agent: implement issue #1 - Doc: request CN version README

src/agent_gitea/rendering.py

@@ -34,8 +34,7 @@ Issue URL: {issue.html_url}
 Implement the requested change in this workspace. Keep the change scoped to this issue.
 Run the relevant tests before finishing.
 
-Write `.agent-output/AGENT_IMPLEMENTATION_REPORT.md` using this exact section contract.
+Write `AGENT_IMPLEMENTATION_REPORT.md` in the workspace root using this exact section contract:
-Keep the section headings exactly as written below, but write the section content in Chinese:
 
 - Summary
 - Files changed
@@ -56,9 +55,7 @@ Issue: #{issue.issue_number} {issue.title}
 Review the implementation currently checked out in this workspace. Focus on correctness,
 scope control, test evidence, and human risks. Do not modify code.
 
-Write `.agent-output/AGENT_REVIEW_REPORT.md` using this exact section contract.
+Write `AGENT_REVIEW_REPORT.md` in the workspace root using this exact section contract:
-Keep the section headings exactly as written below. Keep the Verdict token in English,
-but write the section content and Suggested PR Comment in Chinese:
 
 - Verdict: APPROVE, REQUEST_CHANGES, or NEEDS_HUMAN_DECISION
 - Summary
@@ -72,16 +69,16 @@ but write the section content and Suggested PR Comment in Chinese:
 
 
 def render_pr_body(issue: IssueRecord, implementation_report: str) -> str:
-    return f"""关联 Issue：#{issue.issue_number}
+    return f"""Closes #{issue.issue_number}
 
-## 代理实现报告
+## Agent Implementation Report
 
 {implementation_report.strip()}
 
-## 人工审核
+## Human Review Gate
 
-此 PR 由本地 agent-manager 自动创建，但不会自动合并。
+This PR was opened by the local agent manager. It has not been auto-merged.
-请维护者人工审核、决策并手动合并。
+Human maintainers must review, decide, and merge manually.
 """
 
 
@@ -105,13 +102,13 @@ def extract_section(raw: str, title: str) -> str:
 
 
 def render_human_review_summary(review: ReviewReport) -> str:
-    return f"""## 代理评审摘要
+    return f"""## Agent Review Summary
 
-结论：`{review.verdict}`
+Verdict: `{review.verdict}`
 
 {review.suggested_pr_comment.strip()}
 
-## 需要人工处理
+## Human Action Required
 
-请人工审核该 PR。agent-manager 不会自动合并、关闭 PR 或提交变更请求。
+Please review the PR manually. The agent manager will not merge, close, or request changes automatically.
 """

src/agent_gitea/service.py

@@ -132,15 +132,12 @@ class TaskRunner:
                     error_message="implementer produced no diff",
                     clear_lease=True,
                 )
-            commit_message = f"agent: implement issue #{issue.issue_number} - {issue.title}"
-            commit_id = self.workspace_manager.commit_changes(workspace, commit_message)
-            self.db.add_event(task.id, TaskState.TESTING, TaskState.TESTING, f"committed implementation {commit_id}")
             self.workspace_manager.push_branch(workspace, branch_name)
             pr_body = render_pr_body(issue, implementation_report)
             pr = self.gitea.create_pull_request(
                 owner=repo.owner,
                 name=repo.name,
-                title=f"代理实现：{issue.title}",
+                title=f"Agent: {issue.title}",
                 body=pr_body,
                 head=branch_name,
                 base=repo.default_branch,
@@ -194,9 +191,7 @@ class TaskRunner:
         workspace: Path,
     ) -> str:
         prompt = render_implementer_prompt(repo, issue, branch_name)
-        output_dir = workspace / ".agent-output"
+        prompt_path = workspace / "AGENT_IMPLEMENTER_PROMPT.md"
-        output_dir.mkdir(exist_ok=True)
-        prompt_path = output_dir / "AGENT_IMPLEMENTER_PROMPT.md"
         write_prompt(prompt_path, prompt)
         command = render_command(
             self.config.agents.implementer.command,
@@ -207,7 +202,7 @@ class TaskRunner:
             branch_name=branch_name,
         )
         result = self.command_runner.run(command, workspace, stdin=prompt)
-        report = read_report(output_dir / "AGENT_IMPLEMENTATION_REPORT.md")
+        report = read_report(workspace / "AGENT_IMPLEMENTATION_REPORT.md")
         self.db.add_agent_run(
             task_id=task.id,
             role="implementer",
@@ -231,9 +226,7 @@ class TaskRunner:
         workspace: Path,
     ) -> str:
         prompt = render_reviewer_prompt(repo, issue, pr_number)
-        output_dir = workspace / ".agent-output"
+        prompt_path = workspace / "AGENT_REVIEWER_PROMPT.md"
-        output_dir.mkdir(exist_ok=True)
-        prompt_path = output_dir / "AGENT_REVIEWER_PROMPT.md"
         write_prompt(prompt_path, prompt)
         command = render_command(
             self.config.agents.reviewer.command,
@@ -244,7 +237,7 @@ class TaskRunner:
             pr_number=pr_number,
         )
         result = self.command_runner.run(command, workspace, stdin=prompt)
-        report = read_report(output_dir / "AGENT_REVIEW_REPORT.md")
+        report = read_report(workspace / "AGENT_REVIEW_REPORT.md")
         self.db.add_agent_run(
             task_id=task.id,
             role="reviewer",

src/agent_gitea/workspace.py

@@ -32,22 +32,8 @@ class WorkspaceManager:
         self._git(["clone", repo.clone_url, str(path)], Path.cwd())
         self._git(["checkout", repo.default_branch], path)
         self._git(["checkout", "-B", branch_name], path)
-        self.exclude_runtime_artifacts(path)
         return path
 
-    def exclude_runtime_artifacts(self, workspace: str | Path) -> None:
-        exclude_path = Path(workspace) / ".git" / "info" / "exclude"
-        entries = {
-            ".agent-output/",
-            "AGENT_IMPLEMENTER_PROMPT.md",
-            "AGENT_REVIEWER_PROMPT.md",
-        }
-        existing = exclude_path.read_text(encoding="utf-8") if exclude_path.exists() else ""
-        with exclude_path.open("a", encoding="utf-8") as handle:
-            for entry in sorted(entries):
-                if entry not in existing:
-                    handle.write(f"\n{entry}\n")
-
     def has_diff(self, workspace: str | Path, base_ref: str = "origin/HEAD") -> bool:
         result = self._git(["status", "--porcelain"], Path(workspace), check=False)
         if result.stdout.strip():
@@ -55,16 +41,6 @@ class WorkspaceManager:
         diff = self._git(["diff", "--quiet", f"{base_ref}...HEAD"], Path(workspace), check=False)
         return diff.returncode == 1
 
-    def commit_changes(self, workspace: str | Path, message: str) -> str:
-        workspace_path = Path(workspace)
-        self._git(["add", "-A"], workspace_path)
-        result = self._git(["diff", "--cached", "--quiet"], workspace_path, check=False)
-        if result.returncode == 0:
-            raise RuntimeError("no staged implementation changes to commit")
-        self._git(["commit", "-m", message], workspace_path)
-        commit = self._git(["rev-parse", "--short", "HEAD"], workspace_path)
-        return commit.stdout.strip()
-
     def push_branch(self, workspace: str | Path, branch_name: str) -> None:
         self._git(["push", "-u", "origin", branch_name], Path(workspace))
 

tests/test_gitea_service.py

@@ -136,10 +136,6 @@ class FakeWorkspaceManager:
     def push_branch(self, workspace, branch_name):
         self.pushed.append(branch_name)
 
-    def commit_changes(self, workspace, message):
-        self.commit_message = message
-        return "abc1234"
-
     def cleanup(self, workspace):
         pass
 
@@ -154,16 +150,12 @@ class FakeRunner:
         if role == self.fail_role:
             return AgentResult(exit_code=1, stdout="", stderr="failed")
         if role == "implementer":
-            output_dir = Path(cwd) / ".agent-output"
+            (Path(cwd) / "AGENT_IMPLEMENTATION_REPORT.md").write_text(
-            output_dir.mkdir(exist_ok=True)
-            (output_dir / "AGENT_IMPLEMENTATION_REPORT.md").write_text(
                 "## Summary\nImplemented\n\n## Test commands run\npytest\n",
                 encoding="utf-8",
             )
         if role == "reviewer":
-            output_dir = Path(cwd) / ".agent-output"
+            (Path(cwd) / "AGENT_REVIEW_REPORT.md").write_text(
-            output_dir.mkdir(exist_ok=True)
-            (output_dir / "AGENT_REVIEW_REPORT.md").write_text(
                 "## Verdict\nAPPROVE\n\n## Suggested PR Comment\nLooks good.\n",
                 encoding="utf-8",
             )
@@ -204,12 +196,11 @@ def test_run_task_success_posts_review_comments(db, tmp_path):
             return httpx.Response(201, json={"id": 1})
         return httpx.Response(404)
 
-    workspace_manager = FakeWorkspaceManager(tmp_path / "work")
     task = TaskRunner(
         db=db,
         config=config,
         gitea=make_client(handler),
-        workspace_manager=workspace_manager,
+        workspace_manager=FakeWorkspaceManager(tmp_path / "work"),
         command_runner=FakeRunner(),
         worker_id="worker",
     ).run_once()
@@ -217,11 +208,6 @@ def test_run_task_success_posts_review_comments(db, tmp_path):
     assert task is not None
     assert task.state == TaskState.HUMAN_REVIEW_READY
     assert task.pr_number == 5
-    assert workspace_manager.pushed == ["agent/issue-1-ready-issue"]
-    assert workspace_manager.commit_message == "agent: implement issue #1 - Ready issue"
-    pull_requests = [payload for _, path, payload in requests if path == "/api/v1/repos/acme/service/pulls"]
-    assert pull_requests[0]["title"] == "代理实现：Ready issue"
-    assert "代理实现报告" in pull_requests[0]["body"]
     command = json.loads(db.list_agent_runs(task.id)[0]["command_json"])
     assert command[1] == "--cd"
     assert Path(command[2]).is_absolute()

tests/test_rendering_workspace.py

@@ -38,9 +38,9 @@ def test_prompt_and_pr_body_include_contract_sections(db):
     prompt = render_implementer_prompt(repo, issue, "agent/issue-7-add-thing")
     body = render_pr_body(issue, "## Summary\nDone")
 
-    assert ".agent-output/AGENT_IMPLEMENTATION_REPORT.md" in prompt
+    assert "AGENT_IMPLEMENTATION_REPORT.md" in prompt
-    assert "关联 Issue：#7" in body
+    assert "Closes #7" in body
-    assert "人工审核" in body
+    assert "Human Review Gate" in body
 
 
 def test_review_report_parsing_extracts_verdict_and_comment():