When to use what
A quick map from what you're trying to do to the command that does it.
Understanding code
| You want to… | Use |
|---|---|
| Orient in an unfamiliar repo | arch, then modules |
| Get context for a specific task | context "task" or ask "task" |
| Find code when you don't know its name | semantic "meaning" |
| See one file's shape without bodies | skeleton FILE |
| Read an exact line range | lines FILE 40 80 |
Before editing
| You want to… | Use |
|---|---|
| Know what breaks if I change this function | method-impact QNAME |
| See a symbol's full blast radius | impact QNAME |
| Find the tests that cover this | test-map TARGET |
| Check a plan references real code | verify-plan |
Generating code
| You want to… | Use |
|---|---|
| Match the repo's file conventions | conventions, then scaffold |
| Run the whole gated pipeline | create "task" |
| Audit AI output for fabrications | verify / verify-output |
| Review a diff before committing | review |
Trusting an answer
| You want to… | Use |
|---|---|
| Confirm context is sufficient first | validate "task" --min-coverage 60 |
| Score if the answer is grounded | judge --answer-file … --context-file … |
| Catch fabricated files / symbols | verify --answer-file … |
Proving value
| You want to… | Use |
|---|---|
| See tokens + dollars saved | gain |
| Open a savings dashboard | gain --report |
| Produce an audit-grade evidence pack | evidence "task" |
Next steps
- Full flag list: CLI reference
- The recommended everyday loop: Quick start