_CLV2_PYTHON_CMD" ]; then
+    project_id=$(printf '%s' "$hash_input" | "$_CLV2_PYTHON_CMD" -c "import sys,hashlib; print(hashlib.sha256(sys.stdin.buffer.read()).hexdigest()[:12])" 2>/dev/null)
+  fi
+
+  # Fallback if Python is unavailable or hash generation failed.
+  if [ -z "$project_id" ]; then
+    project_id=$(printf '%s' "$hash_input" | shasum -a 256 2>/dev/null | cut -c1-12 || \
+                 printf '%s' "$hash_input" | sha256sum 2>/dev/null | cut -c1-12 || \
+                 echo "fallback")
+  fi
+
+  # Backward compatibility: if credentials were stripped and the hash changed,
+  # check if a project dir exists under the legacy hash and reuse it
+  if [ "$legacy_hash_input" != "$hash_input" ] && [ -n "$_CLV2_PYTHON_CMD" ]; then
+    local legacy_id=""
+    legacy_id=$(printf '%s' "$legacy_hash_input" | "$_CLV2_PYTHON_CMD" -c "import sys,hashlib; print(hashlib.sha256(sys.stdin.buffer.read()).hexdigest()[:12])" 2>/dev/null)
+    if [ -n "$legacy_id" ] && [ -d "${_CLV2_PROJECTS_DIR}/${legacy_id}" ] && [ ! -d "${_CLV2_PROJECTS_DIR}/${project_id}" ]; then
+      # Migrate legacy directory to new hash
+      mv "${_CLV2_PROJECTS_DIR}/${legacy_id}" "${_CLV2_PROJECTS_DIR}/${project_id}" 2>/dev/null || project_id="$legacy_id"
+    fi
+  fi
+
+  # Export results
+  _CLV2_PROJECT_ID="$project_id"
+  _CLV2_PROJECT_NAME="$project_name"
+  _CLV2_PROJECT_ROOT="$project_root"
+  _CLV2_PROJECT_DIR="${_CLV2_PROJECTS_DIR}/${project_id}"
+
+  # Ensure project directory structure exists
+  mkdir -p "${_CLV2_PROJECT_DIR}/instincts/personal"
+  mkdir -p "${_CLV2_PROJECT_DIR}/instincts/inherited"
+  mkdir -p "${_CLV2_PROJECT_DIR}/observations.archive"
+  mkdir -p "${_CLV2_PROJECT_DIR}/evolved/skills"
+  mkdir -p "${_CLV2_PROJECT_DIR}/evolved/commands"
+  mkdir -p "${_CLV2_PROJECT_DIR}/evolved/agents"
+
+  # Update project registry (lightweight JSON mapping)
+  _clv2_update_project_registry "$project_id" "$project_name" "$project_root" "$remote_url"
+}
+
+_clv2_update_project_registry() {
+  local pid="$1"
+  local pname="$2"
+  local proot="$3"
+  local premote="$4"
+  local pdir="$_CLV2_PROJECT_DIR"
+
+  mkdir -p "$(dirname "$_CLV2_REGISTRY_FILE")"
+
+  if [ -z "$_CLV2_PYTHON_CMD" ]; then
+    return 0
+  fi
+
+  # Pass values via env vars to avoid shell→python injection.
+  # Python reads them with os.environ, which is safe for any string content.
+  _CLV2_REG_PID="$pid" \
+  _CLV2_REG_PNAME="$pname" \
+  _CLV2_REG_PROOT="$proot" \
+  _CLV2_REG_PREMOTE="$premote" \
+  _CLV2_REG_PDIR="$pdir" \
+  _CLV2_REG_FILE="$_CLV2_REGISTRY_FILE" \
+  "$_CLV2_PYTHON_CMD" -c '
+import json, os, tempfile
+from datetime import datetime, timezone
+
+registry_path = os.environ["_CLV2_REG_FILE"]
+project_dir = os.environ["_CLV2_REG_PDIR"]
+project_file = os.path.join(project_dir, "project.json")
+
+os.makedirs(project_dir, exist_ok=True)
+
+def atomic_write_json(path, payload):
+    fd, tmp_path = tempfile.mkstemp(
+        prefix=f".{os.path.basename(path)}.tmp.",
+        dir=os.path.dirname(path),
+        text=True,
+    )
+    try:
+        with os.fdopen(fd, "w") as f:
+            json.dump(payload, f, indent=2)
+            f.write("\n")
+        os.replace(tmp_path, path)
+    finally:
+        if os.path.exists(tmp_path):
+            os.unlink(tmp_path)
+
+try:
+    with open(registry_path) as f:
+        registry = json.load(f)
+except (FileNotFoundError, json.JSONDecodeError):
+    registry = {}
+
+now = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
+entry = registry.get(os.environ["_CLV2_REG_PID"], {})
+
+metadata = {
+    "id": os.environ["_CLV2_REG_PID"],
+    "name": os.environ["_CLV2_REG_PNAME"],
+    "root": os.environ["_CLV2_REG_PROOT"],
+    "remote": os.environ["_CLV2_REG_PREMOTE"],
+    "created_at": entry.get("created_at", now),
+    "last_seen": now,
+}
+
+registry[os.environ["_CLV2_REG_PID"]] = metadata
+
+atomic_write_json(project_file, metadata)
+atomic_write_json(registry_path, registry)
+' 2>/dev/null || true
+}
+
+# Auto-detect on source
+_clv2_detect_project
+
+# Convenience aliases for callers (short names pointing to prefixed vars)
+PROJECT_ID="$_CLV2_PROJECT_ID"
+PROJECT_NAME="$_CLV2_PROJECT_NAME"
+PROJECT_ROOT="$_CLV2_PROJECT_ROOT"
+PROJECT_DIR="$_CLV2_PROJECT_DIR"
+
+if [ -n "$PROJECT_ROOT" ]; then
+  CLV2_OBSERVER_SENTINEL_FILE="${PROJECT_ROOT}/.observer.lock"
+else
+  CLV2_OBSERVER_SENTINEL_FILE="${PROJECT_DIR}/.observer.lock"
+fi
+export CLV2_OBSERVER_SENTINEL_FILE