Syntax and options for `uip codedagent`, a bridge to the Python UiPath CLI for building and running Coded Agents.

uip codedagent is a thin bridge to the Python-based UiPath Coded Agents workflow. It detects a suitable Python interpreter, confirms the uipath Python package is installed, and forwards a fixed whitelist of subcommands into the Python CLI — so new, dev, pack, publish, and other lifecycle verbs come from the Python package rather than this tool. review is the one exception: it's a native command implemented directly in this tool, not forwarded.

Refer to Tools (plugins) for how thin wrappers differ from full tools.

setup, review, and any forwarded command honor the global options (--output, --output-filter, --log-level, --log-file). Exit codes follow the standard contract.

Review a coded agent project against a deterministic rule set — no Python interpreter needed, this runs natively in the CLI. Returns a verdict (PASS/FAIL), a score (0-100), a letter grade (A+...F), and a list of issues.

An unrecognized --checks value fails with Code: "CodedAgentReviewInvalidChecks", listing the known categories.

Detect Python and verify the uipath package is installed. Results are cached in a per-user file so subsequent codedagent commands can resolve the Python binary instantly. The cache is shared with uip function setup, so running either command configures Python for both.

When a virtual environment is active (VIRTUAL_ENV is set), setup binds the Python interpreter and the uipath executable from that environment. A cached result whose uipath executable is outside the active virtual environment is detected again.

None.

A fixed whitelist of subcommand names is forwarded to the uipath Python CLI as codedagent exec <subcommand> ...: add, debug, deploy, dev, eval, init, invoke, list-models, new, pack, publish, pull, push, register, run. Anything outside this list — including a typo — is rejected by Commander as an unknown command before any forwarding is attempted; it is not passed through to Python.

Auth relay: before forwarding, the CLI reads the session created by uip login and injects these environment variables into the Python subprocess so the SDK picks them up automatically:

If you are not logged in, forwarding still works; the Python CLI runs without those variables.

--force is always stripped: this tool defines no command-specific rules for --force, so it is unconditionally and silently stripped from every forwarded command's arguments before they reach the Python CLI. It never reaches Python, and no error is emitted — passing --force to a forwarded subcommand is simply a no-op.

push requires UIPATH_PROJECT_ID — a required environment variable, checked natively (in this tool, before forwarding to Python), not something the Python CLI enforces on its own. Set it once per project:

Get <id> from an existing Coded Agent project in Studio Web. The check is skipped for uip codedagent push --help.

deploy/publish --my-workspace (or -w) requires a personal-workspace-capable login scope — another native, pre-forwarding check. It reads your current uip login access token and requires the umbrella OrchestratorApiUserAccess scope or the granular OR.Users or OR.Users.Read scope (any one satisfies it); a token missing all of them fails before anything reaches Python:

The default uip login client picks up OrchestratorApiUserAccess automatically, so this only bites custom clients or a restricted scope list. Skip the check entirely by deploying to the tenant feed instead: uip codedagent deploy --tenant. This check only applies to deploy/publish when --my-workspace/-w is passed — it does not run for a plain deploy/publish or any other whitelisted command.

Standard exit codes apply. The forwarded subprocess's exit code is relayed directly to the parent shell, so a Python-side failure exits uip codedagent with the same non-zero status.