Walk-and-Fix (preview)
Walk-and-Fix turns autonomous-walker findings into owner-scoped FixIntent records so teams triage, assign, close defects through stable API.
Planning surface only for current deployments: automatic patch generation not implemented.
run_fix_cycleproposes plan but does not execute. Only implemented apply adapter edits files inside legacy local VM; conflicts with current immutable app deployment model and must not be used as production repair path. GitHub PR and provider-redeploy modes not implemented.
Supported workflow
- Run autonomous walk.
- Review finding and evidence.
- Call
propose_fixcreate traceable intent, orrun_fix_cyclecreate intents for bounded set blocker/major findings. - Change source repository through normal review process.
- Build and deploy new immutable variant.
- Re-run walk and compare with prior walk.
- Call
close_findingonly after new evidence demonstrates finding fixed, duplicated, not bug, intentionally not fixed.
Keeps source control, review history, deployed artifact aligned.
Tools and boundaries
| Tool | Current behavior |
|---|---|
propose_fix | Creates FixIntent; does not generate code |
run_fix_cycle | Creates bounded plan; does not apply changes |
get_fix_intent / list_fix_intents | Reads owned intent records |
close_finding | Records disposition for owned finding |
apply_fix | Legacy local-VM full-file replacement; not immutable-deploy workflow |
apply_fix accepts full file contents, not unified diff. Limited to disposable legacy tester deployments where operator explicitly accepts change outside source control and lost on redeploy.
Evidence required to close finding
Successful build or deployment not enough. Capture:
- source revision containing change;
- replacement deployment ID and status;
- new passing scenario or walk exercising affected path; and
- diff showing finding resolved without new blocker or major regression.
If test cannot reproduce original path leave finding open or record evidence limitation in disposition.