A tool should lead to a verified change
An external tool is useful not because it is connected, but because it helps move from signal to an evidence-backed change. The same frame applies to trackers, monitoring, databases, design, documentation, and browser checks.
The tool adds context; it does not replace a task spec, code discovery, tests, or human decisions.
| Scenario | Source | Main verification |
|---|---|---|
| Issue-to-PR | Issue, comments, screenshots | Tests, PR review, TASK_SPEC.md |
| Monitoring triage | Alert, logs, trace | Reproduce, regression test, EVIDENCE_LOG.md |
| Read-only DB | Rows, aggregates, anomalies | Code path, integration test, query review |
| Design-to-code | Mockup, screenshot, design comment | Browser check, screenshot, TASK_SPEC.md |
| Docs lookup | Official documentation | Version, build, tests, migration note |
| Browser check | Actual UI behavior | Repeatable scenario and screenshot |
external signal
→ hypothesis
→ bounded plan
→ small change
→ verification
→ repository artifactChoose a scenario
- Team request or complaint — issue-to-PR.
- System signal — monitoring triage.
- Data anomaly — read-only DB.
- Visual reference — design-to-code.
- Version or API fact — docs lookup.
- Actual UI behavior — browser check.
Issue-to-PR and monitoring triage
- Read the issue, comments, or alert.
- State symptom and current hypothesis.
- Find affected area in the codebase.
- Record scope, non-goals, and reproduction.
- Make the smallest change.
- Confirm it with a test and review.
- Save the conclusion in TASK_SPEC.md or EVIDENCE_LOG.md.
Data, design, and documentation
Read-only database access can show what happened but not why. Find the cause in code paths and tests rather than treating data manually with UPDATE or DELETE.
A mockup is a visual reference, not permission for broad UI refactoring. Documentation clarifies an API for a specific version and must still be compared with the project structure.
Canonical lesson source · JavaRush
Open JavaRush material