Inventory everything an agent injects into each request besides the user's message, such as system prompt, environment info, skill list, tool definitions, per-turn context blocks and reminders, and check necessity, frequency, precision, position and prompt-cache impact. Use when reviewing an agent's prompt assembly, chasing low cache hit rates, or trimming per-turn overhead.

Use when writing, rewriting, or reviewing PR/CL descriptions, commit messages, or code-change summaries that explain what changed and why. TRIGGER on "write PR description", "improve this commit message", "summarize this diff", "CL description", "change description", or "make this PR/CL description easier to review". DO NOT TRIGGER for code review, PR splitting, code restructuring, user-facing changelogs, or reviewer replies unless the requested output is a change description.

Use when planning, splitting, shrinking, or sequencing large features, refactors, migrations, or pull requests into small reviewable PRs or CLs. TRIGGER on "split this PR", "this diff is too large", "stacked PRs", "change sequence", "reviewable chunks", "migration plan", "how should I break this up?", or implementation scope too broad for one review. DO NOT TRIGGER for full code review, PR description writing, or tiny one-file edits that already form one self-contained change.

Review an agent's finished work by reading the evidence before its summary, naming what each check proves and what it misses, pulling out the decisions it made silently, and trying to falsify the result before accepting it. Use when a MiniMax Code session, subagent or cloud agent reports a task done, before merging its PR or acting on its analysis.

Run a harness change as a controlled experiment, with a small end-to-end eval, one change at a time, repeated runs, and looking at flipped tasks, request count, tokens and cache hits, not just the average score. Use before and after changing prompts, tools, skills, subagent layout, compaction or other agent scaffolding, or when asked whether a change helped.

Finds loops where the user keeps acting as a manual step for agents (retrying, judging output, approving, carrying context, starting the same task by hand, doing the agent's work) and proposes how to lift the user out: delete a stale rule, change a skill / AGENTS.md / playbook, add a script or metric, add a skill or scheduled task, or advise the user. Keeps a private loop ledger and promotes Claude Code auto-memory into artifacts. With a task already in the session, reflects on that session; in an empty session, scans recent local Claude Code, MiniMax Code, omp, pi and Codex sessions, a claude.ai export and an Obsidian vault. Triggers: /lifting-loops, retro, post-task reflection, "find repeated work", 反思一下, 复盘这次会话, 总结最近的会话, 找找重复劳动, 哪些事可以不用我; the weekly scheduled run; settling a proposal ("accept P-003", 采纳 P-003, 拒绝 P-003).

Audit how an agent harness runs long shell commands, covering foreground wait limits, moving to background without re-running, timeouts, completion notification, polling, output size, cleanup and behaviour after compaction. Use when building or reviewing an agent's shell or background-task tool, or when an agent hangs on dev servers, long builds or test runs.

Content production in three modes: rewrite, Xiaohongshu (小红书) image post, short video. Rewrite (default): turns an AI-sounding Chinese or mixed Chinese-English draft (post, newsletter, engineering blog, commentary) into the author's restrained style: few emoji, short paragraphs, hedged claims, neutral wording, English terms split by spaces. Triggers: 按我的风格改, 去 AI 味, 润色, 改改这篇, or a vague 帮我改改 when the context is a structured draft. Also for a Xiaohongshu image post from scratch (title, cover, cards, body, hooks); read references/xhs-post.md first. Also turns a document, research, notes or an article into a 60-180 s vertical video (抖音, 小红书, 视频号, B站: real sources, voice-over, captions, cover). Triggers: 做成视频, 做个短视频, 剪个 90 秒版本, 配音, 换成我的声音, or complaints of 字太密, AI 味太重, 容易走神; read references/video-pipeline.md first.

为已排期的需求做方案：列至少两个选项、写决策记录、评估波及面（blast radius）、定验收检查点，全部落进需求文档。Use when a requirement is scheduled and before implementation, or when someone asks 方案怎么定 / 这个改动会影响哪些地方.

上线前做就绪检查（回滚方案、可观测性、完成定义 DoD），全部有证据后再推进 Meegle 状态或节点。Use before a requirement is deployed or marked done, when someone says 可以上线了吗 / 准备发布, or before transitioning the work item to a released state.

按固定节奏跟进进行中的需求，用站会格式记录进展、阻塞和下一步；超过约定时间没有更新的需求要显式标出来，不能悄悄沉下去。Use for daily or weekly requirement check-ins, when asked 这个需求到哪了 / 有没有卡住的, or from a scheduled routine.

Decide whether a piece of work should go to a subagent or stay in the main agent loop. Use before spawning subagents, when designing a multi-agent architecture, or when auditing an existing set of subagents for merging back.

Turn a review correction we keep making on agent output into something that runs on its own, a test, lint rule, hook, CI step or project rule, so the miss does not come back next session. Use when we keep correcting agents on the same mistake (import order, naming, a forbidden API, a missing test), when the same review comment shows up on a second PR, or after a review finds a miss that a check could have caught.