--- name: Proactive Claw description: > 🦞 The most powerful proactive engine for OpenClaw. Your personal assistant that learns from you and helps you be more productive.

version: 1.2.41

metadata: openclaw: requires: bins: [python3] config: [credentials.json, config.json] install: - kind: shell label: "Run scripts/setup.sh --doctor, install required deps manually, then run scripts/setup.sh" ---

🦞 Proactive Claw

Proactive Claw is a proactive execution engine that collaborates with you and learns from you.

It helps you be more productive by making smart suggestions at the right moment — and (by default) only applying what you approve.

Everything can run locally.

---

Safety in one glance ✅

✅ Asks before applying changes (default) ✅ Writes only what you approve (no surprise calendar edits) ✅ Local-first core bundle (no external integration helpers) ✅ Everything can run locally (small/local scoring model recommended) ✅ Local state folder is transparent and deletable

---

Why people install it (what you feel in week 1)

• You show up prepared more often.
• Your calendar becomes more realistic (buffers appear where they matter).
• Deep work blocks stop getting shredded by “just one quick meeting.”
• Follow-ups stop slipping.
• Proactive Claw learns your style quickly:

prep durations, preferred times, what you reject, what you always accept.

In short: it becomes your personal assistant.

---

Graph A — the simple loop (how it works at a glance)

+--------------------------+        +--------------------------+
| Calendar                 |        | Chat                     |
| Google Calendar/Nextcloud|        | (your control loop)      |
+------------+-------------+        +------------+-------------+
             |  events <-> suggestions           | prompts <-> decisions
             |                                   |
             +-------------------+---------------+
                                 |
                                 v
                  +--------------------------------------+
                  |            Proactive Claw             |
                  |  - notices what's coming              |
                  |  - prioritizes what matters           |
                  |  - suggests prep / buffers / followup |
                  |  - learns from your feedback          |
                  |  - applies only what you approve      |
                  +--------------------------------------+

Key idea: Proactive Claw is proactive, but you remain the decision-maker (by default).

---

Quick start (2 minutes)

1) Run readiness checks:

bash scripts/setup.sh --doctor

2) If dependencies are missing, print one install command:

bash scripts/setup.sh --print-install-cmd google
# or: bash scripts/setup.sh --print-install-cmd nextcloud

3) Bootstrap calendar connection:

bash scripts/setup.sh

4) Run a safe preview:

bash scripts/quickstart.sh

5) Approve a suggestion in chat, then run once:

python3 scripts/daemon.py

---

Modes (choose your vibe)

• Suggest (default): asks before applying changes (max_autonomy_level=confirm)
• Background (manual): run local daemon yourself (python3 scripts/daemon.py --loop)
• Autonomous (advanced): explicit opt-in only (not recommended until you’ve used it for a while)

---

Presets (easy setup)

You don’t need to tune dozens of settings. Pick a preset.

Presets are just a starting point. Proactive Claw adjusts to you with every interaction (approvals, edits, rejections), and quickly becomes personal.

A) Calm mode (minimal interruptions)

Choose this if you want only high-value suggestions.

• fewer prompts
• only important events get prep/buffers

B) Builder mode (deep work protected)

Choose this if you code/build and need long focus blocks.

• defends 2–3h deep work windows
• suggests moving meetings that fragment focus
• stronger buffers + recovery time

C) Meetings-heavy week

Choose this when your schedule is packed.

• adds buffers + reset breaks
• prevents back-to-back collapse
• proposes follow-ups so action doesn’t slip

(If you want, these presets can be expressed as a small copy/paste block in config.json.)

---

Chat scoring model (built-in; small/local recommended)

Proactive Claw includes a chat scoring model to rank signals (importance, urgency, disruption risk) so it knows:

• what matters,
• when to ask you,
• and when to stay quiet.

Recommendation: use a small model, ideally local (fast, cheap, private). Your planning model can be bigger; scoring should stay small and fast.

---

Scenarios (mini transcripts)

1) Presentation tomorrow → prep + buffer

You: Tomorrow I’m presenting the roadmap to the board. Claw (score = 0.92): High impact. You usually prep ~70 minutes for presentations. Claw: Want me to reserve 08:40–09:50 for prep + a 10-min buffer before the 10:00 meeting?

2) Deep work protection (developer mode)

Claw (score = 0.78): This meeting would split your deep work block (09:30–12:00). Claw: Prefer to move it to 13:00, or keep it and add a 20-min recovery buffer after?

3) Back-to-back day → buffers + reset

Claw (score = 0.81): Tomorrow is back-to-back from 10:00–14:00. Claw: Add 10-min buffers between meetings + a 25-min reset break around midday?

4) Follow-up that tends to slip

You: I promised I’ll send the summary. Claw (score = 0.66): Time-sensitive. Claw: Want a 20-min follow-up block right after the meeting (or at 17:10)?

5) It becomes personal fast (recurring pattern)

Claw (score = 0.71): I’ve noticed a pattern for “Client Review” meetings:

• you prefer 30-min prep the same day
• 10-min buffer before
• no meetings after 18:00

Claw: Want me to propose this pattern whenever a new “Client Review” appears (still asking you to approve)?

---

Graph B — technical architecture (backend = Google OR Nextcloud)

TECHNICAL ARCHITECTURE (backend = Google OR Nextcloud)

...