Allure Agent Mode turns your existing test command into agent-readable test reporting for Codex, Claude, and other coding agents, so they can review what ran before claiming a change is safe.
Keep your test reporting workflow. Let your AI agent use it too.
I checked checkout coverage with two fresh Allure agent runs:
Covered happy paths:
Main gaps I see:
You ask an agent to check a change. It runs a command, sees green, and says it looks safe. But you still have to ask the real questions: did it run the right tests, did it notice what was skipped, and did it understand what those tests actually proved?
A passing command can hide a lot. The agent may have run too broad a smoke test, missed the feature you cared about, ignored skipped tests, or treated a weak assertion as real validation.
$ npm test -- --grep smoke✓ test/checkout.smoke.test.ts (12) ✓ cart summary ✓ order successTest Files 1 passed (1)Tests 7 passed, 5 skipped (12)Duration 1.4sThe useful story often lives in the report: which tests ran, what steps happened, what retried, what was attached, which screenshots or traces matter, and where the run looks thin.
$ npm test checkout.spec.ts✓ test/checkout.spec.ts (2) ✓ creates an order ✓ declines invalid cardTest Files 1 passed (1)Tests 2 passed (2)Duration 349msWhen an agent says covered or safe, you need more than confidence. You need to see what it checked, what reporting it reviewed, and what gaps it found before you accept the answer.
$ npm test checkout.spec.ts× test/checkout.spec.ts (2) ✓ creates an order × keeps total after retryExpected: "$42.00" Received: "$39.00"Test Files 1 failed (1)Tests 1 failed, 1 passed (2)Duration 612msIt lets your AI agent use the same test reporting workflow your team already uses, so its answer can be based on what actually ran, not just what the terminal happened to print.
index.mdI checked checkout coverage with three fresh Allure agent runs.
checkout.smoke.test.ts passed, but 5 skipped scenarios were present in the same run.checkout.spec.ts covered the legacy checkout route, not the new checkout page.I would not call checkout fully covered yet. The useful next test is the new-page retry path with fresh cart state.
Allure Agent Mode wraps the test command your agent would run anyway and writes agent-readable output, including a Markdown run report. The Allure skills teach Codex, Claude, and other coding agents how to set it up, when to run it, and how to review the result before answering.
Add or change tests without losing the behavior you meant to protect. The agent runs the intended scope, checks what actually executed, and strengthens weak assertions or evidence before calling the work done.
Review what is covered at runtime, not just what appears covered from source code. The agent compares intended scope with observed tests, skipped cases, attachments, and gaps.
Before saying a change is safe, the agent runs the smallest meaningful test scope and reports what passed, what was skipped or retried, and what remains uncertain.
When tests fail, the agent reviews per-test evidence, findings, logs, screenshots, traces, retries, and stderr before deciding whether it is a product bug, test bug, fixture issue, or environment problem.
Separate real failures from unstable tests. The agent can inspect retries, timing, stale state, attachments, and repeated runs before recommending a fix.
Improve tests that pass but are hard to review. The agent adds meaningful steps, attachments, labels, parameters, or descriptions, then reruns the same scope to confirm the report is clearer.
Start by adding the Allure skills. If your project already has Allure reporting, you can go straight to the agent workflow setup. If not, ask your agent to configure reporting first. The examples use different coding agents to show the workflow is portable; use whichever CLI you prefer.
Install the Allure skills in the repository where your tests live.
$ npx skills add allure-framework/skills
Optional. Use this when the project does not already emit useful Allure results.
$ claude "configure allure reporting"
Let the agent discover local wrappers, test commands, result paths, and supported agent options.
$ codex "enable allure agent mode"
Now use normal testing requests. Agent Mode gives the agent a report to review before it answers.
$ opencode run "review checkout test coverage"
Allure Agent Mode does not require MCP, a hosted service, or a specific AI vendor. It works through the test command your coding agent can already run, then gives that agent a readable report to review before it answers.
Readable markdown run reports
Allure reportsVisual reports, charts, history, and artifacts
Keep the judgment with your team. Let Agent Mode give your coding agent the report context it needs before it claims coverage, safety, or a fix.