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