菜单
Use when working with source control for a project within /home/user/workspace/code or code-agent.
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。
Turn a recent human-agent interaction into a portable workflow description.
Work through a proposal by starting a tracked /goal, reading the proposal, breaking it into milestone-sized tasks, and delegating each implementation milestone to fresh subagents instead of implementing directly in the main agent. Use when the user wants proposal-driven execution with explicit progress tracking and milestone-by-milestone delegation.
Draft new proposal documents for a codebase in the local proposal style. Use when the user wants a new proposal written before implementation, or wants an idea turned into a structured design/proposal document with motivation, problem statement, proposal, non-goals, verification, and success criteria.
Manage npm package dependencies in this environment. Use when adding or updating npm packages, editing package.json/package-lock.json, or when network access must be constrained to `run-in-nw-sandbox -- npm install --no-scripts [--save | --save-dev] -- package-spec ...` followed by `npm rebuild` outside the sandbox.
How to use the run-in-nw-sandbox helper to run commands with network access
Manage Rust/Cargo dependencies in this environment. Use when adding, updating, or fetching crates, editing `Cargo.toml`, or working within the constraint that the only approved network command is `cargo fetch` via `run-in-nw-sandbox`.
| name | jj-protected-workspaces |
| description | Use when working with source control for a project within /home/user/workspace/code or code-agent. |
This environment separates canonical project repositories from agent working copies:
/home/user/workspace/code/<repo> contains canonical project repositories.
Treat these as read-only source-of-truth projects./home/user/workspace/code-agent/<workspace> contains writable agent workspaces.
Make file edits and run normal development commands from here.When a task names a repository, prefer the matching workspace under
code-agent/. If you are in code/, do not edit files there directly; create
or switch to the corresponding code-agent/ workspace.
Only a narrow set of jj mutations is currently allowed. Use these exact
argument forms; the allowlist is strict about argument order and shape.
From /home/user/workspace/code/<repo>:
jj workspace add ../../code-agent/<workspace> --name agent-<workspace>
From /home/user/workspace/code-agent/<workspace>:
jj workspace update-stale
jj commit -m "message" [<fileset>...]
jj desc [<revset>] -m "description"
jj describe [<revset>] -m "description"
jj new [<revset>]
jj squash [--from <revset>] [--to <revset>] [<fileset>...]
jj edit <revset>
jj abandon [<revset>...]
Do not use other mutating jj commands.
The immutable-heads configuration is the protection boundary. An agent cannot
rewrite ancestry of default@, bookmarked changes, or remote-bookmarked
changes, but non-immutable unbookmarked revisions are considered mutable
working area and may be changed by allowed revset-based commands.
Read-only inspection commands are permitted and should be used as needed:
jj status
jj log
jj show
jj diff
Use these commands to inspect history, review diffs, and verify the working copy before committing.