feat: Enhance review prompt with detailed instructions and placeholders for empty sections

src/gitea_codex_bot/services/reviewer.py

@@ -114,11 +114,19 @@ def _build_prompt(
     test_output: str | None,
 ) -> str:
     mode = command.mode if command.name in {"review", "rerun"} else "summary"
+    changed_files = diff_context.get("changed_files") or []
+    changed_files_section = os.linesep.join(changed_files) if changed_files else "(none)"
+    unified_diff = str(diff_context.get("diff", ""))
+    unified_diff_section = unified_diff if unified_diff.strip() else "(empty)"
     return (
         "You are reviewing a Gitea pull request.\n\n"
         "Focus only on issues introduced by this PR.\n"
         "Prioritize correctness, security, data loss, broken behavior, bad migrations, and missing tests.\n"
         "Avoid style nitpicks.\n\n"
+        "You do not have internet/network access. Do not try to fetch URLs.\n"
+        "Use only the PR metadata, changed files, diff, and optional file/test content included below.\n"
+        "Never claim that PR content is inaccessible or missing if these sections are present.\n"
+        "If the changed-file list is `(none)` and unified diff is `(empty)`, treat this as a no-op PR and explain that no code changes were detected.\n\n"
         "Return JSON only with schema:\n"
         "{\n"
         '  "verdict": "correct" | "has_issues",\n'
@@ -132,8 +140,8 @@ def _build_prompt(
         f"Trigger message: {command.raw}\n"
         f"Repo focus: {', '.join(repo_cfg.focus)}\n"
         f"Diff truncated: {diff_context['truncated']}\n"
-        f"Changed files:\n{os.linesep.join(diff_context['changed_files'])}\n\n"
-        f"Unified diff:\n{diff_context['diff']}\n\n"
+        f"Changed files:\n{changed_files_section}\n\n"
+        f"Unified diff:\n{unified_diff_section}\n\n"
         f"Changed file content (optional):\n{changed_file_contents or '(not included)'}\n\n"
         f"Test output (optional):\n{test_output or '(not included)'}\n"
     )

src/gitea_codex_bot/workers/container_runner.py

@@ -4,7 +4,6 @@ import base64
 import json
 import logging
 import os
-import re
 import shlex
 import subprocess
 import uuid
@@ -12,14 +11,47 @@ from pathlib import Path
 from typing import Any
 
 from gitea_codex_bot.config import Settings
-from gitea_codex_bot.services.gitea import GiteaClient
-from gitea_codex_bot.services.repo_config import RepoReviewConfig
-from gitea_codex_bot.services.reviewer import normalize_review_result, prepare_review_prompt
+from gitea_codex_bot.services.gitea import GiteaClient, PullRequestContext
+from gitea_codex_bot.services.repo_config import RepoReviewConfig, parse_repo_review_config_text
+from gitea_codex_bot.services.reviewer import normalize_review_result
 from gitea_codex_bot.types import ParsedCommand
 
 CONTAINER_CODEX_HOME = "/root/.codex"
+REVIEW_OUTPUT_FILE = "/tmp/codex-review-result.json"
+REVIEW_SCHEMA_FILE = "/tmp/codex-review-schema.json"
+RESULT_START_MARKER = "__CODEX_REVIEW_RESULT_BEGIN__"
+RESULT_END_MARKER = "__CODEX_REVIEW_RESULT_END__"
 logger = logging.getLogger(__name__)
 
+REVIEW_RESULT_SCHEMA: dict[str, Any] = {
+    "type": "object",
+    "additionalProperties": True,
+    "required": ["verdict", "confidence", "summary", "findings", "markdown_comment"],
+    "properties": {
+        "verdict": {"type": "string", "enum": ["correct", "has_issues"]},
+        "confidence": {"type": "number"},
+        "summary": {"type": "string"},
+        "markdown_comment": {"type": "string"},
+        "findings": {
+            "type": "array",
+            "items": {
+                "type": "object",
+                "additionalProperties": True,
+                "required": ["severity", "file", "line_start", "line_end", "title", "body"],
+                "properties": {
+                    "severity": {"type": "string", "enum": ["low", "medium", "high", "critical"]},
+                    "file": {"type": "string"},
+                    "line_start": {"type": "integer"},
+                    "line_end": {"type": "integer"},
+                    "title": {"type": "string"},
+                    "body": {"type": "string"},
+                    "suggestion": {"type": "string"},
+                },
+            },
+        },
+    },
+}
+
 
 def run_review_ephemeral(
     settings: Settings,
@@ -29,14 +61,27 @@ def run_review_ephemeral(
     command: ParsedCommand,
 ) -> tuple[dict[str, Any], RepoReviewConfig]:
     gitea = GiteaClient(settings)
-    prompt, _diff_context, repo_cfg = prepare_review_prompt(settings, gitea, repo, pr_number, command)
+    pr = gitea.get_pull_request(repo, pr_number)
+    repo_cfg = _load_repo_review_config_from_gitea(gitea, repo, pr.head_sha)
+    _apply_repo_default_review_mode(command, repo_cfg)
+    prompt = _build_review_instructions(command, repo_cfg)
     container_name = f"codex-review-{uuid.uuid4().hex[:12]}"
-    extra_env: dict[str, str] = {}
+    extra_env: dict[str, str] = {
+        "GITEA_TOKEN": settings.gitea_token.get_secret_value(),
+        "GITEA_GIT_USERNAME": settings.gitea_bot_username,
+    }
+    if settings.openai_api_key:
+        extra_env["OPENAI_API_KEY"] = settings.openai_api_key.get_secret_value()
+    if settings.openai_org_id:
+        extra_env["OPENAI_ORG_ID"] = settings.openai_org_id
+    if settings.openai_project_id:
+        extra_env["OPENAI_PROJECT_ID"] = settings.openai_project_id
     if settings.codex_auth_mode == "chatgpt":
         extra_env["CODEX_AUTH_JSON_B64"] = _load_codex_auth_json_b64(settings)
     try:
         completed = _run_ephemeral_container(
             settings,
+            pr=pr,
             container_name=container_name,
             prompt=prompt,
             extra_env=extra_env,
@@ -46,14 +91,18 @@ def run_review_ephemeral(
             logger.info("Ephemeral runner does not support --reasoning-effort; retrying without it.")
             completed = _run_ephemeral_container(
                 settings,
+                pr=pr,
                 container_name=container_name,
                 prompt=prompt,
                 extra_env=extra_env,
                 include_reasoning_effort=False,
             )
         if completed.returncode != 0:
+            compat_failure = _summarize_review_prompt_compat_failure(completed)
+            if compat_failure:
+                raise RuntimeError(compat_failure)
             raise RuntimeError(_format_runner_failure(completed))
-        parsed = _parse_codex_exec_stdout(completed.stdout)
+        parsed = _parse_review_result_from_stdout_artifact(completed.stdout)
         parsed["_meta"] = _extract_result_meta_from_codex_stdout(completed.stdout, settings)
         return normalize_review_result(parsed), repo_cfg
     except Exception as exc:
@@ -64,12 +113,13 @@ def run_review_ephemeral(
 def _run_ephemeral_container(
     settings: Settings,
     *,
+    pr: PullRequestContext,
     container_name: str,
     prompt: str,
     extra_env: dict[str, str],
     include_reasoning_effort: bool,
 ) -> subprocess.CompletedProcess[str]:
-    install_and_run = _build_install_and_run_command(settings, include_reasoning_effort=include_reasoning_effort)
+    install_and_run = _build_install_and_run_command(settings, pr=pr, include_reasoning_effort=include_reasoning_effort)
     cmd = _build_docker_command(settings, container_name=container_name, install_and_run=install_and_run)
     return subprocess.run(
         cmd,
@@ -82,8 +132,25 @@ def _run_ephemeral_container(
     )
 
 
-def _build_install_and_run_command(settings: Settings, *, include_reasoning_effort: bool = True) -> str:
+def _build_install_and_run_command(
+    settings: Settings,
+    *,
+    pr: PullRequestContext,
+    include_reasoning_effort: bool = True,
+) -> str:
     steps = ["set -euo pipefail"]
+    if settings.codex_auth_mode != "chatgpt":
+        steps.extend(
+            [
+                'if [ -z "${OPENAI_API_KEY:-}" ]; then echo "OPENAI_API_KEY missing in runner env" >&2; exit 8; fi',
+            ]
+        )
+    steps.extend(
+        [
+            'if [ -z "${GITEA_TOKEN:-}" ]; then echo "GITEA_TOKEN missing in runner env" >&2; exit 8; fi',
+            'if [ -z "${GITEA_GIT_USERNAME:-}" ]; then echo "GITEA_GIT_USERNAME missing in runner env" >&2; exit 8; fi',
+        ]
+    )
     if settings.codex_auth_mode == "chatgpt":
         steps.extend(
             [
@@ -94,19 +161,50 @@ def _build_install_and_run_command(settings: Settings, *, include_reasoning_effo
         )
     steps.extend(
         [
-            "apt-get update >/tmp/apt-update.log 2>&1 && apt-get install -y --no-install-recommends ca-certificates >/tmp/apt-install.log 2>&1 || { rc=$?; echo 'ca-certificates install failed'; tail -n 80 /tmp/apt-update.log || true; tail -n 80 /tmp/apt-install.log || true; exit $rc; }",
+            "apt-get update >/tmp/apt-update.log 2>&1 && apt-get install -y --no-install-recommends ca-certificates git >/tmp/apt-install.log 2>&1 || { rc=$?; echo 'ca-certificates/git install failed'; tail -n 80 /tmp/apt-update.log || true; tail -n 80 /tmp/apt-install.log || true; exit $rc; }",
             "npm install -g @openai/codex >/tmp/codex-install.log 2>&1 || { rc=$?; echo 'codex install failed'; tail -n 200 /tmp/codex-install.log || true; exit $rc; }",
         ]
     )
+    schema_json = json.dumps(REVIEW_RESULT_SCHEMA, separators=(",", ":"))
+    steps.extend(
+        [
+            f"cat > {REVIEW_SCHEMA_FILE} <<'JSON'\n{schema_json}\nJSON",
+            'auth_b64="$(printf "%s" "${GITEA_GIT_USERNAME}:${GITEA_TOKEN}" | base64 | tr -d \'\\n\')"',
+            f'git -c http.extraHeader="Authorization: Basic $auth_b64" clone --no-tags --depth 80 {shlex.quote(pr.clone_url)} /work/repo',
+            "cd /work/repo",
+            f'git -c http.extraHeader="Authorization: Basic $auth_b64" fetch --no-tags origin {shlex.quote(pr.base_ref)} {shlex.quote(pr.head_ref)}',
+            f"git checkout --detach {shlex.quote(pr.head_sha)}",
+            'resolved_head="$(git rev-parse HEAD)"',
+            f'if [ "$resolved_head" != {shlex.quote(pr.head_sha)} ]; then echo "Checked out SHA mismatch: expected {pr.head_sha}, got $resolved_head" >&2; exit 9; fi',
+            "unset GITEA_TOKEN auth_b64",
+            "git config --global --unset-all http.extraHeader >/dev/null 2>&1 || true",
+        ]
+    )
     model = settings.openai_review_model.strip()
     reasoning_effort = settings.openai_reasoning_effort.strip()
-    codex_exec_parts = ["codex exec --skip-git-repo-check --json"]
+    codex_exec_parts = [
+        "codex exec review",
+        f"--base {shlex.quote(pr.base_sha)}",
+        "--json",
+        "--output-schema",
+        shlex.quote(REVIEW_SCHEMA_FILE),
+        "-o",
+        shlex.quote(REVIEW_OUTPUT_FILE),
+    ]
     if model:
         codex_exec_parts.append(f"-m {shlex.quote(model)}")
     if include_reasoning_effort and reasoning_effort:
         codex_exec_parts.append(f"--reasoning-effort {shlex.quote(reasoning_effort)}")
-    steps.append(" ".join(codex_exec_parts))
-    return "; ".join(steps)
+    codex_exec_parts.append("-")
+    steps.extend(
+        [
+            " ".join(codex_exec_parts),
+            f'echo "{RESULT_START_MARKER}"',
+            f"cat {shlex.quote(REVIEW_OUTPUT_FILE)}",
+            f'echo "{RESULT_END_MARKER}"',
+        ]
+    )
+    return "\n".join(steps)
 
 
 def _needs_reasoning_effort_compat_retry(completed: subprocess.CompletedProcess[str]) -> bool:
@@ -147,6 +245,14 @@ def _build_docker_command(settings: Settings, *, container_name: str, install_an
                 "OPENAI_PROJECT_ID",
             ]
         )
+    cmd.extend(
+        [
+            "-e",
+            "GITEA_TOKEN",
+            "-e",
+            "GITEA_GIT_USERNAME",
+        ]
+    )
     cmd.extend([settings.review_runner_image, "bash", "-lc", install_and_run])
     return cmd
 
@@ -215,27 +321,67 @@ def ensure_workdir(path: str) -> Path:
     return target
 
 
-def _parse_codex_exec_stdout(stdout: str) -> dict[str, Any]:
-    last_text: str | None = None
-    for line in stdout.splitlines():
-        line = line.strip()
-        if not line:
-            continue
+def _load_repo_review_config_from_gitea(gitea: GiteaClient, repo: str, head_sha: str) -> RepoReviewConfig:
+    content = gitea.get_file_content(repo, ".codex-review.yml", ref=head_sha)
+    if content is None:
+        return RepoReviewConfig(configured=False)
+    return parse_repo_review_config_text(content, configured=True)
+
+
+def _apply_repo_default_review_mode(command: ParsedCommand, repo_cfg: RepoReviewConfig) -> None:
+    if command.name != "review" or command.mode_explicit:
+        return
+    configured_mode = repo_cfg.default_mode
+    command.mode = configured_mode if configured_mode in {"summary", "security", "performance", "tests", "full"} else "summary"
+
+
+def _build_review_instructions(command: ParsedCommand, repo_cfg: RepoReviewConfig) -> str:
+    focus = ", ".join(repo_cfg.focus) if repo_cfg.focus else "correctness, security, maintainability"
+    ignore = ", ".join(repo_cfg.ignore) if repo_cfg.ignore else "(none)"
+    lines = [
+        "Review this pull request using local git data in this checkout only.",
+        "Focus on issues introduced by this PR.",
+        "Prioritize correctness, security, data loss, broken behavior, bad migrations, and missing tests.",
+        "Avoid style-only nitpicks.",
+        f"Requested mode: {command.mode}",
+        f"Command: {command.raw}",
+        f"Focus areas: {focus}",
+        f"Ignore patterns: {ignore}",
+        f"Repository include_tests setting: {repo_cfg.include_tests}",
+        f"Full-content review requested: {command.full}",
+        "Return strict JSON matching the provided output schema.",
+    ]
+    return "\n".join(lines)
+
+
+def _parse_review_result_from_stdout_artifact(stdout: str) -> dict[str, Any]:
+    start = stdout.find(RESULT_START_MARKER)
+    end = stdout.find(RESULT_END_MARKER)
+    if start == -1 or end == -1 or end <= start:
+        raise RuntimeError("Runner output did not include final review artifact markers.")
+    artifact = stdout[start + len(RESULT_START_MARKER) : end].strip()
+    if not artifact:
+        raise RuntimeError("Runner output contained empty final review artifact.")
     try:
-            payload = json.loads(line)
-        except json.JSONDecodeError:
-            continue
-        if isinstance(payload, dict) and {"verdict", "summary", "findings"}.issubset(payload.keys()):
+        payload = json.loads(artifact)
+    except json.JSONDecodeError as exc:
+        raise RuntimeError(f"Final review artifact was not valid JSON: {exc}") from exc
+    if not isinstance(payload, dict):
+        raise RuntimeError(f"Final review artifact JSON must be an object, got {type(payload)!r}.")
     return payload
-        extracted = _extract_text(payload)
-        if extracted:
-            last_text = extracted
-            parsed = _parse_review_json_from_text(extracted)
-            if parsed:
-                return parsed
-    if not last_text:
-        raise RuntimeError("codex exec output did not include parseable review payload text")
-    raise RuntimeError(f"codex exec output text did not contain review JSON; text_tail={_tail_text(last_text, 400)}")
+
+
+def _summarize_review_prompt_compat_failure(completed: subprocess.CompletedProcess[str]) -> str | None:
+    text = " ".join([(completed.stdout or "").strip(), (completed.stderr or "").strip()]).lower()
+    has_prompt_conflict = "prompt" in text and (
+        "cannot be used with" in text or "can't be used with" in text or "incompatible" in text
+    )
+    if "--base" not in text or not has_prompt_conflict:
+        return None
+    return (
+        "Installed Codex CLI rejected `codex exec review --base ...` with custom instructions. "
+        "This runner is configured to fail fast on that compatibility issue."
+    )
 
 
 def _extract_result_meta_from_codex_stdout(stdout: str, settings: Settings) -> dict[str, Any]:
@@ -297,49 +443,3 @@ def _find_first_dict_for_key(payload: Any, key: str) -> dict[str, Any] | None:
             if found:
                 return found
     return None
-
-
-def _parse_review_json_from_text(text: str) -> dict[str, Any] | None:
-    candidates: list[str] = [text.strip()]
-    fenced = re.search(r"```(?:json)?\s*(\{.*\})\s*```", text, flags=re.DOTALL | re.IGNORECASE)
-    if fenced:
-        candidates.append(fenced.group(1).strip())
-    start = text.find("{")
-    end = text.rfind("}")
-    if start != -1 and end != -1 and end > start:
-        candidates.append(text[start : end + 1].strip())
-    seen: set[str] = set()
-    for candidate in candidates:
-        if not candidate or candidate in seen:
-            continue
-        seen.add(candidate)
-        try:
-            payload = json.loads(candidate)
-        except json.JSONDecodeError:
-            continue
-        if isinstance(payload, dict) and {"verdict", "summary", "findings"}.issubset(payload.keys()):
-            return payload
-    return None
-
-
-def _extract_text(payload: Any) -> str | None:
-    if isinstance(payload, str):
-        return payload
-    if isinstance(payload, dict):
-        for key in ("text", "message", "content", "output"):
-            value = payload.get(key)
-            text = _extract_text(value)
-            if text:
-                return text
-        for value in payload.values():
-            if not isinstance(value, (dict, list)):
-                continue
-            text = _extract_text(value)
-            if text:
-                return text
-    if isinstance(payload, list):
-        for item in payload:
-            text = _extract_text(item)
-            if text:
-                return text
-    return None

tests/test_container_runner.py

@@ -5,18 +5,37 @@ from pathlib import Path
 import pytest
 
 from gitea_codex_bot.config import get_settings
+from gitea_codex_bot.services.gitea import PullRequestContext
+from gitea_codex_bot.types import ParsedCommand
 from gitea_codex_bot.workers.container_runner import (
     CONTAINER_CODEX_HOME,
+    RESULT_END_MARKER,
+    RESULT_START_MARKER,
     _build_docker_command,
     _build_install_and_run_command,
     _extract_result_meta_from_codex_stdout,
     _load_codex_auth_json_b64,
-    _parse_codex_exec_stdout,
+    _load_repo_review_config_from_gitea,
+    _parse_review_result_from_stdout_artifact,
     _resolve_codex_auth_json_path,
     run_review_ephemeral,
 )
 
 
+def _sample_pr() -> PullRequestContext:
+    return PullRequestContext(
+        repo="acme/repo",
+        pr_number=1,
+        base_ref="main",
+        base_sha="b" * 40,
+        head_ref="feature",
+        head_sha="a" * 40,
+        clone_url="https://gitea.test/acme/repo.git",
+        html_url="https://gitea.test/acme/repo/pulls/1",
+        is_fork=False,
+    )
+
+
 def test_build_docker_command_api_key_mode_uses_openai_env() -> None:
     settings = get_settings()
 
@@ -25,6 +44,8 @@ def test_build_docker_command_api_key_mode_uses_openai_env() -> None:
     assert "OPENAI_API_KEY" in cmd
     assert "OPENAI_ORG_ID" in cmd
     assert "OPENAI_PROJECT_ID" in cmd
+    assert "GITEA_TOKEN" in cmd
+    assert "GITEA_GIT_USERNAME" in cmd
     assert "--mount" not in cmd
 
 
@@ -45,37 +66,40 @@ def test_build_docker_command_chatgpt_mode_mounts_auth_json(
     assert "OPENAI_API_KEY" not in cmd
     assert f"CODEX_HOME={CONTAINER_CODEX_HOME}" in env_items
     assert "CODEX_AUTH_JSON_B64" in env_items
+    assert "GITEA_TOKEN" in env_items
+    assert "GITEA_GIT_USERNAME" in env_items
 
 
-def test_build_install_command_chatgpt_mode_copies_auth_json(monkeypatch: pytest.MonkeyPatch, tmp_path: Path) -> None:
+def test_build_install_command_chatgpt_mode_sets_git_checkout_and_review(monkeypatch: pytest.MonkeyPatch, tmp_path: Path) -> None:
     auth_file = tmp_path / "auth.json"
     auth_file.write_text("{}", encoding="utf-8")
     monkeypatch.setenv("CODEX_AUTH_MODE", "chatgpt")
     monkeypatch.setenv("CODEX_AUTH_JSON_PATH", str(auth_file))
     get_settings.cache_clear()
     settings = get_settings()
+    pr = _sample_pr()
 
-    command = _build_install_and_run_command(settings)
+    command = _build_install_and_run_command(settings, pr=pr)
 
     assert 'printf "%s" "$CODEX_AUTH_JSON_B64" | base64 -d > /root/.codex/auth.json' in command
-    assert "codex exec --skip-git-repo-check --json -m gpt-5.3-codex" in command
-    assert f"--reasoning-effort {settings.openai_reasoning_effort}" in command
-
-
-def test_build_install_command_includes_configured_reasoning_effort(monkeypatch: pytest.MonkeyPatch) -> None:
-    monkeypatch.setenv("OPENAI_REASONING_EFFORT", "medium")
-    get_settings.cache_clear()
-    settings = get_settings()
-
-    command = _build_install_and_run_command(settings)
-
-    assert "--reasoning-effort medium" in command
+    assert "git -c http.extraHeader=" in command
+    assert f"clone --no-tags --depth 80 {pr.clone_url} /work/repo" in command
+    assert f"fetch --no-tags origin {pr.base_ref} {pr.head_ref}" in command
+    assert f"git checkout --detach {pr.head_sha}" in command
+    assert "resolved_head=\"$(git rev-parse HEAD)\"" in command
+    assert "unset GITEA_TOKEN auth_b64" in command
+    assert f"codex exec review --base {pr.base_sha}" in command
+    assert "--output-schema /tmp/codex-review-schema.json" in command
+    assert "-o /tmp/codex-review-result.json" in command
+    assert f'echo "{RESULT_START_MARKER}"' in command
+    assert f'echo "{RESULT_END_MARKER}"' in command
 
 
 def test_build_install_command_can_disable_reasoning_effort_flag() -> None:
     settings = get_settings()
+    pr = _sample_pr()
 
-    command = _build_install_and_run_command(settings, include_reasoning_effort=False)
+    command = _build_install_and_run_command(settings, pr=pr, include_reasoning_effort=False)
 
     assert "--reasoning-effort" not in command
 
@@ -104,6 +128,29 @@ def test_load_codex_auth_json_b64_roundtrip(monkeypatch: pytest.MonkeyPatch, tmp
     assert encoded
 
 
+def test_load_repo_review_config_from_gitea_when_missing() -> None:
+    class _Gitea:
+        def get_file_content(self, *_args, **_kwargs):
+            return None
+
+    cfg = _load_repo_review_config_from_gitea(_Gitea(), "acme/repo", "a" * 40)
+
+    assert cfg.configured is False
+    assert cfg.enabled is True
+
+
+def test_load_repo_review_config_from_gitea_when_present() -> None:
+    class _Gitea:
+        def get_file_content(self, *_args, **_kwargs):
+            return "enabled: false\nreview:\n  default_mode: tests\n"
+
+    cfg = _load_repo_review_config_from_gitea(_Gitea(), "acme/repo", "a" * 40)
+
+    assert cfg.configured is True
+    assert cfg.enabled is False
+    assert cfg.default_mode == "tests"
+
+
 def test_run_review_ephemeral_chatgpt_does_not_fallback_to_api_key_path(
     monkeypatch: pytest.MonkeyPatch,
     tmp_path: Path,
@@ -115,18 +162,22 @@ def test_run_review_ephemeral_chatgpt_does_not_fallback_to_api_key_path(
     get_settings.cache_clear()
     settings = get_settings()
 
-    monkeypatch.setattr(
-        "gitea_codex_bot.workers.container_runner.prepare_review_prompt",
-        lambda *_args, **_kwargs: ("prompt", {"diff": ""}, object()),
-    )
-    monkeypatch.setattr("gitea_codex_bot.workers.container_runner.GiteaClient", lambda _settings: object())
+    class _FakeGiteaClient:
+        def __init__(self, _settings) -> None:
+            pass
+
+        def get_pull_request(self, *_args, **_kwargs):
+            return _sample_pr()
+
+        def get_file_content(self, *_args, **_kwargs):
+            return None
+
+    monkeypatch.setattr("gitea_codex_bot.workers.container_runner.GiteaClient", _FakeGiteaClient)
     monkeypatch.setattr(
         "gitea_codex_bot.workers.container_runner.subprocess.run",
         lambda *_args, **_kwargs: (_ for _ in ()).throw(RuntimeError("docker unavailable")),
     )
 
-    from gitea_codex_bot.types import ParsedCommand
-
     result, _repo_cfg = run_review_ephemeral(
         settings,
         repo="acme/repo",
@@ -142,18 +193,22 @@ def test_run_review_ephemeral_api_key_mode_does_not_fallback_to_host(monkeypatch
     get_settings.cache_clear()
     settings = get_settings()
 
-    monkeypatch.setattr(
-        "gitea_codex_bot.workers.container_runner.prepare_review_prompt",
-        lambda *_args, **_kwargs: ("prompt", {"diff": ""}, object()),
-    )
-    monkeypatch.setattr("gitea_codex_bot.workers.container_runner.GiteaClient", lambda _settings: object())
+    class _FakeGiteaClient:
+        def __init__(self, _settings) -> None:
+            pass
+
+        def get_pull_request(self, *_args, **_kwargs):
+            return _sample_pr()
+
+        def get_file_content(self, *_args, **_kwargs):
+            return None
+
+    monkeypatch.setattr("gitea_codex_bot.workers.container_runner.GiteaClient", _FakeGiteaClient)
     monkeypatch.setattr(
         "gitea_codex_bot.workers.container_runner.subprocess.run",
         lambda *_args, **_kwargs: (_ for _ in ()).throw(RuntimeError("docker unavailable")),
     )
 
-    from gitea_codex_bot.types import ParsedCommand
-
     result, _repo_cfg = run_review_ephemeral(
         settings,
         repo="acme/repo",
@@ -169,12 +224,17 @@ def test_run_review_ephemeral_retries_without_reasoning_effort_when_unsupported(
     get_settings.cache_clear()
     settings = get_settings()
 
-    monkeypatch.setattr(
-        "gitea_codex_bot.workers.container_runner.prepare_review_prompt",
-        lambda *_args, **_kwargs: ("prompt", {"diff": ""}, object()),
-    )
-    monkeypatch.setattr("gitea_codex_bot.workers.container_runner.GiteaClient", lambda _settings: object())
+    class _FakeGiteaClient:
+        def __init__(self, _settings) -> None:
+            pass
 
+        def get_pull_request(self, *_args, **_kwargs):
+            return _sample_pr()
+
+        def get_file_content(self, *_args, **_kwargs):
+            return None
+
+    monkeypatch.setattr("gitea_codex_bot.workers.container_runner.GiteaClient", _FakeGiteaClient)
     calls: list[list[str]] = []
 
     def _fake_run(cmd, *args, **kwargs):
@@ -194,15 +254,18 @@ def test_run_review_ephemeral_retries_without_reasoning_effort_when_unsupported(
             (),
             {
                 "returncode": 0,
-                "stdout": '{"verdict":"correct","confidence":0.9,"summary":"ok","findings":[]}\n',
+                "stdout": (
+                    '{"type":"response.started","model":"gpt-5.3-codex"}\n'
+                    f"{RESULT_START_MARKER}\n"
+                    '{"verdict":"correct","confidence":0.9,"summary":"ok","findings":[],"markdown_comment":"ok"}\n'
+                    f"{RESULT_END_MARKER}\n"
+                ),
                 "stderr": "",
             },
         )()
 
     monkeypatch.setattr("gitea_codex_bot.workers.container_runner.subprocess.run", _fake_run)
 
-    from gitea_codex_bot.types import ParsedCommand
-
     result, _repo_cfg = run_review_ephemeral(
         settings,
         repo="acme/repo",
@@ -218,33 +281,26 @@ def test_run_review_ephemeral_retries_without_reasoning_effort_when_unsupported(
     assert "--reasoning-effort" not in second_shell
 
 
-def test_parse_codex_exec_stdout_from_stream_item_text_json() -> None:
-    stdout = '\n'.join(
-        [
-            '{"type":"thread.started","thread_id":"abc"}',
-            '{"type":"item.completed","item":{"type":"agent_message","text":"{\\"verdict\\":\\"correct\\",\\"confidence\\":0.9,\\"summary\\":\\"ok\\",\\"findings\\":[]}"}}',
-        ]
+def test_parse_review_result_from_stdout_artifact() -> None:
+    stdout = (
+        "noise\n"
+        f"{RESULT_START_MARKER}\n"
+        '{"verdict":"correct","confidence":0.9,"summary":"ok","findings":[],"markdown_comment":"ok"}\n'
+        f"{RESULT_END_MARKER}\n"
     )
-    parsed = _parse_codex_exec_stdout(stdout)
+    parsed = _parse_review_result_from_stdout_artifact(stdout)
     assert parsed["verdict"] == "correct"
     assert parsed["summary"] == "ok"
 
 
-def test_parse_codex_exec_stdout_from_fenced_json_text() -> None:
-    stdout = '\n'.join(
-        [
-            '{"type":"thread.started","thread_id":"abc"}',
-            '{"type":"item.completed","item":{"type":"agent_message","text":"Here is the result:\\n```json\\n{\\"verdict\\":\\"has_issues\\",\\"confidence\\":0.8,\\"summary\\":\\"x\\",\\"findings\\":[]}\\n```"}}',
-        ]
-    )
-    parsed = _parse_codex_exec_stdout(stdout)
-    assert parsed["verdict"] == "has_issues"
-    assert parsed["summary"] == "x"
+def test_parse_review_result_from_stdout_artifact_fails_without_markers() -> None:
+    with pytest.raises(RuntimeError):
+        _parse_review_result_from_stdout_artifact("no markers here")
 
 
 def test_extract_result_meta_from_codex_stdout_collects_model_and_usage() -> None:
     settings = get_settings()
-    stdout = '\n'.join(
+    stdout = "\n".join(
         [
             '{"type":"response.started","model":"gpt-5.3-codex"}',
             '{"type":"response.completed","response":{"usage":{"input_tokens":101,"output_tokens":22,"total_tokens":123}}}',

tests/test_reviewer_fallback.py

@@ -57,6 +57,27 @@ def test_build_prompt_includes_trigger_message() -> None:
     assert "Trigger message: @codex review security\nPlease focus auth." in prompt
 
 
+def test_build_prompt_uses_empty_placeholders_and_no_network_instruction() -> None:
+    pr = type("PR", (), {"html_url": "https://gitea.example/pr/1"})()
+    command = ParsedCommand(name="review", raw="@codex review")
+    diff_context = {"truncated": False, "changed_files": [], "diff": ""}
+    repo_cfg = RepoReviewConfig()
+
+    prompt = _build_prompt(
+        pr,
+        command,
+        diff_context,
+        repo_cfg,
+        changed_file_contents="",
+        test_output=None,
+    )
+
+    assert "You do not have internet/network access. Do not try to fetch URLs." in prompt
+    assert "Never claim that PR content is inaccessible or missing if these sections are present." in prompt
+    assert "Changed files:\n(none)" in prompt
+    assert "Unified diff:\n(empty)" in prompt
+
+
 def test_prepare_review_prompt_applies_repo_default_mode_when_command_mode_not_explicit(monkeypatch, tmp_path) -> None:
     repo_dir = tmp_path / "repo"
     repo_dir.mkdir(parents=True, exist_ok=True)