DO NOT MERGE — responder test sandbox

.github/workflows/review-responder.md

@@ -1,16 +1,21 @@
 ---
-if: "contains(github.event.pull_request.labels.*.name, 'aw')"
 on:
-  pull_request_review:
-    types: [submitted]
-  roles: all
-  bots: [Copilot, copilot-pull-request-reviewer]
+  workflow_dispatch:
+    inputs:
+      pr_number:
+        description: "PR number to address review comments on"
+        required: true
+        type: string
 
 permissions:
   contents: read
   issues: read
   pull-requests: read
 
+checkout:
+  fetch: ["*"]
+  fetch-depth: 0
+
 engine:
   id: copilot
   model: claude-opus-4.6
@@ -28,8 +33,11 @@ safe-outputs:
   noop:
     report-as-issue: false
   push-to-pull-request-branch:
+    target: "*"
+    labels: [aw]
     github-token: ${{ secrets.GH_AW_WRITE_TOKEN }}
   reply-to-pull-request-review-comment:
+    target: "*"
     max: 10
     github-token: ${{ secrets.GH_AW_WRITE_TOKEN }}
   add-labels:
@@ -39,32 +47,28 @@ safe-outputs:
 
 # Review Responder
 
-Address review comments on pull request #${{ github.event.pull_request.number }}.
+Address review comments on pull request #${{ inputs.pr_number }}.
 
 ## Instructions
 
-This workflow runs when a review is submitted on a pull request.
-
-1. First, check if the PR has the `aw` label. If it does NOT have the `aw` label, stop immediately — this workflow only handles agent-created PRs.
-
-2. Check the review that triggered this workflow. If the review has no comments (e.g., a plain approval with no inline comments), stop — there is nothing to address.
+This workflow addresses unresolved review comments on a pull request.
 
-3. Check if the PR already has the label `review-response-attempted`. If it does, add a comment to the PR saying "Review response already attempted — stopping to prevent loops. Manual intervention needed." and stop.
+1. Check if the PR already has the label `review-response-attempted`. If it does, add a comment to the PR saying "Review response already attempted — stopping to prevent loops. Manual intervention needed." and stop.
 
-4. Add the label `review-response-attempted` to the PR.
+2. Add the label `review-response-attempted` to the PR.
 
-5. Read the unresolved review comment threads on the PR (not just the latest review — get all unresolved threads). If there are more than 10 unresolved threads, address the first 10 and leave a summary comment on the PR noting how many remain for manual follow-up.
+3. Read the unresolved review comment threads on the PR (not just the latest review — get all unresolved threads). If there are more than 10 unresolved threads, address the first 10 and leave a summary comment on the PR noting how many remain for manual follow-up.
 
-6. For each unresolved review comment thread (up to 10):
+4. For each unresolved review comment thread (up to 10):
    a. Read the comment and understand what change is being requested
    b. Read the relevant file and surrounding code context
-   c. Make the requested fix in the code (edit the file locally — do NOT push yet)
+   c. Make the requested fix in the code
    d. Reply to the comment thread explaining what you changed
 
-7. After addressing all comments, run the CI checks locally to make sure your fixes don't break anything: `uv sync && uv run ruff check --fix . && uv run ruff format . && uv run pyright && uv run pytest --cov --cov-fail-under=80 -v`
+5. After addressing all comments, run the CI checks locally to make sure your fixes don't break anything: `uv sync && uv run ruff check --fix . && uv run ruff format . && uv run pyright && uv run pytest --cov --cov-fail-under=80 -v`
 
-8. Push all changes in a single commit with message "fix: address review comments". Reply to all threads BEFORE pushing — replies after a push will appear on outdated code.
+6. If CI checks fail, fix the issues and re-run until they pass. Do not push broken code.
 
-If a review comment requests a change that would be architecturally significant or you're unsure about, reply to the thread explaining your concern rather than making the change blindly.
+7. Push all changes in a single commit with message "fix: address review comments".
 
-NOTE: Thread resolution is handled separately by the pipeline orchestrator after this workflow completes. Your job is to fix the code and reply to threads — do NOT attempt to resolve threads.
+If a review comment requests a change that would be architecturally significant or you're unsure about, reply to the thread explaining your concern rather than making the change blindly.

tests/test_responder_sandbox.py

@@ -0,0 +1,16 @@
+# Responder Sandbox — intentionally imperfect code for testing
+
+
+def calculate_total(lst: list[int]) -> int:
+    total = 0
+    for i in lst:
+        total = total + i
+    return total
+
+
+def format_output(data: list[str], verbose: bool) -> str:
+    result = ""
+    for item in data:
+        suffix = "\n" if verbose else ", "
+        result = result + str(item) + suffix
+    return result