assess-and-register

name	assess-and-register
description	Activate before executing any substantive work that produces artifacts or requires multi-step execution (coding, data analysis, report generation, multi-agent chains). Guides you to check past epics for context, register a new Epic/Story/Task when none exists, mark tasks as doing before execution, and close them after completion.

Assess & Register SOP

When to Use

Activate for any request that produces a physical artifact (code, data file, PDF, chart) or requires delegating to a sub-agent.

Skip this skill for:

• Pure conversation, greetings, or time queries
• Creating a reminder or regular work (they have their own tracking)
• User explicitly says "直接做" or "不用記錄"
• Simple one-shot lookup with no artifact output (e.g. get_epics, get_user_profile)

---

Step 1 — Assess Past Records

Call get_epics(user_id=EXTRACTED_ID, status="all") to check all projects (active, completed, archived).

• Relevant epic found → call get_epic_detail(epic_id) to read what was already done. Use this context when planning the current request (continue, extend, or correct prior work).
• No relevant epic → proceed to Step 2.

---

Step 2 — Register the Work

Timing: For single-step requests, register immediately before executing. For multi-step requests requiring user plan confirmation (see multi-agent-chain skill), register after the user confirms the plan and before calling any specialist agent tool.

Create the project structure in order:

1. create_epic(user_id, title, description) — one epic per topic/project
2. create_story(epic_id, user_id, title, priority) — one story per major phase (e.g., "Backend API", "Data Analysis", "Report Generation")
3. create_project_task(...) — one task per concrete unit of work; all tasks are created with status=backlog
4. update_task_status(task_id_1, "doing") — mark only the first task as doing; all other tasks remain backlog until their turn

Then confirm to the user in one short line:

"已建立專案記錄，開始執行…"

---

Step 3 — Per-Task Completion (CRITICAL: close each task immediately, do not batch)

After each sub-agent returns a completion signal, run the acceptance-check skill before marking done:

1. Activate acceptance-check skill — verify all required output files exist
   • Files confirmed → proceed to step 2
   • File missing → skill requeues and retries; only proceed to step 2 after retry passes or task is marked failed
2. add_task_comment(task_id, type="result", content="...") — record the output file path or result summary
3. update_task_status(next_task_id, "doing") — if a next task exists, mark it as doing now, then immediately call its matching specialist agent tool (<agent_name>(request='...'))

After all tasks are done: 4. update_story(story_id, status="done") — close the story