pulseflow（pulseflow 效率）

PulseFlow

Build and operate a simple task system with one human-facing dashboard file and one append-only AI log flow.

Core model

Use one current dashboard file as the source of truth for active work.

• Weekly usage lives in one top summary panel only: AI USAGE THIS WEEK
• Human work lives in four sections only: FOCUS, TODAY, UP NEXT, DONE
• AI work lives in one summary section only: AI DONE TODAY
• Agents do not write the dashboard directly
• Agents append one log line after each completed work unit
• Heartbeat or an explicit refresh reads usage data plus logs and rewrites the AI-derived sections

File roles

Use these files and keep their roles strict:

• todo/NOW.md — current dashboard
• todo/history/YYYY-MM.md — monthly archive
• todo/system/config.json — installation-specific paths and agent list
• todo/system/sync-state.json — last processed offsets/checkpoints
• reports/<agent>-ai-log-YYYY-MM-DD.jsonl — append-only per-agent daily work logs

Initialization workflow

When setting up a new installation:

1. Create todo/, todo/history/, and todo/system/
2. Create todo/NOW.md from the dashboard template in references/now-template.md
3. Create todo/history/<current-month>.md from references/history-template.md if missing
4. Create todo/system/config.json from references/config-template.json
5. Create todo/system/sync-state.json from references/sync-state-template.json
6. Fill installation-specific values:

- dashboard path - history directory - reports directory per agent - optional agentsFilePath per agent - enabled agent list - timezone

1. If an older dashboard exists, migrate the human task sections into the new todo/NOW.md
2. Do not import old AI activity retroactively unless explicitly requested
3. Create today's empty AI log files for enabled agents
4. Install or refresh managed AI logging rules in configured AGENTS.md files
5. If the installation explicitly wants scheduled summaries, fill notifications.summaryCrons in config and run scripts/install_summary_crons.js

Agent write contract

Each agent must append one JSON object per completed work unit to its own daily JSONL log.

Required fields:

• ts — ISO timestamp with timezone
• agent — agent name
• task — one-line work summary
• tokens — optional integer token count for that work unit; may be 0 or omitted

Example line:

{"ts":"2026-04-05T14:10:00+08:00","agent":"main","task":"完成量化交易系统回测模块初版","tokens":12000}

Rules:

• Append only; never edit prior lines in-place
• Write after each completed work unit, not only at day end
• One line should be readable without extra context
• Do not depend on per-task token totals for daily usage accounting
• Do not write to todo/NOW.md directly from agents

Heartbeat sync workflow

On heartbeat or explicit refresh:

1. Read todo/system/config.json
2. Read todo/system/sync-state.json
3. For each enabled agent, locate today's JSONL log
4. Parse all valid lines for today
5. Query OpenClaw usage summary for the current week and rebuild AI USAGE THIS WEEK
6. Rebuild the AI DONE TODAY section from the log contents for today
7. Compute today's total tokens from OpenClaw usage data, not by summing task lines
8. Rewrite only the AI-derived sections in todo/NOW.md
9. Update todo/system/sync-state.json

Important:

• Treat the dashboard AI section as a derived view, not a write-ahead log
• Rebuild from source logs for the current day instead of incrementally appending into Markdown
• Skip malformed lines; never let one bad line break the whole sync

Dashboard format

Keep the dashboard simple and stable.

Top usage section:

• ## AI USAGE THIS WEEK
• table columns: Date / Total Tokens / Input / Output / Cache / Hit Rate

Human task sections:

• ## FOCUS
• ## TODAY
• ## UP NEXT
• ## DONE

AI section:

• ## AI DONE TODAY
• then bullet items formatted as:

- <agent>: <task>

Archive workflow

At end of day or next-day rollover:

1. Append completed human tasks and the AI section snapshot into todo/history/YYYY-MM.md
2. Update the month file's AI Usage Daily Summary block with finalized per-day usage rows
3. Group archived day entries by clipped week sections inside the month file, and keep one AI Usage Weekly Summary table per visible week block
4. Reset DONE in todo/NOW.md
5. Reset AI DONE TODAY for the new day on first sync
6. Rebuild AI USAGE THIS WEEK for the new current week
7. Move unfinished FOCUS and unfinished TODAY into the new day's TODAY
8. Keep unfinished UP NEXT unchanged
9. Reset FOCUS to an empty placeholder

Scripts

• scripts/append_ai_log.js — appends one JSONL AI work record to today's per-agent log
• scripts/install_agent_log_rules.js — installs or updates managed AI log rule blocks in configured AGENTS.md files
• scripts/install_summary_crons.js — optionally installs or updates template-driven notification cron jobs from notifications.summaryCrons in config; these jobs supplement rollover and do not replace it
• scripts/init_system.js — creates missing dashboard, history, config, state files, today's empty AI logs, and installs managed agent log rules when configured; it does not install cron jobs by default
• scripts/repair_system.js — repairs missing runtime files without overwriting healthy ones
• scripts/rollover_now.js — daily rollover script; archives yesterday's human done + AI snapshot into a week-grouped month file, updates that week's AI Usage Weekly Summary, clears DONE, resets AI DONE TODAY, carries unfinished tasks forward, and updates rollover-state.json
• scripts/sync_ai_done.js — deterministic sync script for heartbeat or manual refresh; reads config, queries weekly usage, scans today's agent JSONL logs, rebuilds AI USAGE THIS WEEK plus AI DONE TODAY, and updates sync state
• scripts/validate_system.js — runs a local end-to-end validation covering init, sync, repair, rollover, and rollover idempotency

Run rollover with:

node <skill-dir>/scripts/rollover_now.js

Recommended scheduler:

• daily cron at 00:15 Asia/Shanghai

Run with:

node <skill-dir>/scripts/sync_ai_done.js

Optional config override:

AI_WORKLOG_CONFIG=/path/to/config.json node <skill-dir>/scripts/sync_ai_done.js

Validation:

node <skill-dir>/scripts/validate_system.js

Install or refresh agent rules:

AI_WORKLOG_CONFIG=/absolute/path/to/todo/system/config.json node <skill-dir>/scripts/install_agent_log_rules.js

Optionally install or update summary cron jobs after filling notifications.summaryCrons in config:

AI_WORKLOG_CONFIG=/absolute/path/to/todo/system/config.json node <skill-dir>/scripts/install_summary_crons.js

Review the planned cron changes first with:

AI_WORKLOG_CONFIG=/absolute/path/to/todo/system/config.json node <skill-dir>/scripts/install_summary_crons.js --dry-run

Recommended production invocation:

AI_WORKLOG_CONFIG=/absolute/path/to/todo/system/config.json node <skill-dir>/scripts/sync_ai_done.js

Templates and references

Read these files as needed:

• references/now-template.md — base dashboard template
• references/history-template.md — monthly archive template
• references/config-template.json — install config shape, including optional summary-cron settings
• references/sync-state-template.json — sync checkpoint shape
• references/agent-log-format.md — log schema and examples
• references/heartbeat-checklist.md — heartbeat operating checklist
• references/midday-summary-template.md — optional 15:30 summary structure
• references/daily-close-template.md — optional 00:05 previous-day report structure, intended to run before rollover
• docs/portability.md — env overrides, installation assumptions, and optional cron-install notes

Output standard

When asked to initialize, produce or update the installation files only. When asked to sync, update only the AI-derived sections and sync state unless the user asks for more. When asked to design or migrate, keep the system minimal and reusable; avoid installation-specific assumptions outside config.json.

Runtime note

The JavaScript runtime is the only supported implementation.