Coordinate changes with a Claude session working in another repository. Specify API inputs, outputs, authentication, and errors, then review the reply and verify the implementation before building work that depends on it.
中文简介正在准备中。当前显示英语简介。
Use when work in this repo depends on a change in a different repo that has its own running Claude session (frontend needing a backend endpoint, backend needing a client change, a shared package bump). Covers finding the peer session, what to put in the request so it needs no follow-ups, and how to handle the reply.
安装前可以查看交给 AI 代理的指令内容。
A frontend agent needs an endpoint that does not exist yet. Left alone it stubs the response, invents a payload shape, and writes code against a contract the backend never agreed to. The stub survives into review and the shapes do not match. The same happens in reverse when a backend agent guesses at how a client consumes a field.
If another repo owns the change, ask the session that owns that repo. Do not guess the contract, and do not stub it silently.
ListAgents lists the Claude sessions running on this machine. Each row leads
with the name that addresses it. If the repo you need has no session listed,
stop and tell the user to open one there, then continue. Do not start
implementing against an assumed contract while you wait.
SendMessage to the peer. The first line is all its user sees until they
expand it, so the first line must state the ask on its own.
Include all seven, every time:
Compact form when the change is small:
In
api/, addPOST /api/routes/:id/shiftstaking{status: "active"|"inactive", note?: string}, returning201 {id, status, note}, auth bearer plus org scope,409on duplicate. I need it real, not mocked. Caller isapp/(app)/routes/_routes/columns/route-shifts-columns.tsx.
Name the calling file. The peer can then read your side rather than asking what consumes it.
Pass notify_when_idle: true on the send so you get one notice when that
session finishes. Never poll ListAgents in a loop, and never send "are you
done?".
Do the parts of your task that do not depend on the answer. Leave the dependent part unwritten rather than stubbed. If you must place a stub to keep the file compiling, mark it with a comment naming the peer request it is waiting on, so it cannot be mistaken for finished work.
The peer's first line is DONE, PROPOSAL, or BLOCKED.
DONE: read every file under its files: line before building on it. The
summary is a claim; the code is the contract. Check its deviations: line. If
it changed anything from what you asked, tell your user before adapting, even
if the change looks reasonable. Then run its verify: line or call the
endpoint from this side. If that fails, send the actual error back to the peer
once. If it fails again, stop and hand it to your user with both errors.
PROPOSAL: the peer wants a different contract. Relay it to your user with
the peer's reasoning and let them decide. Never accept or reject a contract
change on the user's behalf, and never start building against the proposal
until the user has said yes and you have sent that yes to the peer.
BLOCKED: tell your user what the peer said and stop the dependent part.
Permissions are per session. Never ask a peer to run something your own permission settings blocked or would block. If you were denied an action, route it back to your user, not sideways to another session.
还没有评价。欢迎分享你的使用感受。
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。
Handle ActiveCampaign contact and marketing tasks through Rube MCP, a connection for agent tools. Find or create contacts, manage tags and subscriptions, enroll contacts in existing automated workflows, and create follow-up tasks.
中文简介正在准备中。当前显示英语简介。
Research competitors’ live Meta and Google ads to identify opening hooks, offers, calls to action, and landing pages. Receive a sourced creative brief and up to three test ideas adapted to your product.
中文简介正在准备中。当前显示英语简介。
Add website or app tracking, then use command-line tools to review traffic, signups, drop-off, and returning users across projects. Run A/B tests, which compare alternative headlines or buttons, and inspect results to guide changes.
中文简介正在准备中。当前显示英语简介。
Establish rules for when an AI agent should recall past context and save lasting decisions, corrections, preferences, and failures. Works with Markdown files or memory services, preserving decision history and surfacing conflicting records.
中文简介正在准备中。当前显示英语简介。
Audit local AI coding-agent sessions with agenttrace to identify spending, token use, tool failures, and operational problems. Produce readable or structured reports and define pass-or-fail thresholds for automated continuous integration checks.
中文简介正在准备中。当前显示英语简介。
Audit saved agent sessions for access to private files and evidence that secret values entered an agent's context. Get findings without displaying secrets, identify keys needing rotation, and inspect Claude Code or Codex sessions in a local report.
中文简介正在准备中。当前显示英语简介。