用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/Pubx-ai/skills --skill planner命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Evaluate whether AdCP (Ad Context Protocol, agenticadvertising.org) — or a concrete AdCP implementation such as a seller agent, buyer agent, or orchestrator — is safe, interoperable, operationally viable, and ready for pilot or production use. Combines a weighted decision scorecard, hard safety gates that override any numerical score, and an observable pilot test suite run against a sandboxed or live agent. Use when asked to "score AdCP", "evaluate AdCP", "run the AdCP scorecard", "is AdCP (or our seller agent) production-ready", "pilot-test the AdCP agent", "assess AdCP readiness", or to compare AdCP against another agent protocol with a weighted rubric. DO NOT USE for reviewing code changes for AdCP spec conformance (use adcp-review) or for routine operations against a seller agent — discovery, media buys, updates, deliveries (use adcp-seller-agent).
Review code changes for conformance with the AdCP (Ad Context Protocol, agenticadvertising.org) technical spec. Use whenever reviewing a PR, diff, commit, or code change in a repository that implements AdCP — seller agents, buyer agents, orchestrators, or SDK integrations — even if the request is just "review this PR" or "check my changes" without mentioning AdCP explicitly. Covers task request/response wire shapes, async task lifecycle and status handling, idempotency, error envelopes, authentication/signing, webhook security, and media-buy semantics. Also use when asked whether a change "breaks the protocol", "is spec-compliant", or "will interop with other AdCP agents".
Load every comment on a pull request, understand and triage each one against the current code, then resolve them — fixing what's valid, pushing back with technical reasoning on what isn't. Use this skill whenever the user asks to "address the PR comments", "review the PR comments", "respond to review feedback", "handle the CodeRabbit / Bugbot / reviewer comments", "triage the review comments", "fix the PR feedback", or "resolve the comments" on a pull request. Works for human and bot reviewers alike. If a skill named "receiving-code-review" is available in the session, this skill applies its verification discipline rather than duplicating it: verify before implementing, no performative agreement, technical correctness over social comfort.
基于 SOC 职业分类
正在显示 SKILL.md
| name | planner |
| description | Turn a completed requirements document into a detailed, sequential implementation plan. Alias for the create-plan skill. |
| disable-model-invocation | true |
Run a /dev:create-plan session.
