Run review without checking out PR

Author: trask

.github/scripts/pr-triage/common.py

@@ -39,6 +39,7 @@ class Summary:
     restored_branch: str | None = None
     restoration_note: str | None = None
     outcome: str = ""
+    changed_files_label: str = "Files changed"
     changed_files: list[str] = field(default_factory=list)
     commits: list[str] = field(default_factory=list)
     push_result: str | None = None
@@ -63,7 +64,7 @@ def print_text(self) -> None:
             for failure in self.failures:
                 print(f"- {failure}")
         if self.changed_files:
-            print("Files changed:")
+            print(f"{self.changed_files_label}:")
             for path in self.changed_files:
                 print(f"- {path}")
         if self.commits:
@@ -177,7 +178,16 @@ def restore_original_branch(summary: Summary) -> None:
     try:
         branch = current_branch()
     except WorkflowError as e:
-        summary.restoration_note = str(e)
+        if merge_in_progress():
+            summary.restoration_note = "not restored because a merge is in progress"
+            return
+        if status_porcelain().strip():
+            summary.restoration_note = "not restored because the working tree has uncommitted changes"
+            return
+        progress(f"Restoring original branch from detached HEAD: {summary.original_branch}")
+        git(["checkout", summary.original_branch], summary)
+        summary.restored_branch = summary.original_branch
+        summary.restoration_note = "restored from detached HEAD"
         return
     if branch == summary.original_branch:
         summary.restored_branch = branch
@@ -240,14 +250,22 @@ def checkout_pr_no_push_check(pr: int, summary: Summary) -> None:
     remember_pr_url(pr_view(pr, summary), summary)
 
 
-def run_pr_workflow(pr: int, body: Callable[[Summary], int], *, push_required: bool = True) -> int:
+def run_pr_workflow(
+    pr: int,
+    body: Callable[[Summary], int],
+    *,
+    push_required: bool = True,
+    checkout_required: bool = True,
+) -> int:
     summary = Summary(pr=pr)
     try:
         require_clean_worktree(summary)
         summary.original_branch = current_branch(summary)
+        if push_required and not checkout_required:
+            raise WorkflowError("push workflows require checking out the PR branch")
         if push_required:
             checkout_pr(pr, summary)
-        else:
+        elif checkout_required:
             checkout_pr_no_push_check(pr, summary)
         return body(summary)
     except Exception as e:
@@ -328,8 +346,22 @@ def extract_job_id(check: dict[str, Any]) -> int | None:
     return int(value) if isinstance(value, int) else None
 
 
-def invoke_copilot(prompt: str, summary: Summary) -> str:
-    cmd = ["copilot", "-p", prompt, "--allow-all-tools", "--model", COPILOT_MODEL]
+def invoke_copilot(
+    prompt: str,
+    summary: Summary,
+    event_log_path: Path | None = None,
+    tool_usage_path: Path | None = None,
+    *,
+    allow_all_tools: bool = True,
+    extra_args: list[str] | None = None,
+) -> str:
+    cmd = ["copilot", "-p", prompt, "--model", COPILOT_MODEL]
+    if allow_all_tools:
+        cmd.append("--allow-all-tools")
+    if extra_args is not None:
+        cmd.extend(extra_args)
+    if event_log_path is not None:
+        cmd.extend(["--output-format", "json"])
     progress(f"Handing off to Copilot CLI using {COPILOT_MODEL}; streaming output below")
     proc = subprocess.Popen(
         cmd,
@@ -342,19 +374,71 @@ def invoke_copilot(prompt: str, summary: Summary) -> str:
         bufsize=1,
     )
     output_parts: list[str] = []
+    message_parts: list[str] = []
+    tool_uses: list[dict[str, Any]] = []
+    event_log = event_log_path.open("w", encoding="utf-8") if event_log_path is not None else None
     if proc.stdout is not None:
-        for line in proc.stdout:
-            print(line, end="", flush=True)
-            output_parts.append(line)
+        try:
+            for line in proc.stdout:
+                if event_log is not None:
+                    event_log.write(line)
+                    event_log.flush()
+                    output_parts.append(line)
+                    try:
+                        event = json.loads(line)
+                    except json.JSONDecodeError:
+                        continue
+                    event_type = event.get("type")
+                    data = event.get("data") or {}
+                    if event_type == "assistant.message":
+                        content = data.get("content")
+                        if isinstance(content, str) and content:
+                            print(content, flush=True)
+                            message_parts.append(content)
+                    elif event_type == "tool.execution_start":
+                        tool_name = data.get("toolName")
+                        if isinstance(tool_name, str):
+                            tool_uses.append({"tool": tool_name, "arguments": data.get("arguments")})
+                    continue
+                print(line, end="", flush=True)
+                output_parts.append(line)
+        finally:
+            if event_log is not None:
+                event_log.close()
     returncode = proc.wait()
-    output = "".join(output_parts)
+    output = "\n".join(message_parts) if event_log_path is not None else "".join(output_parts)
     if returncode != 0:
         raise subprocess.CalledProcessError(
             returncode,
-            ["copilot", "-p", "<generated prompt>", "--allow-all-tools", "--model", COPILOT_MODEL],
-            output,
+            ["copilot", "-p", "<generated prompt>", "--model", COPILOT_MODEL],
+            "".join(output_parts),
             "",
         )
+    if tool_usage_path is not None:
+        counts: dict[str, int] = {}
+        for tool_use in tool_uses:
+            tool = tool_use["tool"]
+            counts[tool] = counts.get(tool, 0) + 1
+        tool_usage_path.write_text(
+            json.dumps(
+                {
+                    "tools": [{"tool": tool, "count": count} for tool, count in sorted(counts.items())],
+                    "tool_uses": tool_uses,
+                },
+                indent=2,
+                sort_keys=True,
+            )
+            + "\n",
+            encoding="utf-8",
+        )
+        summary.notes.append(f"Copilot tool usage summary: {display_path(str(tool_usage_path))}")
+        if counts:
+            tools = ", ".join(f"{tool} ({count})" for tool, count in sorted(counts.items()))
+            summary.notes.append(f"Copilot tools used: {tools}")
+        else:
+            summary.notes.append("Copilot tools used: none")
+    if event_log_path is not None:
+        summary.notes.append(f"Copilot event log: {display_path(str(event_log_path))}")
     progress("Copilot CLI handoff completed")
     return output.strip()
 

.github/scripts/pr-triage/fix_ci.py

@@ -37,6 +37,11 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument("pr", type=int, help="pull request number")
     parser.add_argument("--no-push", action="store_true", help="commit locally but do not push to the PR")
     parser.add_argument("--keep-temp", action="store_true", help="reuse and retain the temp bundle directory")
+    parser.add_argument(
+        "--capture-tool-usage",
+        action="store_true",
+        help="capture Copilot JSONL events and tool usage summary in the work bundle",
+    )
     return parser.parse_args()
 
 
@@ -182,7 +187,14 @@ def body(summary: Summary) -> int:
 
         commit_message_path = bundle_dir / "commit-message.txt"
         prompt_improvement_path = bundle_dir / "prompt-improvement.md"
-        response = invoke_copilot(copilot_prompt(args.pr, bundle_checks, commit_message_path, prompt_improvement_path), summary)
+        event_log_path = bundle_dir / "copilot-events.jsonl" if args.capture_tool_usage else None
+        tool_usage_path = bundle_dir / "copilot-tools-used.json" if args.capture_tool_usage else None
+        response = invoke_copilot(
+            copilot_prompt(args.pr, bundle_checks, commit_message_path, prompt_improvement_path),
+            summary,
+            event_log_path=event_log_path,
+            tool_usage_path=tool_usage_path,
+        )
         (bundle_dir / "copilot-response.txt").write_text(response + "\n", encoding="utf-8")
         read_prompt_improvement(prompt_improvement_path, summary)
 

.github/scripts/pr-triage/resolve_conflicts.py

@@ -27,6 +27,11 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument("--upstream", default="upstream", help="upstream remote name")
     parser.add_argument("--no-push", action="store_true", help="commit locally but do not push to the PR")
     parser.add_argument("--keep-temp", action="store_true", help="reuse and retain the temp bundle directory")
+    parser.add_argument(
+        "--capture-tool-usage",
+        action="store_true",
+        help="capture Copilot JSONL events and tool usage summary in the work bundle",
+    )
     return parser.parse_args()
 
 
@@ -100,7 +105,14 @@ def workflow(summary: Summary) -> int:
         bundle_dir = make_temp_dir("otel-conflicts", args.pr, args.keep_temp)
         plan_path = write_conflict_bundle(bundle_dir, summary)
 
-        response = invoke_copilot(copilot_prompt(plan_path), summary)
+        event_log_path = bundle_dir / "copilot-events.jsonl" if args.capture_tool_usage else None
+        tool_usage_path = bundle_dir / "copilot-tools-used.json" if args.capture_tool_usage else None
+        response = invoke_copilot(
+            copilot_prompt(plan_path),
+            summary,
+            event_log_path=event_log_path,
+            tool_usage_path=tool_usage_path,
+        )
         (bundle_dir / "copilot-response.txt").write_text(response + "\n", encoding="utf-8")
 
         remaining = unmerged_paths(summary)