Skip to content

An always-on second brain you talk to. Voice notes in Telegram → typed, linked knowledge in your Obsidian vault. Runs 24/7 on the Claude subscription you already have.

feat(processor): improve reliability and add process goals methodology

smixs4bd627730c15
diff --git a/scripts/process.sh b/scripts/process.sh
index 089677b..ae7482e 100755
--- a/scripts/process.sh
+++ b/scripts/process.sh
@@ -21,40 +21,68 @@ if [ -z "$TELEGRAM_BOT_TOKEN" ]; then
exit 1
fi
+# MCP timeout for stdio server (default 5 sec is too short)
+export MCP_TIMEOUT=30000
+export MAX_MCP_OUTPUT_TOKENS=50000
+
# Date and chat_id
TODAY=$(date +%Y-%m-%d)
CHAT_ID="${ALLOWED_USER_IDS//[\[\]]/}" # remove brackets from [123456]
-cd "$PROJECT_DIR" # MCP configured for project root
-
echo "=== d-brain processing for $TODAY ==="
-# Run Claude with --dangerously-skip-permissions and MCP
+# Run Claude from vault/ for context (reads vault/.claude/CLAUDE.md)
+cd "$VAULT_DIR"
REPORT=$(claude --print --dangerously-skip-permissions \
--mcp-config "$PROJECT_DIR/mcp-config.json" \
- -p "Today is $TODAY. Execute daily processing according to dbrain-processor skill." \
+ -p "Today is $TODAY. Execute daily processing according to dbrain-processor skill.
+
+CRITICAL: MCP loads in 10-30 seconds. You are NOT in subprocess — MCP IS running, just initializing.
+
+Algorithm:
+1. Call mcp__todoist__find-tasks-by-date
+2. Error? Wait 10 sec, read goals/, daily/ files
+3. Call again
+4. Error again? Wait 20 more sec
+5. Third call — GUARANTEED to work
+
+DO NOT say MCP unavailable. It is available. Just wait and call." \
2>&1) || true
+cd "$PROJECT_DIR"
echo "=== Claude output ==="
echo "$REPORT"
echo "===================="
+# Remove HTML comments (break Telegram HTML parser)
+REPORT_CLEAN=$(echo "$REPORT" | sed '/<!--/,/-->/d')
+
+# Rebuild vault graph (keeps structure up to date)
+echo "=== Rebuilding vault graph ==="
+cd "$VAULT_DIR"
+uv run .claude/skills/graph-builder/scripts/analyze.py || echo "Graph rebuild failed (non-critical)"
+cd "$PROJECT_DIR"
+
# Git commit
git add -A
git commit -m "chore: process daily $TODAY" || true
git push || true
# Send to Telegram
-if [ -n "$REPORT" ] && [ -n "$CHAT_ID" ]; then
+if [ -n "$REPORT_CLEAN" ] && [ -n "$CHAT_ID" ]; then
echo "=== Sending to Telegram ==="
- curl -s -X POST "https://api.telegram.org/bot$TELEGRAM_BOT_TOKEN/sendMessage" \
+ RESULT=$(curl -s -X POST "https://api.telegram.org/bot$TELEGRAM_BOT_TOKEN/sendMessage" \
-d "chat_id=$CHAT_ID" \
- -d "text=$REPORT" \
- -d "parse_mode=HTML" || \
- # Fallback: send without HTML parsing
- curl -s -X POST "https://api.telegram.org/bot$TELEGRAM_BOT_TOKEN/sendMessage" \
- -d "chat_id=$CHAT_ID" \
- -d "text=$REPORT"
+ -d "text=$REPORT_CLEAN" \
+ -d "parse_mode=HTML")
+
+ # If HTML failed, send without formatting
+ if echo "$RESULT" | grep -q '"ok":false'; then
+ echo "HTML failed: $RESULT"
+ curl -s -X POST "https://api.telegram.org/bot$TELEGRAM_BOT_TOKEN/sendMessage" \
+ -d "chat_id=$CHAT_ID" \
+ -d "text=$REPORT_CLEAN"
+ fi
fi
echo "=== Done ==="
diff --git a/vault/.claude/CLAUDE.md b/vault/.claude/CLAUDE.md
index 4c4f2f1..8f9e4b0 100644
--- a/vault/.claude/CLAUDE.md
+++ b/vault/.claude/CLAUDE.md
@@ -207,7 +207,15 @@ Analyze and maintain vault link structure. Use `/graph` command or invoke `graph
See `skills/graph-builder/` for full documentation.
+## Learnings (from experience)
+
+1. **Don't rewrite working code** without reason (KISS, DRY, YAGNI)
+2. **Don't add checks** that weren't there — let the agent decide
+3. **Don't propose solutions** without studying git log/diff first
+4. **Don't break architecture** (process.sh → Claude → skill is correct)
+5. **Problems are usually simple** (e.g., sed one-liner for HTML fix)
+
---
-*System Version: 2.2*
-*Updated: 2026-01-30*
+*System Version: 2.3*
+*Updated: 2026-02-01*
diff --git a/vault/.claude/skills/dbrain-processor/SKILL.md b/vault/.claude/skills/dbrain-processor/SKILL.md
index 0c51d51..66886c8 100644
--- a/vault/.claude/skills/dbrain-processor/SKILL.md
+++ b/vault/.claude/skills/dbrain-processor/SKILL.md
@@ -35,7 +35,7 @@ mcp__todoist__find-tasks-by-date — Check workload
## CRITICAL: MCP Tool Usage
-**НИКАКИХ WORKAROUNDS. НИКАКИХ "добавь вручную". ТОЛЬКО ПРЯМЫЕ ВЫЗОВЫ.**
+**СНАЧАЛА ВЫЗОВИ TOOL. ПОТОМ ДУМАЙ.**
У тебя ЕСТЬ доступ к MCP tools:
- `mcp__todoist__add-tasks`
@@ -44,17 +44,29 @@ mcp__todoist__find-tasks-by-date — Check workload
- `mcp__todoist__complete-tasks`
- `mcp__todoist__update-tasks`
-ЗАПРЕЩЕНО:
-- Писать "MCP недоступен"
-- Предлагать "добавь вручную"
-- Использовать subprocess для вызова CLI
-- Делать HTTP запросы к API напрямую
-- Выводить команды для копирования
-
-ОБЯЗАТЕЛЬНО:
-- Вызывать `mcp__todoist__add-tasks` tool напрямую
-- Если tool вернул ошибку — включить её в отчёт
-- Если task создан — включить task ID в отчёт
+### Обязательный алгоритм:
+
+1. ВЫЗОВИ: mcp__todoist__find-tasks-by-date
+ ↓
+ Получил результат? → Продолжай
+ ↓
+ Ошибка? → Читай файлы 30 секунд, потом ВЫЗОВИ СНОВА
+ ↓
+ 3 ошибки подряд? → Покажи ТОЧНЫЙ текст ошибки
+
+### ЗАПРЕЩЕНО:
+- ❌ Писать "MCP недоступен"
+- ❌ Предлагать "добавь вручную"
+- ❌ Использовать subprocess для вызова CLI
+- ❌ Делать HTTP запросы к API напрямую
+- ❌ Выводить команды для копирования
+- ❌ Решать что не работает БЕЗ вызова tool
+
+### ОБЯЗАТЕЛЬНО:
+- ✅ Вызывать MCP tool напрямую
+- ✅ Если ошибка — подождать, вызвать снова
+- ✅ 3 retry перед любыми выводами
+- ✅ Если task создан — включить task ID в отчёт
При ошибке MCP tool — показать ТОЧНУЮ ошибку от tool, не придумывать отговорки.
@@ -62,10 +74,105 @@ mcp__todoist__find-tasks-by-date — Check workload
1. Load context — Read goals/3-weekly.md (ONE Big Thing), goals/2-monthly.md
2. Check workload — find-tasks-by-date for 7 days
-3. Read daily — daily/YYYY-MM-DD.md
-4. Process entries — Classify → task or thought
-5. Build links — Connect notes with [[wiki-links]]
-6. Generate HTML report — RAW HTML for Telegram
+3. **Check process goals** — find-tasks with labels: ["process-goal"]
+4. Read daily — daily/YYYY-MM-DD.md
+5. Process entries — Classify → task or thought
+6. Build links — Connect notes with [[wiki-links]]
+7. **Log actions to daily** — append action log entry
+8. **Evolve MEMORY.md** — update long-term memory if needed
+9. Generate HTML report — RAW HTML for Telegram
+
+## Process Goals Check (Step 3)
+
+**ОБЯЗАТЕЛЬНО выполни при каждом /process:**
+
+### 1. Проверь существующие process goals
+Используй mcp__todoist__find-tasks с labels: ["process-goal"]
+
+### 2. Если отсутствуют — создай
+Читай goals/ и генерируй process commitments:
+
+| Goal Level | Source | Process Pattern |
+|------------|--------|-----------------|
+| Weekly ONE Big Thing | goals/3-weekly.md | 2h deep work ежедневно |
+| Monthly Top 3 | goals/2-monthly.md | 1 action/день на приоритет |
+| Yearly Focus | goals/1-yearly-*.md | 30 мин/день на стратегию |
+
+Создавай recurring tasks с label "process-goal" (max 5-7 активных).
+
+### 3. Включи в отчёт
+
+```html
+<b>📋 Process Goals:</b>
+• 2h deep work → ✅ активен
+• 1 outreach/день → ⚠️ просрочен
+{N} активных | {M} требуют внимания
+```
+
+See: references/process-goals.md for patterns and examples.
+
+## Logging to daily/ (Step 7)
+
+**После ЛЮБЫХ изменений в vault — СРАЗУ пиши в `daily/YYYY-MM-DD.md`:**
+
+Format:
+```
+## HH:MM [text]
+{Description of actions}
+
+**Created/Updated:**
+- [[path/to/file|Name]] — description
+```
+
+What to log:
+- Files created in thoughts/
+- Tasks created in Todoist (with task ID)
+- Links built between notes
+
+Example:
+```
+## 14:30 [text]
+Daily processing complete
+
+**Created tasks:** 3
+- "Follow-up client" (id: 8501234567, p2, tomorrow)
+- "Prepare proposal" (id: 8501234568, p2, friday)
+
+**Saved thoughts:** 1
+- [[thoughts/ideas/product-launch|Product Launch]] — new idea
+```
+
+## Evolve MEMORY.md (Step 8)
+
+**GOAL:** Keep MEMORY.md current. Don't append — EVOLVE.
+
+### When to update:
+- ✅ Key decisions with impact (pivot, tool choice, architecture change)
+- ✅ New patterns/insights (learnings)
+- ✅ Changes in Active Context (new ONE Big Thing, Hot Projects)
+
+### When NOT to update:
+- ❌ Daily trivia (meetings, calls without impact)
+- ❌ Temporary notes (stay in daily/)
+- ❌ Duplicates of what's already there
+
+### How to update (evolve, not append):
+
+| Situation | Action |
+|-----------|--------|
+| New contradicts old | REPLACE old information |
+| New complements old | Add to existing section |
+| Info is outdated | Delete or archive |
+
+Use Edit tool for precise changes.
+
+### In report (if updated):
+
+```html
+<b>🧠 MEMORY.md updated:</b>
+• Active Context → Hot Projects changed
+• Key Decisions → +1 new decision
+```
## Entry Format
@@ -108,6 +215,11 @@ Output RAW HTML (no markdown, no code blocks):
<b>✅ Создано задач:</b> {M}
• {task} <i>({priority}, {due})</i>
+<b>📋 Process Goals:</b>
+• {process goal 1} → {status}
+• {process goal 2} → {status}
+{N} активных | {M} требуют внимания
+
<b>📅 Загрузка на неделю:</b>
Пн: {n} | Вт: {n} | Ср: {n} | Чт: {n} | Пт: {n} | Сб: {n} | Вс: {n}
@@ -125,6 +237,10 @@ Output RAW HTML (no markdown, no code blocks):
<b>📈 Прогресс:</b>
• {goal}: {%} {emoji}
+<b>🧠 MEMORY.md:</b>
+• {section} → {change description}
+<i>(if updated)</i>
+
---
<i>Обработано за {duration}</i>
@@ -137,6 +253,11 @@ If all entries have `<!-- ✓ processed -->` marker, return status report:
<b>🎯 Текущий фокус:</b>
{ONE_BIG_THING}
+<b>📋 Process Goals:</b>
+• {process goal 1} → {status}
+• {process goal 2} → {status}
+{N} активных | {M} требуют внимания
+
<b>📅 Загрузка на неделю:</b>
Пн: {n} | Вт: {n} | Ср: {n} | Чт: {n} | Пт: {n} | Сб: {n} | Вс: {n}
@@ -177,6 +298,7 @@ Read these files as needed:
- references/classification.md — Entry classification rules
- references/todoist.md — Task creation details
- references/goals.md — Goal alignment logic
+- references/process-goals.md — Process vs outcome goals, transformation patterns
- references/links.md — Wiki-links building
- references/rules.md — Mandatory processing rules
- references/report-template.md — Full HTML report spec
diff --git a/vault/.claude/skills/dbrain-processor/references/process-goals.md b/vault/.claude/skills/dbrain-processor/references/process-goals.md
new file mode 100644
index 0000000..d3c5ea2
--- /dev/null
+++ b/vault/.claude/skills/dbrain-processor/references/process-goals.md
@@ -0,0 +1,108 @@
+# Process Goals
+
+## Concept
+
+**Outcome goals** focus on WHAT you want to achieve:
+- "Close the deal"
+- "Hit $1M ARR"
+- "Launch the product"
+
+**Process goals** focus on HOW you do it:
+- "Reach out to 100 potential clients"
+- "Send 5 proposals per week"
+- "Conduct 3 customer interviews this week"
+
+## Why Process Goals Work Better
+
+Based on research (meta-analysis of 27 experiments in sports):
+
+1. **Control** — You control actions, not outcomes. "Close deal" creates anxiety; "send 3 follow-ups" is actionable.
+2. **Confidence** — Process goals turn challenge into achievable steps, boosting self-efficacy.
+3. **Focus** — Daily clarity on what to do, not vague aspiration.
+
+## How to Transform
+
+| Outcome Goal | Process Goal |
+|--------------|--------------|
+| "Close the deal with X" | "Send 1 follow-up per day to client X" |
+| "Launch product" | "2h deep work on MVP every morning" |
+| "Get 10 clients" | "Write 5 outreach messages per day" |
+| "Prepare the program" | "30 min research analogs daily" |
+| "Find a partner" | "1 networking message per day" |
+
+## Patterns for Process Goals
+
+### Time Blocks
+Regular dedicated time for deep work:
+- "2h deep work on [project] every morning 6-8am"
+- "1h research [topic] after lunch"
+
+### Frequency Commitments
+Consistent action cadence:
+- "1 outreach message per day"
+- "3 customer calls per week"
+- "5 proposals per week"
+
+### Checkpoints
+Review milestones:
+- "Show draft to 1 person before Friday"
+- "Review progress with mentor on Wednesday"
+
+### Streaks
+Daily habits:
+- "30 min movement every day"
+- "10 min journaling in the morning"
+
+## When Processing Voice Entries
+
+If user says outcome goal, consider transforming:
+
+```
+User: "Need to close the deal with client X"
+
+BAD: Create task "Close deal with client X" p1
+→ Vague, creates anxiety, unclear next action
+
+GOOD: Create task "Send follow-up to client X with updated proposal" p1 due:tomorrow
+→ Specific, actionable, controllable
+```
+
+## For Weekly Planning
+
+When generating Process Commitments for ONE Big Thing:
+
+1. **Ask:** What daily/weekly habits would make this inevitable?
+2. **Limit:** 3-5 process commitments max
+3. **Prioritize:** Time blocks > Frequency > Checkpoints
+4. **Create:** Recurring tasks in Todoist with appropriate patterns
+
+## Recurring Task Patterns
+
+Map process goals to Todoist dueString:
+
+| Process | dueString |
+|---------|-----------|
+| "every morning" | "every day at 6am" |
+| "every day" | "every day" |
+| "every workday" | "every weekday" |
+| "3 times per week" | "every monday, wednesday, friday" |
+| "once a week" | "every week" |
+| "by Friday" | "friday" (one-time) |
+
+## Anti-Patterns
+
+Avoid creating process goals that are:
+- Too vague: "Work on project" (how much? when?)
+- Too ambitious: "4h deep work 5 days a week" (unsustainable)
+- Untrackable: "Think about strategy" (not measurable)
+
+Good process goals are:
+- Specific in action
+- Bounded in time/frequency
+- Trackable (done/not done)
+
+## Reference
+
+Based on: Meta-analysis of 27 experiments in sports showing process goals significantly outperform outcome/performance goals.
+
+See also: "The Success Equation" by Mauboussin — Results = Skill + Luck. Focus on skill (process) is the only controllable strategy.