task-management

Task Management (Lark-native)

Lark Task is the system of record. TASKS.md / dashboard.html in the working directory are an optional local mirror for offline editing — never the source of truth. Always read live state from Lark before answering, and write changes back to Lark.

Read ../../CONNECTORS.md and the depth core (LARK-PATTERNS, LARK-RECIPES, LARK-FUSION). For anything beyond create/list/complete, delegate to the installed lark-task skill.

Read — "what's on my plate" / "what's overdue" / "due this week"

1. lark_task_my with a jq projection (P3 token economy):
   • on your plate now: { "complete": false, "jq": ".data.items[] | {id, summary, due, completed}" }
   • overdue: add due_end = now; due this week: due_start=now, due_end=+7d.
2. Rank by deadline × importance (for a real ranking, delegate to the task-prioritizer skill).
3. Surface as an interactive card, not plain text (P4): a brief header, one item row per top task with a side button (e.g. Complete / Snooze), and an actions footer. Use lark_im_card_send with print_json: true first to validate. Fall back to plain text only if the user explicitly wants it inline.

Add — "add a task" / "remind me to" / "assign X to "

1. If the task is for someone else, resolve the assignee first (P1): lark_contact_search(query="<name>") → open_id.
2. lark_task_create with dry_run: true (P2) — show summary + assignee + due — then commit on confirm. Pass due as date:YYYY-MM-DD / relative:+2d, and assignee as the open_id.
3. Mirror into local TASKS.md only if the user keeps the offline dashboard.

Complete — "done with X" / "finished X"

1. Find the task id via lark_task_my (match on summary).
2. Complete it with lark_task_complete (curated, safe-mutation): pass the task id and dry_run: true first to preview, then commit on confirm.
3. Confirm completion back to the user.

Extract — from meetings & conversations (P6)

When summarizing a meeting or chat, offer to capture action items into Lark Task:

1. Prefer real Minutes artifacts: lark_minutes_search(participant_ids="me", query=...) → pull the AI action items instead of re-deriving them. (im search for chat-only threads.)
2. For each item: resolve owner (P1) → lark_task_create(dry_run) → confirm → commit.
3. Optionally log decisions to a Base (P5) and post a recap card (P4).

Always ask before adding — never auto-create tasks without confirmation (P2).

Conventions

• Bold/clear task summaries; include "for [person]" for commitments and an explicit due.
• Keep replies scannable; a card with ≤7 rows beats a long list.
• Tasks assigned to others aren't in lark_task_my — use the task list endpoint (LARK-RECIPES).

Local mirror (optional)

If the user wants the offline dashboard, keep TASKS.md + dashboard.html in the working dir as a view, and reconcile on /productivity:update: pull lark_task_my, diff against TASKS.md, and offer to add/complete the deltas — Lark wins on conflict.