Read test and coverage evidence
Use Bee to read reports without loading their full contents into an agent’s context. These commands work offline without bee init, MongoDB or a model. Create a new directory and save the following UTF-8 files there. They are synthetic parser examples, not tests you have executed.
1 · Summarize a failing test
Section titled “1 · Summarize a failing test”Save report.trx:
<TestRun xmlns="http://microsoft.com/schemas/VisualStudio/TeamTest/2010" id="11111111-1111-4111-8111-111111111111"> <TestDefinitions><UnitTest id="22222222-2222-4222-8222-222222222222"><TestMethod className="Docs.Synthetic" name="Example"/></UnitTest></TestDefinitions> <Results><UnitTestResult testId="22222222-2222-4222-8222-222222222222" executionId="33333333-3333-4333-8333-333333333333" testName="Synthetic example" outcome="Failed"/></Results> <ResultSummary outcome="Failed"><Counters total="1" executed="1" passed="0" failed="1" notExecuted="0"/></ResultSummary></TestRun>bee test summarize report.trx --limit 5 --budget 8000 --jsonExpect exit 1 because the completed report contains one failed test named Synthetic example: total 1, executed 1, passed 0, failed 1. This is a successful reading of a failure. bee test summarize does not execute tests. With real reports, retain the original runner’s exit code, source revision and full report. Exit 5 means incomplete/no executed tests; inspect the diagnostics before drawing a conclusion. Check the omitted-detail counts when limiting output. The summary reference describes the supported TRX shape.
2 · Read coverage without inventing missing measurements
Section titled “2 · Read coverage without inventing missing measurements”Save coverage.info:
TN:syntheticSF:Price.csDA:5,1DA:6,0LF:2LH:1end_of_recordbee code coverage import coverage.info --format lcov --jsonExpect exit 0, status: complete and files[0].lines: state: known, total: 2, covered: 1, percent: 50. Branches are unknown; no branch records were supplied. Price.cs is a report label: the importer does not open that file. checkedSource: not-read and sourceBinding: not-established remain explicit. Exit 0 proves the report parsed, not that today’s code has 50% coverage. Fix malformed/unsupported input after exit 5; do not silently replace it with zero coverage. For format details use bee code coverage import --help or the command reference.
3 · Inspect what a QA assessment still needs
Section titled “3 · Inspect what a QA assessment still needs”Prerequisites: a Git checkout with at least one commit, Git on PATH, and its physical root directory. Run from that root. The bundled bee-cli profile targets Bee CLI integration on linux-x64, not arbitrary applications; this is a diagnostic demonstration, not a project-specific QA policy. Bash/zsh:
bee qa assess --root "$(pwd -P)" --base HEAD --head worktree --profile bee-cli --limit 5 --budget 16000 --jsonPowerShell:
bee qa assess --root (Get-Location).Path --base HEAD --head worktree --profile bee-cli --limit 5 --budget 16000 --jsonExpect exit 5, decision: collect_evidence, reasonCounts.missing_evidence: 1, mode: diagnostics, trusted: false and acceptance: false. Every assessment exits 5 by design. It has not run tests or granted QA approval. --base HEAD compares with the current commit; choose your actual comparison commit for real work.
Continue with the complete QA bundle example: preview source/policy, attach a hashed report outside the checkout, assess it, then inspect named failures and missing scope. Do not relabel an old report with a new source hash. If the profile does not fit your project, use the test runner and review policy your project actually defines.
4 · Hand off enough to verify
Section titled “4 · Hand off enough to verify”Keep the command, true process exit, source revision, artifact hash, omissions and original report together. A parser summary, analysis result, test outcome and review decision answer different questions. Increase a bounded output only when the named omission matters. Then report your plan step with a summary of what was actually checked.