phoenixclaw

# PhoenixClaw: Zero-Tag Passive Journaling

PhoenixClaw automatically distills daily conversations into meaningful reflections using semantic intelligence.

Automatically identifies journal-worthy moments, patterns, and growth opportunities.

## 🛠️ Core Workflow

> [!critical] **MANDATORY: Complete Workflow Execution**
> This 9-step workflow MUST be executed in full regardless of invocation method:
> - **Cron execution** (10 PM nightly)
> - **Manual invocation** ("Show me my journal", "Generate today's journal", etc.)
> - **Regeneration requests** ("Regenerate my journal", "Update today's entry")
> 
> **Never skip steps.** Partial execution causes:
> - Missing images (session logs not scanned)
> - Missing finance data (Ledger plugin not triggered)
> - Incomplete journals (plugins not executed)

PhoenixClaw follows a structured pipeline to ensure consistency and depth:

1. **User Configuration:** Check for `~/.phoenixclaw/config.yaml`. If missing, initiate the onboarding flow defined in `references/user-config.md`.
2. **Context Retrieval:** 
   - **Scan memory files (NEW):** Read `memory/YYYY-MM-DD.md` and `memory/YYYY-MM-DD-*.md` files for manually recorded daily reflections. These files contain personal thoughts, emotions, and context that users explicitly ask the AI to remember via commands like "记一下" (remember this). **CRITICAL**: Do not skip these files - they contain explicit user reflections that session logs may miss.
   - **Scan session logs:** Call `memory_get` for the current day's memory, then **CRITICAL: Scan ALL raw session logs and filter by message timestamp**. Session files are often split across multiple files. Do NOT classify images by session file `mtime`:
      ```bash
      # Read all session logs from ALL known OpenClaw locations, then filter by per-message timestamp
      # Use timezone-aware epoch range to avoid UTC/local-day mismatches.
      TARGET_DAY="$(date +%Y-%m-%d)"
      TARGET_TZ="${TARGET_TZ:-Asia/Shanghai}"
      read START_EPO...