integrate

/integrate

Run the installed integration utility from the target repository's working directory. The utility resolves the Git root itself and must be invoked from a repository. It performs the safety checks, optional test-server cleanup, main branch update, and fresh-branch creation described below.

Runtime script

Use the user-scope installed script:

${CLAUDE_HOME:-$HOME/.claude}/scripts/integrate.sh

Require that path to be executable before running it. Do not look for, require, copy, or invoke ./integrate.sh or any other repository-local copy. The repository's scripts/integrate.sh is the export source used by the installer; the runtime entry point is the copy under the active Claude home.

Argument mapping

Interpret the text after /integrate as the script's explicit argv list. The supported forms are one optional branch name and these flags:

• no arguments: run the script with no arguments;
• <branch-name>: pass that branch name as one positional argument;
• --force: pass the force flag;
• --new-branch: pass the new-branch flag;
• --help or -h: pass the help flag.

Preserve every supplied argument and its order when constructing argv. Do not use $@, eval, unquoted expansion, or pass the entire $ARGUMENTS string as one shell argument. Let the script apply its own argument validation and multiple-name warning rather than silently changing that behavior. Run it from the current target repository directory and report its exit status and output.

Examples:

/integrate
/integrate feature/my-feature
/integrate fix/bug-123 --force
/integrate --new-branch feature/my-feature

Workflow contract

• The script fetches origin/main and creates a timestamped dev<epoch> branch when no branch name is supplied.
• It skips test-server cleanup on main and detached HEAD.
• If test_server_manager.sh is absent at the Git root, it reports SKIPPED and continues.
• If the manager is present but not executable, or its stop command fails, it reports the original error and exits before checkout, stash, or branch deletion.
• A successful manager call proves only that the manager returned success; it does not prove that no orphaned processes remain.
• --force and --new-branch retain the script's documented safety semantics.

After the script succeeds, invoke /learn as a required completion step when learning persistence is authorized. A user request for /integrate includes this learning step; do not require separate approval or skip it because the user did not also type /learn. When an explicit no-memory instruction or higher-priority storage restriction applies, skip /learn and report it as skipped; otherwise, follow the canonical learn skill for all persistence targets and report each result. Honor higher-priority storage restrictions, report any blocked target, and never claim learning was saved when persistence is blocked. Run /testserver start when the new branch needs its test server started. Do not claim server cleanup or branch integration until the script output and exit status establish it, or learning capture until its persistence results are verified.