autonomous-manual-test
Generate and execute a manual test plan against Taskrail task acceptance criteria in a temporary sandbox
autonomous-manual-test
Generate and execute a manual test plan against task acceptance criteria, then produce a structured report.
Required Flow
- Run
go run ./cmd/taskrail validate. - Read the target task file and extract its acceptance criteria.
- Create a manual test plan at
planning/artifacts/manual-test/<task-id>/<timestamp>/plan.md. - Derive numbered test steps from the acceptance criteria.
- Execute each test step in order.
- If a step fails, decide whether a code fix is needed, apply the smallest fix, and re-run only the affected step.
- Write
planning/artifacts/manual-test/<task-id>/<timestamp>/report.mdwith per-step results and a final verdict. - Clean up any temporary manual test code after the report is written.
Test Modes
Sandbox Mode (default)
Use sandbox mode for most Taskrail work.
- Run commands in a temporary directory or temporary repository.
- Exercise normal
go run ./cmd/taskrail ...flows. - Use small temporary helper programs only when direct shell commands are not enough.
- Delete any
cmd/manual-test-*/helper directories after writing the report.
Optional manual_test Go-tag mode
Use this only when shell-driven validation is awkward for a real CLI flow.
- Write
_manual_test.gofiles with build tag//go:build manual_test. - Name tests
TestManual_<Name>. - Run them with
go test -tags=manual_test ./<package> -run TestManual_<Name> -v -count=1. - Delete the
_manual_test.gofiles after writing the report.
Artifact Format
plan.md
- task id
- generated timestamp
- test steps derived from acceptance criteria
- commands to run
- expected outcomes
report.md
- task id
- executed timestamp
- verdict: pass, pass-with-fixes, or fail
- per-step observations
- any fixes that were applied
Rules
- Manual testing here is for Taskrail's internal development workflow, not a required Taskrail product invariant.
- Never substitute ordinary automated tests for manual test evidence when the change needs end-to-end judgment.
- Re-run only the affected test step after a fix.
- Do not auto-select another task after manual testing completes.
- Never commit temporary manual test code.