What happened
On PR #5, the code agent created a PR in openshift-helm/console-fullsend. However, this repo is a staging/fork repo — the actual upstream is openshift/console. User sowmya-sl commented: the desired behavior is for the code agent to push changes to a branch without creating a PR, so the user can submit a cross-repo PR to the upstream manually.
The code agent's post-script always creates a PR in the source repo. There is no config option to skip PR creation. This forces repos that serve as staging areas into a workflow mismatch: the agent creates PRs that are never intended to be merged in-repo, triggering unnecessary review agent runs and creating noise.
What could go better
The code agent post-script should support a configuration option to push changes to a branch without creating a PR. This is useful for repos that serve as staging areas or forks where the actual PR target is a different upstream repo.
Confidence: High — the user explicitly requested this behavior and explained the rationale. The pattern of fork/staging repos where PRs go upstream is common in the OpenShift ecosystem (which typically contributes to repos like openshift/console from org-specific forks).
Uncertainty: The specific implementation location (post-code script vs. harness config vs. fullsend config) depends on how the platform structures output modes.
Proposed change
Add a code.output_mode or code.skip_pr option to .fullsend/config.yaml that controls what the code agent's post-script does after committing changes. Options:
pr (default): Current behavior — push branch and create a PR.branch: Push the branch only, no PR creation. Post a comment on the issue with the branch name and instructions for the user to create a cross-repo PR.
Example config:
code:
output_mode: branch # push branch only, no PR
The post-code script (in the fullsend platform) would check this config value and skip gh pr create when set to branch. The review agent dispatch (triggered by PR creation / ready-for-review label) would naturally not fire since no PR exists.
Validation criteria
After implementation: (1) When code.output_mode: branch is set in .fullsend/config.yaml, the code agent pushes changes to a branch but does not create a PR. (2) The agent posts a comment on the originating issue with the branch name and a note that no PR was created. (3) No review agent dispatch is triggered. Test on openshift-helm/console-fullsend with the next /fs-code invocation.
Generated by retro agent from openshift-helm/console-fullsend#5
What happened
On PR #5, the code agent created a PR in
openshift-helm/console-fullsend. However, this repo is a staging/fork repo — the actual upstream isopenshift/console. Usersowmya-slcommented: the desired behavior is for the code agent to push changes to a branch without creating a PR, so the user can submit a cross-repo PR to the upstream manually.The code agent's post-script always creates a PR in the source repo. There is no config option to skip PR creation. This forces repos that serve as staging areas into a workflow mismatch: the agent creates PRs that are never intended to be merged in-repo, triggering unnecessary review agent runs and creating noise.
What could go better
The code agent post-script should support a configuration option to push changes to a branch without creating a PR. This is useful for repos that serve as staging areas or forks where the actual PR target is a different upstream repo.
Confidence: High — the user explicitly requested this behavior and explained the rationale. The pattern of fork/staging repos where PRs go upstream is common in the OpenShift ecosystem (which typically contributes to repos like openshift/console from org-specific forks).
Uncertainty: The specific implementation location (post-code script vs. harness config vs. fullsend config) depends on how the platform structures output modes.
Proposed change
Add a
code.output_modeorcode.skip_proption to.fullsend/config.yamlthat controls what the code agent's post-script does after committing changes. Options:pr(default): Current behavior — push branch and create a PR.branch: Push the branch only, no PR creation. Post a comment on the issue with the branch name and instructions for the user to create a cross-repo PR.Example config:
The post-code script (in the fullsend platform) would check this config value and skip
gh pr createwhen set tobranch. The review agent dispatch (triggered by PR creation /ready-for-reviewlabel) would naturally not fire since no PR exists.Validation criteria
After implementation: (1) When
code.output_mode: branchis set in.fullsend/config.yaml, the code agent pushes changes to a branch but does not create a PR. (2) The agent posts a comment on the originating issue with the branch name and a note that no PR was created. (3) No review agent dispatch is triggered. Test onopenshift-helm/console-fullsendwith the next/fs-codeinvocation.Generated by retro agent from openshift-helm/console-fullsend#5