Improve context persistence and memory daily log defaults

.github/workflows/privacy-scan.yml

@@ -15,17 +15,48 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: Skip push scan when PR exists
+        id: pr_guard
+        if: github.event_name == 'push'
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          REPO: ${{ github.repository }}
+          OWNER: ${{ github.repository_owner }}
+          BRANCH: ${{ github.ref_name }}
+        run: |
+          if [ "${BRANCH}" = "main" ]; then
+            echo "skip_scan=false" >> "$GITHUB_OUTPUT"
+            echo "Main branch push detected; running push scan."
+            exit 0
+          fi
+
+          prs=$(curl -sS \
+            -H "Authorization: Bearer ${GH_TOKEN}" \
+            -H "Accept: application/vnd.github+json" \
+            "https://api.github.com/repos/${REPO}/pulls?state=open&head=${OWNER}:${BRANCH}&per_page=1")
+
+          if [ "$(echo "$prs" | python3 -c "import sys, json; d=json.load(sys.stdin); print(len(d) if isinstance(d, list) else 0)")" -gt 0 ]; then
+            echo "skip_scan=true" >> "$GITHUB_OUTPUT"
+            echo "Open PR detected for branch ${BRANCH}; skipping push scan."
+          else
+            echo "skip_scan=false" >> "$GITHUB_OUTPUT"
+            echo "No open PR detected; running push scan."
+          fi
+
       - name: Cleanup stale gitleaks temp file
+        if: steps.pr_guard.outputs.skip_scan != 'true'
         run: |
           sudo rm -f /tmp/gitleaks.tmp || rm -f /tmp/gitleaks.tmp || true
 
       - name: Run gitleaks
+        if: steps.pr_guard.outputs.skip_scan != 'true'
         uses: gitleaks/gitleaks-action@v2
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Determine diff range
         id: range
+        if: steps.pr_guard.outputs.skip_scan != 'true'
         run: |
           if [ "${{ github.event_name }}" = "pull_request" ]; then
             echo "base=${{ github.event.pull_request.base.sha }}" >> "$GITHUB_OUTPUT"
@@ -44,6 +75,7 @@ jobs:
           echo "three_dot=false" >> "$GITHUB_OUTPUT"
 
       - name: Run PII guard on added lines
+        if: steps.pr_guard.outputs.skip_scan != 'true'
         run: |
           if [ "${{ steps.range.outputs.three_dot }}" = "true" ]; then
             python3 scripts/pii_guard.py \

README.md

@@ -158,6 +158,19 @@ Detailed runbook: `docs/obsidian-operations.md`.
 
 The main configuration is in `config/openclaw.json` (JSON5 format). See `config/AGENTS.md` for complete reference.
 
+### Memory Persistence Preferences
+
+This project uses explicit memory persistence preferences to reduce context loss between sessions:
+
+- **Long idle session window**: `session.reset.idleMinutes` is set to `1440` (24h) so conversations are less likely to reset before memory safeguards can run.
+- **Pre-compaction memory flush enabled**: `agents.defaults.compaction.memoryFlush.enabled: true` with `softThresholdTokens: 4000` and `reserveTokensFloor: 40000`.
+- **Memory checkpoint cron**: state repos should include a recurring checkpoint job that updates the **memory daily log** at `memory/YYYY-MM-DD.md` incrementally.
+- **Dedup cursor file**: `memory/flush-state.json` tracks checkpoint progress to reduce repeated entries in the same day.
+
+Terminology note:
+- Use **memory daily log** for `memory/YYYY-MM-DD.md`.
+- Use **Obsidian daily note** only for vault files under `07-Daily/`.
+
 ### Auxiliary ML Service (Optional)
 
 The auxiliary ML service runs in a dedicated container and is designed for queue-based, long-running jobs (minutes are acceptable). It currently starts with OCR (`glm-ocr`) and keeps a modular model registry for future additions.
@@ -238,7 +251,8 @@ josemar-assistente/
 │   ├── .sync-manifest              # Files to version
 │   ├── .gitignore                  # Security ignore list
 │   ├── skills/                     # User-owned state skills
-│   └── cron/jobs.json              # Cron definitions synced from state repo
+│   ├── cron/jobs.json              # Cron definitions synced from state repo
+│   └── memory/flush-state.json     # Checkpoint cursor for memory daily log dedup
 ├── config/                         # OpenClaw configuration
 │   ├── AGENTS.md                   # Config reference
 │   └── openclaw.json               # Main config

config/openclaw.json

@@ -186,6 +186,13 @@
       workspace: "~/.openclaw/workspace",
       userTimezone: "${JOSEMAR_TIMEZONE}",
       envelopeTimezone: "${JOSEMAR_TIMEZONE}",
+      compaction: {
+        reserveTokensFloor: 40000,
+        memoryFlush: {
+          enabled: true,
+          softThresholdTokens: 4000,
+        },
+      },
       model: {
         primary: "ollama/kimi-k2.6:cloud",
         fallbacks: ["zai/glm-5", "zai/glm-4.7", "deepseek/deepseek-chat", "ollama/glm-5.1:cloud"],
@@ -274,7 +281,7 @@
     scope: "per-sender",
     reset: {
       mode: "idle",
-      idleMinutes: 60,
+      idleMinutes: 1440,
     },
     store: "~/.openclaw/agents/josemar/sessions/sessions.json",
   },

templates/agent-state-template/.sync-manifest

@@ -13,6 +13,7 @@ skills/*/*
 
 # Memory logs (daily files matching YYYY-MM-DD.md)
 memory/[0-9][0-9][0-9][0-9]-[0-9][0-9]-[0-9][0-9].md
+memory/flush-state.json
 
 # Avatars
 avatars/*

templates/agent-state-template/README.md

@@ -40,6 +40,8 @@ Personality files (`SOUL.md`, `IDENTITY.md`, `USER.md`, `AGENTS.md`) are created
 
 Operational files (`TOOLS.md`, `BOOT.md`, `HEARTBEAT.md`) are pre-seeded in this template to document core capabilities available out of the box.
 
+This template also includes a memory checkpoint cron job (`cron/jobs.json`) plus `memory/flush-state.json` to keep daily memory logs updated incrementally with reduced duplication.
+
 ## Skill Ownership Model
 
 This project separates skills by ownership:

templates/agent-state-template/cron/jobs.json

@@ -1,4 +1,31 @@
 {
   "$schema": "https://openclaw.ai/schemas/cron-jobs.json",
-  "jobs": []
+  "jobs": [
+    {
+      "id": "9bb8cc42-bca2-4516-b7c0-4305d0315fc9",
+      "agentId": "josemar",
+      "sessionKey": "agent:josemar:main",
+      "name": "Checkpoint de memória",
+      "description": "Checkpoint incremental de memória para garantir memory daily log (memory/YYYY-MM-DD.md) sem duplicações",
+      "enabled": true,
+      "schedule": {
+        "expr": "0 10,15,21 * * *",
+        "kind": "cron",
+        "tz": "America/Sao_Paulo"
+      },
+      "sessionTarget": "main",
+      "wakeMode": "now",
+      "payload": {
+        "kind": "agentTurn",
+        "message": "Checkpoint de memória (sem resposta ao usuário).\n\n1. Use NO_REPLY.\n2. Garanta que exista memory/YYYY-MM-DD.md para hoje; crie se não existir.\n3. Leia memory/YYYY-MM-DD.md e memory/flush-state.json antes de escrever.\n4. Faça um resumo incremental do que aconteceu desde o último checkpoint, sem repetir itens já registrados hoje.\n5. Escreva apenas fatos relevantes (decisões, mudanças de configuração, ações executadas, pendências).\n6. Se não houver novidade real desde o último checkpoint, não altere o memory daily log (memory/YYYY-MM-DD.md).\n7. Atualize memory/flush-state.json com lastCheckpointAt (ISO-8601) e lastSummaryHash (hash do resumo incremental).",
+        "toolsAllow": [
+          "read",
+          "write"
+        ]
+      },
+      "delivery": {
+        "mode": "none"
+      }
+    }
+  ]
 }

templates/agent-state-template/memory/flush-state.json

@@ -0,0 +1,4 @@
+{
+  "lastCheckpointAt": null,
+  "lastSummaryHash": null
+}