- Overview
- Get started
- Concepts
- Using UiPath CLI
- How-to guides
- CI/CD recipes
- Command reference
- Overview
- Exit codes
- Global options
- uip codedagent
- uip coder
- uip context-grounding
- uip docsai
- uip function
- uip guardrails
- uip llm-configuration
- uip llm-gateway
- uip model-hub
- add-test-data-entity
- add-test-data-queue
- add-test-data-variation
- analyze
- build
- create-project
- diff
- find-activities
- get-analyzer-rules
- get-default-activity-xaml
- get-errors
- get-manual-test-cases
- get-manual-test-steps
- get-library-object-repository
- get-object-repository
- get-versions
- get-workflow-example
- indicate-application
- indicate-element
- inspect-package
- install-data-fabric-entities
- install-or-update-packages
- list-data-fabric-entities
- list-instances
- list-workflow-examples
- pack
- publish
- remote
- restore
- run, debug & execution
- run-file
- search-templates
- start-studio
- stop-execution
- tm
- uia
- uip tasks
- uip traces
- uip traces feedback
- Migration
- Reference & support
uip agent review
Score a low-code agent project against a deterministic rule set with `uip agent review`.
uip agent review runs a deterministic rule set over a low-code agent project and returns a verdict (PASS/FAIL), a score (0-100), a letter grade (A+ through F), and a categorized list of issues and strengths. It's a quality pass, distinct from uip agent validate's pass/fail schema check — a structurally valid project can still score low or fail review when the rules flag authoring issues, such as too few eval datapoints, prompt placeholder problems, or misconfigured guardrails.
No login required. When the project declares an llm_as_judge guardrail and a CLI session is active, review also checks the judge model against the models available to the tenant; without a session, that one rule is skipped and the rest of the review runs.
Synopsis
uip agent review [path] [--checks <categories>]
uip agent review [path] [--checks <categories>]
All invocations honor the global options (--output, --output-filter, --log-level, --log-file). Exit codes follow the standard contract.
Arguments
[path](optional, default.) — Agent project directory.
Options
| Flag | Default | Purpose |
|---|---|---|
--checks <categories> | all categories | Comma-separated rule categories to run. Known categories: evals, schema, tools, guardrails, code, general. |
An empty or all-whitespace --checks value fails with AgentReviewInvalidChecks; an unrecognized category name fails the same way, listing the known set.
Verdict and score
- The score starts at 100 and drops for each error, warning, and info finding. Strengths add a small bonus only when there are no errors.
- The verdict is
FAILwhen the project has any error finding or the score is below 50. Otherwise it isPASS. - A
FAILverdict caps the grade atC+.
Eval coverage rules
The evals category checks that the project has enough eval datapoints across all its eval sets under evals/eval-sets/. Both rules have error severity, so either one fails the review.
| Rule ID | Fires when |
|---|---|
NO_EVALS | The project has no eval datapoints at all: no eval-set files, or every eval set has an empty evaluations array. |
TOO_FEW_EVALS | The project has eval datapoints, but fewer than five across all eval sets. |
When the project has at least one datapoint, an eval set whose evaluations array is empty is reported separately as an info finding (LOWCODE_EVAL_SET_EMPTY).
Examples
# Review a project at a specific path
uip agent review ./my-agent
# Run only the schema and general rule categories
uip agent review --checks schema,general
# Review a project at a specific path
uip agent review ./my-agent
# Run only the schema and general rule categories
uip agent review --checks schema,general
Data shape (--output json)
{
"Code": "AgentReview",
"Data": {
"ProjectDir": "./my-agent",
"Verdict": "PASS",
"Score": 100,
"Grade": "A+",
"Issues": [],
"Strengths": [],
"Stats": {
"errors": 0,
"warnings": 0,
"infos": 0,
"byCategory": { "evals": 0, "schema": 0, "tools": 0, "guardrails": 0, "code": 0, "general": 0 }
}
}
}
{
"Code": "AgentReview",
"Data": {
"ProjectDir": "./my-agent",
"Verdict": "PASS",
"Score": 100,
"Grade": "A+",
"Issues": [],
"Strengths": [],
"Stats": {
"errors": 0,
"warnings": 0,
"infos": 0,
"byCategory": { "evals": 0, "schema": 0, "tools": 0, "guardrails": 0, "code": 0, "general": 0 }
}
}
}
Issues and Strengths are populated with rule-specific entries when the project has findings. Each issue carries ruleId, category, severity (error, warning, or info), and description, plus file, line, element, evidences, and suggestedFix when the rule provides them. Review failing to run at all (bad path, unreadable project) uses Code: "AgentReviewFailed".
Related
uip agent review-history add— record this command's grade over time.uip agent validate— the schema-conformance check to run alongside review.