Proactive-Do

A lightweight workflow for proactive execution of a human's todo list with recurring reviews, clear state labels, structured documentation, and concise reporting.

Quick start

1) Source of truth

• Todos live in todo/todo.md, grouped by date headings (e.g., "## YYYY-MM-DD").
• Items are bullets with a state label: "- [new] ..., - [wip] ..., - [done] ...".
• Only these three states are valid.
• Strict preservation rule: never delete existing todo items from todo/todo.md automatically. You may append new items and revise existing items in place (for example, changing only the state label or updating the text of the same line), but you must not remove lines/items unless the owner/requester explicitly asks for deletion.

2) Initialize phase (install-time only; do not run on every review)

• Run these setup actions when the skill is first installed or first adopted in a workspace, not during each 3h review or 2.5h follow-up.
• Ensure these paths exist; create them if missing:
• todo/
• todo/todo.md
• agent_work/
• agent_work/heartbeat_emails/
• agent_work/proactive-do/
• If todo/todo.md does not exist, create it with a minimal starter structure:
• a top-level heading or today's date section ("## YYYY-MM-DD")
• If agent_work/proactive-do/delivery_prefs.md is referenced later but missing, do not fail; create the parent folder and continue with fallback behavior per notification policy.
• Initialization must be non-destructive: create missing paths only, do not remove or overwrite existing content.

3) Cadence (recommended)

2) Cadence (recommended)

• Every 3 hours: scan all [new] items and pick the top 3 to address this pass (leave the rest for later) to reduce load.
• Every 2.5 hours: reconcile what was worked on; mark [done] if finished, else [wip] with a one-line reason + next action.
• Use OpenClaw cron to schedule system events that trigger these reviews (see Cron payloads below).

3) Decision policy

• If a selected [new] item appears doable within ~1 hour: do it now during the 3h review.
• If it likely needs >1 hour: write a simple, structured plan (steps + rough time estimate) and leave the item [new] until kickoff; when starting work, flip to [wip].
• If assistance/approval is needed: halt, report to the owner/requester, and wait.
• Search online when needed.
• If blocked twice on the same point: stop, record the failure and next step, mark [wip], and report.

4) Selection heuristics (for "top 3")

Default ranking when not specified by the owner/requester:

1) Explicit priority hints in the text (e.g., "priority: high"),

2) Fit for ≤1 hour (quick wins first),

3) Recency/clarity of the item.

When in doubt, take the first three [new] items in file order.

5) Per-task journal (agent_work/)

• When beginning work on an item: create a folder agent_work/YYYY-MM-DD_HHMM_nickname.
• Inside, maintain a running log: start time, linked todo text, decisions, commands, files created, errors, attempts, results.
• For harder tasks include a small plan with time estimate and critical steps.
• Place any generated files/documents inside this folder.

6) Reporting

• Every report must name the todo/project it concerns.
• For ≤1h tasks done now: include Result, Key steps, and any links to outputs.
• For >1h tasks: include Plan (steps + estimate), Current status ([new|wip]), and where the journal folder lives.

Categorization policy

When the owner/requester says "add a todo":

• Assign exactly one primary category from: quick search | setup | prototype | try out | learn | write.
• Add one generation tag: code | no-code.
• Append to todo/todo.md (today's section) using:
• "- [new]