菜单
Align project background, initialize runtime trace / memory files, and output the minimum startup summary for the current task.
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。
基于 SOC 职业分类
Used for high-quality experiment reports, analysis reports, and technical summaries. Evidence must be enough, charts must be real, structure must be clear, and empty talk is forbidden.
Open several fundamentally different routes for one problem fast, so the flow does not keep doing small tweaks on one path only.
Split a complex problem into a multi-step reasoning chain, and make key assumptions, branch points, validation order, and rollback points explicit.
Manual start skill: read `.opencode/` first, then read the task files at the project root and start execution. It is like replacing the first prompt line.
Built-in Python tool priors covering time series, statistics, machine learning, and related ecosystems. It reduces repeated search cost and gives the first coverage of method families and tool families.
In time-series, deep-learning, and LLM projects, use ztxexp to unify experiment config, batch runs, result tracking, ablation analysis, and failure diagnosis.
| name | workspace-init |
| description | Align project background, initialize runtime trace / memory files, and output the minimum startup summary for the current task. |
[Skills: workspace-init] Follow: <rules / context constraints that really apply now>; Current step: <one-line note>.[Rules: <rule>] ..., then continue with the real work.bash .opencode/skills/workspace-init/init.sh
README.md, task notes, and existing .opencode/trace.md, to align project background and context.interface mode from execution strategy: run / tui are interface modes, while autonomous / interactive are execution strategies. If there is no stronger signal, the default execution strategy is autonomous; run + interactive is not a supported default path..opencode/trace.md or any .opencode/memory/*.md already exists, read its current contents before any later write, append, or overwrite. Do not treat runtime files like empty templates that may be rebuilt blindly..opencode/memory/initial-prompt.md is initialized, the main agent must write the original user prompt, earliest task goal, explicit metrics, and non-goals for this round into it at once as an append-only baseline. Later it may only append clarifications, not overwrite the original prompt block..opencode/memory/context-snapshot.md is initialized, the main agent must refresh the first snapshot before workspace-init closes: at minimum it should write task anchors, current phase, explicit constraints, the most obvious gaps, and the default next-dispatch focus. This is not an optional summary. It is the canonical entry point for later multi-agent compaction.workspace-init baseline is established, it should create the first local git commit soon, then keep committing again at plan switches, key milestones, major rollback points, and the final stable pre-delivery state..opencode/protocols/ and .opencode/evals/, especially dispatch, compaction, memory-sync, and runtime-events, so it does not initialize memory only and miss runtime protocols or eval contracts..venv already exists.python-version, pyproject.toml, environment.yml, requirements*.txt, or uv.lock.venv > follow project constraints > create a workspace-root .venv.autonomous and the environment detection found no real conflict, follow that default priority directly instead of asking the user about interpreter preferences.interactive, still do the same local detection first; only when multiple equally reasonable environment candidates remain and they would materially change the later path may this fork be escalated back to the main agent for a short user confirmation.interactive does not mean asking on every step. It only means exposing real forks to the user after local detection has already narrowed the space.