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/nicsuzor/academicOps --skill pull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。
Standing experiment protocol for the academicOps project — every piece of real work is also evidence about the framework. Use when starting framework work of any type, testing instructions blind, trialling a runtime mechanism, or reviewing how well the framework served a task just completed.
Creating diagrams in any style — Mermaid flowcharts (structured, code-based) or Excalidraw (hand-drawn, organic). Use style parameter to select.
Use when driving a framework run you intend to score — choosing a surface, dispatching a worker, or when asked to "debug a polecat", "run a polecat container interactively", "attach to a polecat session", "check polecat logs", or to verify that a change to plugins, hooks, lib/, skills, or the Dockerfile actually works inside a real container. Spins up a `polecat run` container under tmux for live interaction, says where the durable host-side session state lands, and walks the layered check that separates "installed in the image" from "actually fires".
正在显示 SKILL.md
| name | pull |
| description | Claim a queued task, execute it, record the result on the task, and hand over. |
Each task was vetted before it became available. Execute it faithfully: it is the whole of your obligation and the limit of your authority.
Work counts only once it is recorded on the task. In an ephemeral container or worktree your changes exist nowhere else — halt without completing the handover and they are destroyed.
Claim your task through the knowledge base and synchronise it with your internal TODO list.
The claimed task is your unit of delivery, children included. Existing children are input to your plan: execute them directly in sequence or in parallel where the work allows. The return contract attaches to the claimed task — one deliverable, with evidence and an output URL — never a spray of per-child deliverables.
Track every step and deliverable on your native task list, including all outstanding subtasks, then a verification step, and finally "Handover".
Execute the steps systematically, sequencing in parallel where the work allows.
Refuse and attempt. Refuse any choice not derivable with reasonable confidence from the axioms plus the context you were given — that is the same limit on your authority, applied to decisions. Attempt everything that does not depend on a refused choice, then take dump's partial path.
Failures are routine and provide valuable information.
Complete what steps you can and cut at a clean seam. Mark what is incomplete and why.
Check your work against the literal requirements and acceptance criteria set out in the task, and carry the evidence for each into the report you hand back — the brief's evidence bar is what your claims are admitted against. Technical compliance is not sufficient and quality assurance is not a checklist; the bar is excellence. Rectify what falls short.
Do not certify a task complete without certainty that it is delivered in full.
Invoke the dump skill. It records your work and lets the task proceed to the next stage.
dump skill or your work will be DESTROYED.