feat(ralph): smart gates + DAG + dashboard live (W3)

Restructurare Ralph QC loop pe smart gate dispatcher tag-driven (în loc de 5 faze fixe), DAG dependsOn cu propagare blocked, retry guard 3-strike, rate limit detection, plus dashboard live cu polling 5s. Changes: - tools/ralph_prd_generator.py: parametru optional final_plan_path; când e furnizat, invocă Claude Opus pe final-plan.md pentru extragere user stories cu schema extinsă (tags, dependsOn, acceptanceCriteria 3-5). Backward compat păstrat — fără final_plan_path, fallback la heuristic-ul vechi. - tools/ralph/prd-template.json: schema W3 (tags[], dependsOn[], retries, failed, blocked, failureReason, requiresDesignReview). - tools/ralph/prompt.md: 4 faze (impl, base quality, smart gates, commit) + dispatcher pe story.tags. Tags vide → run-all-gates fallback (safe default). - tools/ralph_dag.py (nou): tag validation heuristic anti-silent-regression (force ui dacă diff atinge .vue/.tsx/.html/.css/.scss; force db pentru migrations sau .sql; force vercel dacă există vercel.json) + topological sort cu blocked propagation + atomic prd.json updates. - tools/ralph/ralph.sh: --max-turns 30, DAG-aware story selection, retry counter cu auto-fail la 3, rate limit detection (sleep 30min + 1 retry), CLI subcommands prin tools/ralph_dag.py helper. - dashboard/handlers/ralph.py (nou): /api/ralph/status + /<slug>/log + /prd + /stop. Defensive vs corrupt prd.json. Sandbox-ed PID kill. - dashboard/ralph.html (nou): live cards 3/2/1 col responsive, polling 5s, drawer pentru log/PRD viewer, status colors (--status-running/blocked/ failed/complete declarate inline), Lucide icons cu aria-labels. - dashboard/api.py: mount /api/ralph/* (GET status/log/prd, POST stop). - tests/: 72 teste noi (smart gates, DAG, retry, dashboard endpoint). Note arhitecturale: - Polling 5s ales peste SSE/WebSocket (suficient pentru iter Ralph 8-15min) - Tag validation rulează POST-iter pe diff git pentru anti-silent-regression - Rate limit retry: 1 dată per rulare, apoi mark failed=rate_limited Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-26 18:36:35 +00:00
parent e06a79d98c
commit 655ed3ae09
11 changed files with 2282 additions and 189 deletions
--- a/tools/ralph/ralph.sh
+++ b/tools/ralph/ralph.sh
@@ -1,7 +1,13 @@
 #!/bin/bash
-# Ralph pentru Claude Code - Loop autonom de agent AI
+# Ralph pentru Claude Code - Loop autonom de agent AI (W3: smart gates + DAG + rate limit)
 # Adaptat din Ralph original (snarktank/ralph) pentru Claude Code CLI
 # Usage: ./ralph.sh [max_iterations] [project_dir]
+#
+# Env vars (opționale):
+#   RALPH_MAX_TURNS        — --max-turns per iter (default 30)
+#   RALPH_RATE_LIMIT_SLEEP — sleep după rate limit detection (default 1800 = 30min)
+#   RALPH_DAG_HELPER       — path la tools/ralph_dag.py (auto-detect default)
+#   RALPH_PYTHON           — interpreter Python pentru DAG helper (default python3)

 set -e

@@ -15,6 +21,24 @@ SCREENSHOTS_DIR="$SCRIPT_DIR/screenshots"
 LAST_BRANCH_FILE="$SCRIPT_DIR/.last-branch"
 PROMPT_FILE="$SCRIPT_DIR/prompt.md"

+# W3 config
+MAX_TURNS=${RALPH_MAX_TURNS:-30}
+RATE_LIMIT_SLEEP=${RALPH_RATE_LIMIT_SLEEP:-1800}
+RALPH_PYTHON=${RALPH_PYTHON:-python3}
+
+# DAG helper auto-detect: prefer co-located cu echo-core; fallback la $SCRIPT_DIR
+if [ -n "$RALPH_DAG_HELPER" ] && [ -f "$RALPH_DAG_HELPER" ]; then
+  DAG_HELPER="$RALPH_DAG_HELPER"
+elif [ -f "/home/moltbot/echo-core/tools/ralph_dag.py" ]; then
+  DAG_HELPER="/home/moltbot/echo-core/tools/ralph_dag.py"
+elif [ -f "/home/moltbot/echo-core-qc/tools/ralph_dag.py" ]; then
+  DAG_HELPER="/home/moltbot/echo-core-qc/tools/ralph_dag.py"
+elif [ -f "$SCRIPT_DIR/ralph_dag.py" ]; then
+  DAG_HELPER="$SCRIPT_DIR/ralph_dag.py"
+else
+  DAG_HELPER=""
+fi
+
 # Verifică că jq este instalat
 if ! command -v jq &> /dev/null; then
    echo "Eroare: jq nu este instalat. Rulează: apt install jq"
@@ -132,6 +156,51 @@ check_all_complete() {
  [ "$incomplete" -eq 0 ]
 }

+# W3: nimic eligibil = toate sunt fie passes, fie failed, fie blocked
+check_no_eligible() {
+  local n=$(jq '[.userStories[] | select(.passes != true and .failed != true and .blocked != true)] | length' "$PRD_FILE" 2>/dev/null || echo "999")
+  [ "$n" -eq 0 ]
+}
+
+# W3: alege next eligible story via DAG helper. Print story ID sau "" dacă nimic.
+dag_next_story() {
+  if [ -n "$DAG_HELPER" ]; then
+    "$RALPH_PYTHON" "$DAG_HELPER" next-story "$PRD_FILE" 2>/dev/null || echo ""
+  else
+    # Fallback simplu (fără DAG): primul story cu passes!=true && failed!=true && blocked!=true, priority asc
+    jq -r '[.userStories[] | select(.passes != true and .failed != true and .blocked != true)] | sort_by(.priority) | .[0].id // ""' "$PRD_FILE"
+  fi
+}
+
+dag_incr_retry() {
+  local sid="$1"
+  if [ -n "$DAG_HELPER" ]; then
+    "$RALPH_PYTHON" "$DAG_HELPER" incr-retry "$PRD_FILE" "$sid" 2>/dev/null || echo "0"
+  else
+    echo "0"
+  fi
+}
+
+dag_mark_failed() {
+  local sid="$1" reason="$2"
+  if [ -n "$DAG_HELPER" ]; then
+    "$RALPH_PYTHON" "$DAG_HELPER" mark-failed "$PRD_FILE" "$sid" "$reason" 2>/dev/null || true
+  fi
+}
+
+dag_force_tags() {
+  local sid="$1"
+  if [ -n "$DAG_HELPER" ]; then
+    "$RALPH_PYTHON" "$DAG_HELPER" force-tags "$PRD_FILE" "$sid" "$PROJECT_DIR" 2>/dev/null || true
+  fi
+}
+
+# W3: detectează rate limit în output Claude (heuristic — Anthropic nu are exit code dedicat)
+is_rate_limited() {
+  local output="$1"
+  echo "$output" | grep -qiE "rate limit|rate_limit_exceeded|429|too many requests"
+}
+
 # Afișare status inițial
 echo ""
 echo "======================================================================="
@@ -155,6 +224,9 @@ if check_all_complete; then
  exit 0
 fi

+# Tracker pentru rate limit retry (max 1 retry de iterație-rate-limit per rulare)
+RATE_LIMIT_RETRY_USED=0
+
 # Loop principal
 for i in $(seq 1 $MAX_ITERATIONS); do
  echo ""
@@ -162,17 +234,38 @@ for i in $(seq 1 $MAX_ITERATIONS); do
  echo "  Ralph Iterația $i din $MAX_ITERATIONS"
  echo "==================================================================="

+  # W3: alege next story via DAG (propagă blocked dacă vreun dep a eșuat)
+  CURRENT_STORY=$(dag_next_story)
+  if [ -z "$CURRENT_STORY" ]; then
+    echo ""
+    echo "==================================================================="
+    if check_all_complete; then
+      echo "  TOATE STORY-URILE DIN PRD SUNT COMPLETE!"
+      exit 0
+    else
+      echo "  NICIUN STORY ELIGIBIL (toate fie complete, fie failed, fie blocked)"
+      echo "  Stories incomplete:"
+      jq -r '.userStories[] | select(.passes != true) | "  - \(.id): \(.title) [failed=\(.failed // false) blocked=\(.blocked // false) retries=\(.retries // 0)]"' "$PRD_FILE"
+      exit 0
+    fi
+  fi
+
  # Status curent
  COMPLETE_NOW=$(jq '[.userStories[] | select(.passes == true)] | length' "$PRD_FILE")
-  NEXT_STORY=$(jq -r '[.userStories[] | select(.passes != true)] | sort_by(.priority) | .[0] | "\(.id): \(.title)"' "$PRD_FILE")
+  NEXT_TITLE=$(jq -r --arg id "$CURRENT_STORY" '.userStories[] | select(.id == $id) | "\(.id): \(.title)"' "$PRD_FILE")
+  STORY_TAGS=$(jq -r --arg id "$CURRENT_STORY" '.userStories[] | select(.id == $id) | (.tags // []) | join(",")' "$PRD_FILE")
+  STORY_RETRIES=$(jq -r --arg id "$CURRENT_STORY" '.userStories[] | select(.id == $id) | (.retries // 0)' "$PRD_FILE")
  echo "  Progress: $COMPLETE_NOW / $TOTAL_STORIES stories complete"
-  echo "  Next: $NEXT_STORY"
+  echo "  Next: $NEXT_TITLE  [tags: ${STORY_TAGS:-<none>}, retries: $STORY_RETRIES]"
  echo ""

  # Pregătește prompt-ul cu context
  FULL_PROMPT=$(cat <<EOF
 # Context pentru această iterație Ralph

+## Story țintă (DAG-eligible):
+$CURRENT_STORY  (tags: ${STORY_TAGS:-<none>})
+
 ## PRD (prd.json):
 $(cat "$PRD_FILE")

@@ -188,10 +281,34 @@ EOF
  LOG_FILE="$SCRIPT_DIR/logs/iteration-$i-$(date +%Y%m%d-%H%M%S).log"
  mkdir -p "$SCRIPT_DIR/logs"

-  # --output-format json avoids streaming mode issues
-  echo "$FULL_PROMPT" | claude -p --dangerously-skip-permissions --output-format json 2>&1 | tee "$LOG_FILE" || true
+  # --output-format json + --max-turns pentru control runtime
+  set +e
+  echo "$FULL_PROMPT" | claude -p \
+    --dangerously-skip-permissions \
+    --output-format json \
+    --max-turns "$MAX_TURNS" \
+    2>&1 | tee "$LOG_FILE"
+  CLAUDE_EXIT=${PIPESTATUS[1]}
+  set -e
  OUTPUT=$(cat "$LOG_FILE")

+  # W3: rate limit detection (max 1 retry per rulare)
+  if is_rate_limited "$OUTPUT" || [ "$CLAUDE_EXIT" = "29" ]; then
+    if [ "$RATE_LIMIT_RETRY_USED" = "0" ]; then
+      echo ""
+      echo "  ⏸️  Rate limit detectat. Sleep ${RATE_LIMIT_SLEEP}s, apoi retry o dată."
+      RATE_LIMIT_RETRY_USED=1
+      echo "## Rate limit la iter $i — sleep $RATE_LIMIT_SLEEP" >> "$PROGRESS_FILE"
+      sleep "$RATE_LIMIT_SLEEP"
+      continue  # retry aceeași iterație
+    else
+      echo "  ❌ Rate limit din nou — abort run, mark $CURRENT_STORY rate_limited"
+      dag_mark_failed "$CURRENT_STORY" "rate_limited"
+      echo "## Rate limit final la iter $i — abort" >> "$PROGRESS_FILE"
+      exit 2
+    fi
+  fi
+
  # Verifică dacă toate task-urile sunt complete
  if echo "$OUTPUT" | grep -q "<promise>COMPLETE</promise>"; then
    echo ""
@@ -211,6 +328,23 @@ EOF
    exit 0
  fi

+  # W3: tag validation post-iter — chiar dacă Opus a marcat docs, dacă diff atinge .vue/.tsx, force ui
+  dag_force_tags "$CURRENT_STORY" >/dev/null 2>&1 || true
+
+  # W3: dacă story-ul curent ÎNCĂ nu trece (passes==false), incrementăm retries
+  STILL_INCOMPLETE=$(jq -r --arg id "$CURRENT_STORY" '.userStories[] | select(.id == $id) | (.passes == true)' "$PRD_FILE")
+  if [ "$STILL_INCOMPLETE" != "true" ]; then
+    NEW_RETRY=$(dag_incr_retry "$CURRENT_STORY")
+    echo "  Story $CURRENT_STORY încă incomplet. Retries: $NEW_RETRY/3"
+    if [ "$NEW_RETRY" -ge 3 ] 2>/dev/null; then
+      echo "  ❌ $CURRENT_STORY failed: max_retries — sare la următorul"
+      # mark-failed e deja făcut de incr-retry când >=3, dar idempotent o re-aplicăm
+      dag_mark_failed "$CURRENT_STORY" "max_retries"
+    fi
+  else
+    echo "  ✅ Story $CURRENT_STORY marcat passes=true în iterația asta."
+  fi
+
  echo "  Iterația $i completă. Continuăm..."
  sleep 2
 done