Menu
Use when working with source control for a project within /home/user/workspace/code or code-agent.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| 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.
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`.