Background
This is an experimental, for-fun POC.
Goal is to see if OpenClaw can read a generated note and turn it into useful operational actions in a safe sandbox.
Not for clinical decision-making. Not production automation.
How This Works (End-to-End)
- User finishes recording -> note is generated as usual.
- If
OpenClaw Mode = Off: nothing new happens.
- If
Suggest Only: note is sent to OpenClaw planner, which returns a structured action list.
- UI shows a “Claw Report”:
- Proposed actions
- Confidence
- Why it suggested each action
- Blocked actions (with reason)
- If
Tiny Actions Only is enabled:
- Only allowlisted low-risk actions execute
- Everything else is blocked and logged
POC Modes
OFF (default)
SUGGEST_ONLY (recommended)
TINY_ACTIONS_ONLY (still gated by allowlist)
POC Action Contract
OpenClaw returns JSON like:
type (e.g., create_task, set_followup_reminder, tag_encounter)
payload (action data)
confidence (0-1)
reason (short explanation)
Initial Allowlist (tiny/safe actions only)
create_task (local task list only)
set_followup_reminder (local reminder only)
tag_encounter (local metadata tag)
Everything else: blocked.
Scope
- Add mode toggle in Settings with “for-fun experimental” copy.
- Add OpenClaw planner call after note generation.
- Add policy engine (allowlist + confidence threshold).
- Add Claw Report UI/log.
- Add audit entries for proposed/executed/blocked actions.
- Keep normal note workflow unchanged when mode is off.
Success Criteria
- Feature is clearly labeled as POC/experimental.
- Default behavior unchanged (
OFF).
SUGGEST_ONLY never executes actions.
TINY_ACTIONS_ONLY executes only allowlisted actions.
- Blocked actions are visible with reasons.
- OpenClaw failure does not break note generation.
- Tests cover mode gating, allowlist blocking, and failure handling.
Why This Is a Good POC
You get real signal on “note -> action” usefulness, while keeping blast radius tiny and the chaos contained to a sandbox.
Background
This is an experimental, for-fun POC.
Goal is to see if OpenClaw can read a generated note and turn it into useful operational actions in a safe sandbox.
Not for clinical decision-making. Not production automation.
How This Works (End-to-End)
OpenClaw Mode = Off: nothing new happens.Suggest Only: note is sent to OpenClaw planner, which returns a structured action list.Tiny Actions Onlyis enabled:POC Modes
OFF(default)SUGGEST_ONLY(recommended)TINY_ACTIONS_ONLY(still gated by allowlist)POC Action Contract
OpenClaw returns JSON like:
type(e.g.,create_task,set_followup_reminder,tag_encounter)payload(action data)confidence(0-1)reason(short explanation)Initial Allowlist (tiny/safe actions only)
create_task(local task list only)set_followup_reminder(local reminder only)tag_encounter(local metadata tag)Everything else: blocked.
Scope
Success Criteria
OFF).SUGGEST_ONLYnever executes actions.TINY_ACTIONS_ONLYexecutes only allowlisted actions.Why This Is a Good POC
You get real signal on “note -> action” usefulness, while keeping blast radius tiny and the chaos contained to a sandbox.