本文へ移動
cccskills
無料GitHub で公開

morning-standup

Use when: Good morning Team, start-of-day briefing, or morning status request. Architect orchestrates; Builder, Tester, and Router return their sections as subagents.

インストール方法を見る

含まれるファイル(1)

  • SKILL.md7.4 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Morning Standup

Start-of-day briefing routine. Triggered by the team lead saying "Good morning Team" (or variants). Architect orchestrates; Builder, Tester, and Router are invoked as subagents and return their sections to Architect.

Trigger

Phrases: "Good morning Team", "Good morning", "Morning Team".

Read your agent identity and follow the section below that matches.


Architect — Orchestrator

Rules

  • Read-only. No file writes, no git operations, no memory updates.
  • Bullet points only. The team lead will ask for detail if needed.
  • Dispatch Builder, Tester, and Router as subagents in parallel before compiling the briefing.
  • Read the live memory file first. Use only the latest available archive snapshot for additional context on carry-forward items or yesterday's outcomes.
  • Do not scan all historical archive files by default.

Protocol

Step 1 — Read Architect's memory

  • Read the HOT, WARM, and COLD sections via the active storage plugin using read-memory('architect')
  • Read the latest available archive snapshot via the active storage plugin if one exists (the concrete path is defined in the active plugin file)
  • Prefer the live memory record for current state. Use the archive snapshot only for context on carry-forward items, blockers, and yesterday's outcomes.

Step 2 — Dispatch subagents

Run Builder, Tester, and Router as subagents simultaneously with the morning-standup prompt. Also dispatch the coordination agent (CAO) as a subagent with the same morning-standup prompt when and only when CAO's own profile file exists in the hub — probe the adapter-matching path (e.g. .claude/agents/cao.md for Claude Code); skip silently if absent. Collect their reports. When dispatching Router, use the fastest/economy-tier model your runtime offers for the dispatch — Router is economy-tier per its manifest. This skill file ships identically to every adapter, so it never asserts one runtime's literal model-selection syntax; see ROUTING.md's "Dispatch Builder" model-selection guidance for how each adapter's own generated instructions resolve this to a concrete identifier.

Step 3 — Compile and deliver briefing

Good morning. Team brief for [DATE].

**Architect**
- [HOT item 1 — one line status]
- [HOT item 2 — one line status]
- Today's focus: [from architect.md or latest archive Carry-Forward section]
- Decisions needed: [any team lead decision or approval needed, or "None"]

**Builder**
[paste Builder's bullets verbatim]

**Tester**
[paste Tester's bullets verbatim]

**Router**
[paste Router's bullets verbatim]

**CAO**
[paste CAO's bullets verbatim if present, or omit this block if CAO is not configured]

**Needs team lead today**
- [consolidated list of decisions, approvals, or inputs required across all agents — or "Nothing urgent"]

No closing questions. No narrative. Deliver and stop.


Builder — Subagent

Rules

  • Invoked as a subagent by Architect.
  • Read-only. No file writes, no git operations, no memory updates.
  • Return bullet points only to Architect — do not deliver directly to the team lead.

Protocol

Step 1 — Read Builder's memory

  • Read the live memory record via the active storage plugin using read-memory('builder') if it exists
  • Read the latest available archive snapshot via the active storage plugin if one exists (the concrete path is defined in the active plugin file)
  • Prefer the live memory record for current state. Use the archive snapshot only for context on carry-forward items and yesterday's outcomes.

Step 2 — Return brief

**Builder**
- Active: [current branch or ticket — one line, or "No active work"]
- Yesterday: [one-line outcome summary, or "No sessions"]
- Blockers: [blocked on decision or environment issue, or "None"]

Three bullets maximum unless there are multiple active items. If nothing to report, return:

**Builder**
- Clear. No active work.

Tester — Subagent

Rules

  • Invoked as a subagent by Architect.
  • Read-only. No file writes, no git operations, no memory updates.
  • Return bullet points only to Architect — do not deliver directly to the team lead.

Protocol

Step 1 — Read Tester's memory

  • Read the live memory record via the active storage plugin using read-memory('tester') if it exists
  • Read the latest available archive snapshot via the active storage plugin if one exists (the concrete path is defined in the active plugin file)
  • Prefer the live memory record for current state. Use the archive snapshot only for context on carry-forward items and yesterday's outcomes.

Step 2 — Return brief

**Tester**
- QA queue: [tickets in state:ready-for-qa — list briefly, or "Clear"]
- Yesterday: [one-line outcome summary, or "No QA runs"]
- Blockers: [environment issues, missing data, or "None"]

Three bullets maximum unless there are multiple queue items. If nothing to report, return:

**Tester**
- QA clear. Nothing pending.

Router — Subagent

Rules

  • Invoked as a subagent by Architect.
  • Read-only. No file writes, no git operations, no memory updates.
  • Return bullet points only to Architect — do not deliver directly to the team lead.
  • If no remote team is configured, return: **Router** - Not configured. No relay activity.

Protocol

Step 1 — Read Router's memory

  • Read the live memory record via the active storage plugin using read-memory('router') if it exists
  • Read the latest available archive snapshot via the active storage plugin if one exists (the concrete path is defined in the active plugin file)
  • Prefer the live memory record for current state. Use the archive snapshot only for context on carry-forward items and yesterday's outcomes.

Step 2 — Return brief

**Router**
- Routing: [any unresolved classifications or pending escalations — or "Clear"]
- Yesterday: [one-line summary of routing activity, or "No routing events"]

Two bullets maximum. If nothing to report, return:

**Router**
- Relay clear. Nothing pending.

CAO — Subagent

Rules

  • Invoked as a subagent by Architect when CAO's own profile file exists in the hub.
  • Read-only. No file writes, no git operations, no memory updates.
  • Return bullet points only to Architect — do not deliver directly to the team lead.
  • If CAO is not configured, this section does not run.

Protocol

Step 1 — Read CAO's memory

  • Read the live memory record via the active storage plugin using read-memory('cao') if it exists
  • Read the latest available archive snapshot via the active storage plugin if one exists (the concrete path is defined in the active plugin file)
  • Prefer the live memory record for current state. Use the archive snapshot only for context on carry-forward items and yesterday's outcomes.

Step 2 — Return brief

**CAO**
- Active: [current active leads, offers, or calls — one line, or "No active work"]
- Yesterday: [one-line outcome summary of lead triage or deals, or "No sessions"]
- Blockers: [pending approvals or founder decisions, or "None"]

Three bullets maximum unless there are multiple active items. If nothing to report, return:

**CAO**
- Clear. No active work.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Universal guardrails loaded by every agent. Defines Verify Before Reporting (VBR), Write-Ahead Log (WAL), and the security baseline. Always-on, role-independent.

日本語の概要は準備中です。原文の説明を表示しています。

Agistra/agistra.dev52026年10月7日 更新

Use when: architecture mode, architect this, ADR, C4, PlantUML, system boundary, design decision, PRD gap, high ambiguity, multiple technical approaches, security architecture, data architecture, integration risk, implementability gap, story ticket, or epic ticket.

日本語の概要は準備中です。原文の説明を表示しています。

Agistra/agistra.dev52026年10月7日 更新

Use when: assumptions audit, audit assumptions, check for hidden assumptions, plan/ticket has ambiguous scope edges, pre-flight before finalizing a ticket, unstated assumptions, or 'what am I assuming'. Surfaces unstated assumptions, ambiguous scope edges, and untested preconditions in a finished plan/ticket/ADR before it is handed to Builder. Optional pass — Architect judges when to apply it; not a mandatory gate on every ticket.

日本語の概要は準備中です。原文の説明を表示しています。

Agistra/agistra.dev52026年10月7日 更新

Use when: Tester needs UI state evidence for VBR, or Builder needs to verify an integration wire is observable end-to-end. Powered by agent-browser MCP — token-efficient browser automation (200–400 tokens/page).

日本語の概要は準備中です。原文の説明を表示しています。

Agistra/agistra.dev52026年10月7日 更新

Conducts multi-axis code review. Use before merging any change. Use when reviewing code written by yourself, another agent, or a human. Use when you need to assess code quality across multiple dimensions before it enters the main branch.

日本語の概要は準備中です。原文の説明を表示しています。

Agistra/agistra.dev52026年10月7日 更新

Writing skill for freelancers and independent consultants. Covers project proposals, bids, client emails, and scope summaries. Leads with the client's problem, not the consultant's background.

日本語の概要は準備中です。原文の説明を表示しています。

Agistra/agistra.dev52026年10月7日 更新

Agistra のスキルをすべて見る

このスキルの問題を報告する