跳到正文
cccskills
免费在 GitHub 公开

team-relay

Use when: remote team, delegate, delegated lane, delegated implementation, delegated QA, or non-sensitive ticket routing to the remote team's implementation lane.

中文简介正在准备中。当前显示原文说明。

集成工具Telegram
查看安装方法

包含的文件(1)

  • SKILL.md6.1 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Team Relay

Use this skill when Builder needs to coordinate delegated work with the remote team. The remote team has a single AI agent (the remote agent) who does both implementation and first-pass QA.

Builder remains the accountable engineering lead throughout the delegated lane. The remote agent implements and runs first-pass QA. Builder reviews all code before QA. The team lead is the merge and close authority.

All inter-team communication runs through Router via the configured relay channel — see telegram-relay. Builder never contacts the remote agent directly; Builder dispatches Router when a state transition warrants a notification.

Delegation Decision

  • Sensitive tickets stay in Builder's direct lane and route Builder → Tester.
  • Non-sensitive tickets may be delegated and route Builder → remote agent → Builder review → remote agent (QA) → Builder → team lead.
  • Tester is outside the default delegated lane and returns only when the team lead explicitly requests re-validation.

A ticket is sensitive if it touches: credentials, payment flows, privileged data, or confidential business logic. When in doubt, keep it in the direct lane.

If scope drifts into a sensitive area mid-ticket, pull it back into Builder's direct lane before more code is written.

Delegated Lane Workflow

  1. Builder reviews the ticket and confirms it is safe to delegate.
  2. Builder writes the handoff in the GitHub issue using the template below and moves the ticket to state:ready-for-implementation.
  3. Builder dispatches Router to notify the remote agent per telegram-relay state-transition trigger table.
  4. Remote agent implements the scoped work, opens the PR, runs validation, moves the ticket to state:ready-for-review.
  5. Builder reviews the code. If changes are needed, Builder moves to state:changes-requested and dispatches Router with a notification; loop continues until Builder accepts the implementation.
  6. Only after Builder accepts the implementation does Builder move the ticket to state:ready-for-qa. Builder dispatches Router with the QA-ready notification.
  7. Remote agent runs first-pass QA and returns a verdict:
    • On PASS: moves the ticket to state:qa-passed and dispatches Router with the QA-passed notification.
    • On FAIL / PARTIAL PASS: moves the ticket to state:changes-requested with defect evidence on the ticket; dispatches Router with the QA-failed notification.
  8. On FAIL, Builder triages: either the remote agent loops on the fix, or Builder pulls the work back into the direct lane if the defect implies a sensitive change.
  9. On PASS, Builder adds the final engineering recommendation while the ticket sits in state:qa-passed for team lead review.
  10. The team lead remains the only merge and close authority.

Builder Responsibilities in the Delegated Lane

  • Scope the ticket fully before delegating (all acceptance criteria testable, all assumptions documented).
  • Write the scoped handoff for the remote agent (template below).
  • Dispatch Router at each state transition that crosses the inter-team boundary, per telegram-relay.
  • Review all remote-agent code before moving to state:ready-for-qa.
  • Triage remote-agent QA findings and decide whether Tester re-validation is needed (only when the team lead asks).
  • Prepare the final engineering recommendation for the team lead after PASS.

Delegated Lane State Table

StateMeaningOwner
state:ready-for-implementationBuilder has scoped the ticket and assigned the delegated laneBuilder
state:in-progressRemote agent is actively implementingRemote agent
state:ready-for-reviewDelegated implementation complete and waiting for Builder reviewBuilder
state:changes-requestedBuilder rejected the implementation, or remote-agent QA returned defectsRemote agent (to fix)
state:ready-for-qaBuilder approved the code; remote agent may now run first-pass QARemote agent (QA)
state:qa-passedRemote-agent QA passed. Builder adds engineering recommendation for team lead review.Builder (recommendation) → team lead (merge)

Builder → Remote Agent Handoff Template

Post this as a comment on the GitHub issue when the ticket moves to state:ready-for-implementation.

## Builder → Remote Agent Handoff

**Ticket:** #N
**Implementation owner:** Remote agent
**Branch plan:** [new branch name or "remote agent chooses"]

**Acceptance Criteria**
- [ ] [criterion 1]
- [ ] [criterion 2]

**Environment:** [URL or environment name for the deployed build]

**Credentials or Test Data Source:** [where the remote agent gets access — not the secret itself]

**Scope Boundaries**
- In scope: [what to build]
- Out of scope: [explicit exclusions]

**Assumptions**
- [assumption and verification need]

### Test Focus (used by remote agent in QA mode after Builder review)
- [area to focus testing on]

### Validation Required (before remote agent moves ticket to state:ready-for-review)
- [ ] Build passes
- [ ] Lint passes
- [ ] Unit tests pass
- [ ] Manual smoke test: [specific steps]

### Communication Protocol
- After implementation: return owner Builder, state state:ready-for-review.
- After Builder approves and moves to state:ready-for-qa, remote agent runs first-pass QA.
- Dispatch Router on each state transition per `telegram-relay`.

Tester Re-Validation

Only include Tester if the team lead explicitly requests re-validation of a remote-agent-QA'd ticket. The route in that case:

  1. The team lead asks Builder for Tester re-validation.
  2. Builder creates a fresh Developer → QA handoff using ticket-lifecycle-mode.
  3. Tester runs full QA as Architect classifies it: a subagent in Full QA mode for non-browser tickets, a direct Tester session for browser-driven QA.
  4. Tester's verdict updates the ticket state. Router is not involved in this route since it is an internal team escalation.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

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

中文简介正在准备中。当前显示原文说明。

Agistra/agistra.dev52026年10月19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.

中文简介正在准备中。当前显示原文说明。

Agistra/agistra.dev52026年10月19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.

中文简介正在准备中。当前显示原文说明。

Agistra/agistra.dev52026年10月19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).

中文简介正在准备中。当前显示原文说明。

Agistra/agistra.dev52026年10月19日 更新

Use when drafting a bug or defect ticket, reviewing one someone else wrote, or deciding whether a reported problem is ready to file. Keeps what the reporter actually saw separate from what the agent later found out, so a reader can tell which lines are testimony and which are collected proof. Trigger phrases: 'file a bug', 'write up this defect', 'triage this report', 'is this ticket ready to file', 'what did the reporter actually see'.

中文简介正在准备中。当前显示原文说明。

Agistra/agistra.dev52026年10月19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.

中文简介正在准备中。当前显示原文说明。

Agistra/agistra.dev52026年10月19日 更新

查看 Agistra 的全部技能

报告此技能的问题