"""하네스 플러그인 훅과 장부 도구 시험. 표준 라이브러리만 쓴다. python3 -m unittest discover -s plugins/harness/tests 훅 스크립트에 Claude Code가 주는 것과 같은 JSON을 stdin으로 넣고 종료 코드와 출력을 본다. """ import json import os import shutil import subprocess import sys import tempfile import unittest SCRIPTS = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "scripts") REPO = os.path.dirname(os.path.dirname(os.path.dirname(SCRIPTS))) # 시험용 checkout을 둘 /tmp 밖 후보. H12는 /tmp 아래 rm을 일부러 허용하므로 /tmp 안에서는 작업 폴더 판정이 안 보인다 OUTSIDE_TMP_BASES = ("/var/tmp", os.path.expanduser("~")) def run(script, args=(), payload=None, env=None, cwd=None): full_env = {k: v for k, v in os.environ.items() if k != "CLAUDE_PROJECT_DIR"} full_env.update(env or {}) stdin = payload if isinstance(payload, str) else json.dumps(payload or {}) return subprocess.run([sys.executable, "-I", os.path.join(SCRIPTS, script), *args], input=stdin, capture_output=True, text=True, env=full_env, cwd=cwd, timeout=30) def bash(command, cwd=REPO): return {"session_id": "t", "hook_event_name": "PreToolUse", "tool_name": "Bash", "tool_input": {"command": command}, "cwd": cwd} def git(cwd, *args): """시험용 git. 사용자 설정(서명·훅·기본 브랜치)에 따라 결과가 달라지지 않게 부른다.""" subprocess.run(["git", "-c", "user.name=t", "-c", "user.email=t@example.com", "-c", "commit.gpgsign=false", "-c", "core.hooksPath=/dev/null", "-c", "init.defaultBranch=main", "-C", cwd, *args], check=True, capture_output=True, text=True, timeout=30) def outside_tmp_base(): """OUTSIDE_TMP_BASES 중 쓸 수 있는 첫 폴더. 없으면 시험을 건너뛰지 않고 오류로 멈춘다.""" for base in OUTSIDE_TMP_BASES: real = os.path.realpath(base) if os.path.isdir(real) and os.access(real, os.W_OK) and not (real + "/").startswith("/tmp/"): return real raise RuntimeError("/tmp 밖에 시험용 checkout을 만들 곳이 없다: %s" % ", ".join(OUTSIDE_TMP_BASES)) class H12Asserts: ALWAYS = {"HARNESS_H12": "always"} def assert_blocked(self, command, cwd=REPO, env=None): r = run("guard_bash.py", payload=bash(command, cwd), env={**self.ALWAYS, **(env or {})}) self.assertEqual(r.returncode, 2, "막혀야 한다: %s\n%s" % (command, r.stderr)) self.assertIn("[H12", r.stderr) def assert_passed(self, command, cwd=REPO, env=None): r = run("guard_bash.py", payload=bash(command, cwd), env={**self.ALWAYS, **(env or {})}) self.assertEqual(r.returncode, 0, "통과해야 한다: %s\n%s" % (command, r.stderr)) class GuardBashTest(H12Asserts, unittest.TestCase): """H12. 여기 사례는 판정의 대표만 둔다. 많은 사례는 소유자 기계 전역 훅 시험에 있다.""" def test_force_push_to_hosting_remote_is_blocked(self): for c in ("git push --force origin main", "git push -f", "git push origin +feature/x", "git push --force-with-lease origin claude/x", "git push --mirror origin", "ssh build 'cd ~/w/x && git push -f origin main'"): self.assert_blocked(c) def test_path_remote_transport_passes_but_not_onto_main(self): self.assert_passed("git push -q -f desktop:/home/u/w/x v3/check:refs/heads/v3/check") self.assert_blocked("git push -f desktop:/home/u/w/x HEAD:main") def test_normal_push_and_own_branch_delete_pass(self): self.assert_passed("git push -u origin claude/x") self.assert_passed("git push origin --delete review/x") self.assert_blocked("git push origin --delete main") def test_rm_outside_or_too_broad_is_blocked(self): for c in ("rm -rf /", "rm -rf ~", "rm -rf .git", 'rm -rf "$UNSET/"', "ssh build 'rm -rf ~/workspace/some-repo'"): self.assert_blocked(c) def test_rm_inside_project_tmp_and_known_vars_pass(self): for c in ("rm -rf build dist", "rm -rf /tmp/some-build-x", 'T=$(mktemp -d) && rm -rf "$T"', "S=/tmp/x; rm -rf $S/out", 'rm -rf "${OUT:?}/build"', "ssh build 'rm -rf ~/x-logs'"): self.assert_passed(c) def test_data_deletion_is_blocked(self): for c in ("psql -c 'DROP DATABASE app'", "redis-cli FLUSHALL", "./gradlew :app:flywayClean", "ssh build 'docker compose down -v'", "docker volume rm app_db", "kubectl delete pvc data-0"): self.assert_blocked(c) def test_scoped_teardown_and_data_as_text_pass(self): for c in ("docker compose -p lane-1 -f compose.yml down -v", "docker compose down", "kubectl delete pod web-1", "grep -rn 'DROP TABLE' docs", "cat > V2__drop.sql <<'EOF'\nDROP TABLE old;\nEOF"): self.assert_passed(c) def test_skips_when_owner_global_hook_has_h12(self): with tempfile.TemporaryDirectory() as home: os.makedirs(os.path.join(home, ".claude")) settings = {"hooks": {"PreToolUse": [{"matcher": "Bash", "hooks": [ {"type": "command", "command": "python3 -I /x/.claude/hooks/guard_bash.py H12"}]}]}} with open(os.path.join(home, ".claude", "settings.json"), "w") as f: json.dump(settings, f) r = run("guard_bash.py", payload=bash("git push --force origin main"), env={"HOME": home}) self.assertEqual(r.returncode, 0, r.stderr) settings["hooks"]["PreToolUse"][0]["hooks"][0]["command"] = "python3 -I /x/guard_bash.py H1" with open(os.path.join(home, ".claude", "settings.json"), "w") as f: json.dump(settings, f) r = run("guard_bash.py", payload=bash("git push --force origin main"), env={"HOME": home}) self.assertEqual(r.returncode, 2, r.stderr) def test_broken_input_does_not_block(self): r = run("guard_bash.py", payload="{not json", env=self.ALWAYS) self.assertEqual(r.returncode, 1) class GuardRmCheckoutTest(H12Asserts, unittest.TestCase): """H12가 작업 폴더 안 삭제는 통과시키고 작업 폴더 자신·옆 폴더 삭제는 막는지. 이 repo의 checkout을 cwd로 쓰면 결과가 checkout 위치에 따라 달라진다(/tmp 아래면 /tmp 허용에 걸려 통과하고, main checkout 안의 worktree면 main checkout 허용에 걸려 통과한다). 그래서 /tmp 밖에 checkout과 그 안의 worktree를 직접 만들어 cwd로 넘긴다. """ @classmethod def setUpClass(cls): base = tempfile.mkdtemp(prefix="h12-test-", dir=outside_tmp_base()) cls.addClassCleanup(shutil.rmtree, base, True) cls.checkout = os.path.join(base, "repo") git(base, "init", "-q", cls.checkout) git(cls.checkout, "commit", "-q", "--allow-empty", "-m", "init") cls.worktree = os.path.join(cls.checkout, ".claude", "worktrees", "w") cls.sibling = os.path.join(cls.checkout, ".claude", "worktrees", "w2") for wt in (cls.worktree, cls.sibling): git(cls.checkout, "worktree", "add", "-q", "--detach", wt) # scratchpad처럼 /tmp 아래 둔 worktree. /tmp 아래는 일부러 허용하므로 통째로 지워도 통과해야 한다 cls.tmp_worktree = os.path.join(tempfile.mkdtemp(prefix="h12-test-", dir="/tmp"), "wt") cls.addClassCleanup(shutil.rmtree, os.path.dirname(cls.tmp_worktree), True) git(cls.checkout, "worktree", "add", "-q", "--detach", cls.tmp_worktree) def test_rm_inside_checkout_and_worktree_passes(self): # 둘 다 허용 폴더로 잡힌다. 아래 시험이 막히는 것은 위치를 몰라서가 아니라 대상이 넓어서다 for cwd in (self.checkout, self.worktree): for c in ("rm -rf build", "rm -rf build dist node_modules/.cache", "rm -rf ./out/*"): self.assert_passed(c, cwd=cwd) self.assert_passed("rm -rf ../../build", cwd=self.worktree) # main checkout 안, worktree 밖(.claude/build) self.assert_passed("rm -rf .claude/worktrees/w/build", cwd=self.checkout) # main checkout이 worktree 안을 치운다 def test_rm_tmp_worktree_passes(self): for cwd in (self.checkout, self.worktree): self.assert_passed("rm -rf %s" % self.tmp_worktree, cwd=cwd) def test_rm_whole_checkout_or_sibling_is_blocked(self): for c in ("rm -rf ./*", "rm -rf ../other-repo/src", "rm -rf .claude/worktrees", "rm -rf .claude/worktrees/w"): self.assert_blocked(c, cwd=self.checkout) def test_rm_nested_worktree_itself_or_sibling_is_blocked(self): """main checkout 안 worktree에서 worktree 자신, 옆 worktree, 그 안, 둘을 품은 폴더를 지우지 못한다.""" for c in ("rm -rf ./*", "rm -rf ../w2", "rm -rf ../w2/src", "rm -rf ..", "rm -rf ../.."): self.assert_blocked(c, cwd=self.worktree) def test_project_dir_main_checkout_still_guards_nested_worktree(self): """앱이 worktree 세션의 CLAUDE_PROJECT_DIR를 main checkout으로 줄 때도 같은 판정이다(세션 cwd가 먼저다).""" env = {"CLAUDE_PROJECT_DIR": self.checkout} for c in ("rm -rf ./*", "rm -rf ../w2", "rm -rf ../w2/src"): self.assert_blocked(c, cwd=self.worktree, env=env) for c in ("rm -rf build", "rm -rf ../../build"): self.assert_passed(c, cwd=self.worktree, env=env) self.assert_passed("rm -rf .claude/worktrees/w/build", cwd=self.checkout, env=env) class NudgeSecretsTest(unittest.TestCase): def nudge(self, content, tool="Write"): key = "content" if tool == "Write" else "new_string" r = run("nudge_secrets.py", payload={"tool_name": tool, "tool_input": {"file_path": "a.yml", key: content}}) self.assertEqual(r.returncode, 0, r.stderr) return json.loads(r.stdout)["hookSpecificOutput"]["additionalContext"] if r.stdout.strip() else "" def test_real_looking_secrets_nudge_and_are_masked(self): token = "ghp_" + "a1B2" * 9 for content in ("-----BEGIN OPENSSH PRIVATE KEY-----", "token: " + token, 'db:\n password: "s3cr3t-Value"', "AWS_KEY=AKIA" + "ABCDEFGHIJKLMNOP"): text = self.nudge(content) self.assertIn("[N4", text, content) self.assertNotIn(token, text) def test_edit_new_string_is_checked(self): self.assertIn("[N4", self.nudge("api_key = 'q8Zr2LmP0xY'", tool="Edit")) def test_placeholders_and_references_are_quiet(self): for content in ("password: ${DB_PASSWORD}", "password=changeme", "api_key: ", "AWS_KEY=AKIAIOSFODNN7EXAMPLE", "max_tokens: 4096", "TOKEN_TTL=3600", "password = request.getParameter(\"pw\")", "secret: dummy-secret-value"): self.assertEqual(self.nudge(content), "", content) class LedgerTest(unittest.TestCase): def setUp(self): self.tmp = tempfile.TemporaryDirectory() self.dir = self.tmp.name self.env = {"HARNESS_STATE_DIR": os.path.join(self.dir, "state")} self.transcript = os.path.join(self.dir, "abc.jsonl") def tearDown(self): self.tmp.cleanup() def write_transcript(self, lines, mode="w"): with open(self.transcript, mode) as f: for line in lines: f.write(json.dumps(line) + "\n") @staticmethod def assistant(msg_id, ts, out, cache_read=100, inp=10, created=5): return {"type": "assistant", "timestamp": ts, "message": {"id": msg_id, "usage": { "input_tokens": inp, "cache_creation_input_tokens": created, "cache_read_input_tokens": cache_read, "output_tokens": out}}} def test_usage_dedupes_message_ids_and_caps_gaps(self): self.write_transcript([ {"type": "user", "timestamp": "2026-11-07T10:00:00Z"}, self.assistant("m1", "2026-11-07T10:01:00Z", 50), self.assistant("m1", "2026-11-07T10:01:01Z", 70), # 같은 메시지가 나뉘어 기록됨: 한 번, 큰 값 self.assistant("m2", "2026-11-07T10:31:01Z", 30, cache_read=400), # 30분 빈틈은 5분만 ]) sub = os.path.join(self.dir, "abc", "subagents") os.makedirs(sub) with open(os.path.join(sub, "agent-1.jsonl"), "w") as f: f.write(json.dumps(self.assistant("s1", "2026-11-07T10:02:00Z", 20)) + "\n") r = run("ledger.py", ["usage", self.transcript], env=self.env) self.assertEqual(r.returncode, 0, r.stderr) u = json.loads(r.stdout) self.assertEqual((u["calls"], u["subagent_calls"], u["output_tokens"]), (3, 1, 120)) self.assertEqual(u["cache_read_tokens"], 600) self.assertEqual(u["max_context"], 415) self.assertEqual((u["minutes"], u["active_minutes"]), (31, 6)) def test_stop_hook_reads_incrementally_and_matches_full_scan(self): self.write_transcript([self.assistant("m1", "2026-11-07T10:00:00Z", 10)]) hook = {"session_id": "s-1", "transcript_path": self.transcript, "cwd": self.dir} self.assertEqual(run("ledger.py", ["stop"], payload=hook, env=self.env).returncode, 0) self.write_transcript([self.assistant("m2", "2026-11-07T10:02:00Z", 5)], mode="a") with open(self.transcript, "a") as f: f.write('{"type": "assistant", "timestamp": "2026-11-07T10:0') # 쓰는 중인 줄은 다음에 self.assertEqual(run("ledger.py", ["stop"], payload=hook, env=self.env).returncode, 0) with open(os.path.join(self.env["HARNESS_STATE_DIR"], "usage", "s-1.json")) as f: saved = json.load(f)["summary"] self.assertEqual((saved["calls"], saved["output_tokens"], saved["active_minutes"]), (2, 15, 2)) def test_session_start_records_and_hooks_never_print(self): hook = {"session_id": "s-2", "transcript_path": self.transcript, "cwd": self.dir, "source": "startup"} r = run("ledger.py", ["session-start"], payload=hook, env=self.env) self.assertEqual((r.returncode, r.stdout), (0, "")) with open(os.path.join(self.env["HARNESS_STATE_DIR"], "sessions.jsonl")) as f: self.assertEqual(json.loads(f.readline())["session_id"], "s-2") r = run("ledger.py", ["stop"], payload="{broken", env=self.env) self.assertEqual((r.returncode, r.stdout), (0, "")) def test_append_checks_fields_and_actual_computes_within(self): ledger = os.path.join(self.dir, "docs", "ledger", "estimates.jsonl") bad = {"kind": "estimate", "task": "t1", "minutes": {"p30": 1}} r = run("ledger.py", ["append", "--ledger", ledger], payload=bad, env=self.env) self.assertEqual(r.returncode, 1) self.assertFalse(os.path.exists(ledger)) good = {"kind": "estimate", "task": "t1", "task_kind": "docs", "size": "small", "minutes": {"p30": 1, "p50": 2, "p80": 10}, "basis": "시험"} self.assertEqual(run("ledger.py", ["append", "--ledger", ledger], payload=good, env=self.env).returncode, 0) self.write_transcript([self.assistant("m1", "2026-11-07T10:00:00Z", 10), self.assistant("m2", "2026-11-07T10:03:00Z", 10)]) r = run("ledger.py", ["actual", "--task", "t1", "--ledger", ledger, self.transcript], env=self.env) row = json.loads(r.stdout) self.assertEqual((row["active_minutes"], row["within"]), (3, "p80")) r = run("ledger.py", ["precedents", "--ledger", ledger], env=self.env) self.assertIn("| t1 | docs | small | 1/2/10 |", r.stdout) retro = {"kind": "retro", "task": "t1", "cause": "모름", "level": "gate"} self.assertEqual(run("ledger.py", ["append", "--ledger", ledger], payload=retro, env=self.env).returncode, 1) def write_ledger(self, path, rows): os.makedirs(os.path.dirname(path), exist_ok=True) with open(path, "a", encoding="utf-8") as f: for row in rows: f.write(json.dumps(row, ensure_ascii=False) + "\n") @staticmethod def pair(task, active, why="", at="2026-11-01T10:00+09:00", p=(10, 20, 30), calls=None): est = {"kind": "estimate", "task": task, "minutes": dict(zip(("p30", "p50", "p80"), p)), "basis": "시험", "recorded_at": at} act = {"kind": "actual", "task": task, "active_minutes": active, "why": why, "recorded_at": at} if calls: est["calls"], act["calls"] = {"p30": 1, "p50": calls[0], "p80": 99}, calls[1] return [est, act] def test_pending_lists_misses_then_tracks_fix_until_verdict(self): ledger = os.path.join(self.dir, "estimates.jsonl") self.write_ledger(ledger, self.pair("in", 25) + self.pair("under", 3) + self.pair("over", 31) + self.pair("noted", 15, why="검증 대기가 길었다")) r = run("ledger.py", ["pending", "--ledger", ledger], env=self.env) self.assertEqual([line.split(":")[0] for line in r.stdout.splitlines()], ["- under", "- over", "- noted"]) retro = {"kind": "retro", "task": "under", "cause": "예측 근거 없음", "level": "architecture", "note": "시험", "fix": "f1", "recorded_at": "2026-11-02T10:00+09:00"} self.assertEqual(run("ledger.py", ["append", "--ledger", ledger], payload=retro, env=self.env).returncode, 0) self.write_ledger(ledger, [{"kind": "estimate", "task": "under", "correction": True, "recorded_at": "2026-11-02T12:00+09:00"}, # 회고한 작업 자신은 세지 않는다 {"kind": "estimate", "task": "c", "minutes": {"p30": 1, "p50": 2, "p80": 9}, "basis": "시험", "recorded_at": "2026-11-02T12:00+09:00"}]) # 실측 전은 세지 않는다 r = run("ledger.py", ["pending", "--ledger", ledger], env=self.env) self.assertNotIn("- under", r.stdout) self.assertIn("- 판정 대기 f1(under, 2026-11-02 적용): 뒤 실측 0건\n", r.stdout) for task, at in (("a", "2026-11-03T10:00+09:00"), ("b", "2026-11-04T10:00+09:00")): self.write_ledger(ledger, [{"kind": "estimate", "task": task, "minutes": {"p30": 1, "p50": 2, "p80": 9}, "basis": "시험", "recorded_at": at}]) r = run("ledger.py", ["append", "--ledger", ledger], env=self.env, payload={"kind": "actual", "task": task, "active_minutes": 40, "recorded_at": at}) self.assertIn("예측 밖이다(예측 1/2/9분, 실측 40분)", r.stdout) self.assertIn("효과를 판정할 개선 1건", r.stdout) self.assertIn("뒤 실측 2건. 판정할 때다", run("ledger.py", ["pending", "--ledger", ledger], env=self.env).stdout) for bad in ({"fix": "f9", "effect": "effective"}, {"fix": "f1", "effect": "모름"}): r = run("ledger.py", ["append", "--ledger", ledger], env=self.env, payload={"kind": "verdict", "task": "b", **bad}) self.assertEqual(r.returncode, 1, bad) verdict = {"kind": "verdict", "task": "b", "fix": "f1", "effect": "reverted", "ref": "abc1234"} self.assertEqual(run("ledger.py", ["append", "--ledger", ledger], payload=verdict, env=self.env).returncode, 0) self.assertNotIn("판정 대기", run("ledger.py", ["pending", "--ledger", ledger], env=self.env).stdout) def test_precedents_shows_bias_across_ledgers(self): one, two = os.path.join(self.dir, "one.jsonl"), os.path.join(self.dir, "two.jsonl") self.write_ledger(one, self.pair("x", 5, calls=(100, 40))) self.write_ledger(two, self.pair("y", 15, calls=(100, 60)) + self.pair("z", 40)) r = run("ledger.py", ["precedents", "--ledger", one, "--ledger", two], env=self.env) self.assertIn("실측/p50 중앙값 시간 0.75(3건) · 호출 0.50(2건). 시간 실측이 p30 아래 1건, p80 위 1건", r.stdout) def test_usage_measures_start_context_and_subagent_returns(self): """평가(v1·v2 비교)에 쓰는 칸: 첫 호출 컨텍스트, Agent 호출 수, 서브에이전트가 돌려준 글자 수.""" self.write_transcript([ self.assistant("m1", "2026-11-07T10:00:00Z", 10, cache_read=900, inp=50, created=50), {"type": "assistant", "timestamp": "2026-11-07T10:00:05Z", "message": { "id": "m2", "usage": {"input_tokens": 1, "output_tokens": 5}, "content": [{"type": "tool_use", "id": "tu1", "name": "Agent", "input": {"prompt": "x"}}, {"type": "tool_use", "id": "tu2", "name": "Bash", "input": {"command": "ls"}}]}}, ]) # 결과가 다음 조각에 온다(Stop 훅은 이어 읽는다) self.write_transcript([ {"type": "user", "timestamp": "2026-11-07T10:01:00Z", "message": {"content": [ {"type": "tool_result", "tool_use_id": "tu1", "content": [{"type": "text", "text": "가" * 40}]}, {"type": "tool_result", "tool_use_id": "tu2", "content": "bash 출력은 세지 않는다"}]}}, {"type": "assistant", "timestamp": "2026-11-07T10:01:05Z", "message": { "id": "m3", "usage": {"output_tokens": 1}, "content": [{"type": "tool_use", "id": "tu3", "name": "Task", "input": {}}]}}, {"type": "user", "timestamp": "2026-11-07T10:02:00Z", "message": {"content": [ {"type": "tool_result", "tool_use_id": "tu3", "content": "short"}]}}, # 백그라운드 서브에이전트의 결과는 대기열 알림으로 온다. 사용자가 끼워 넣은 말은 세지 않는다 {"type": "attachment", "timestamp": "2026-11-07T10:03:00Z", "attachment": {"type": "queued_command", "prompt": "0123456789"}}, {"type": "attachment", "timestamp": "2026-11-07T10:03:01Z", "attachment": {"type": "queued_command", "prompt": "사용자가 중간에 보낸 말"}}, ], mode="a") sub = os.path.join(self.dir, "abc", "subagents") os.makedirs(sub) for i in (1, 2): with open(os.path.join(sub, "agent-%d.jsonl" % i), "w") as f: f.write(json.dumps(self.assistant("s%d" % i, "2026-11-07T10:00:30Z", 7, cache_read=5000)) + "\n") u = json.loads(run("ledger.py", ["usage", self.transcript], env=self.env).stdout) self.assertEqual(u["start_context"], 1000) # 본 기록 첫 호출만, 서브에이전트 첫 호출이 아니다 self.assertEqual((u["agent_calls"], u["agent_return_chars"]), (2, 45 + 35)) # 알림 머리 25자 + 본문 10자 self.assertEqual((u["subagents"], u["subagent_output_tokens"]), (2, 14)) hook = {"session_id": "s-3", "transcript_path": self.transcript, "cwd": self.dir} with open(self.transcript) as f: lines = f.readlines() with open(self.transcript, "w") as f: f.writelines(lines[:2]) # Agent 호출까지만 쓰인 때 한 번 run("ledger.py", ["stop"], payload=hook, env=self.env) with open(self.transcript, "a") as f: f.writelines(lines[2:]) run("ledger.py", ["stop"], payload=hook, env=self.env) with open(os.path.join(self.env["HARNESS_STATE_DIR"], "usage", "s-3.json")) as f: saved = json.load(f)["summary"] self.assertEqual((saved["agent_calls"], saved["agent_return_chars"]), (2, 45 + 35)) # 알림 머리 25자 + 본문 10자 def test_stop_hook_survives_state_saved_by_older_version(self): self.write_transcript([self.assistant("m1", "2026-11-07T10:00:00Z", 10)]) folder = os.path.join(self.env["HARNESS_STATE_DIR"], "usage") os.makedirs(folder) old = {"offset": 0, "ids": {}, "first": None, "last": None, "active": 0.0} # 0.2.0이 저장한 모양 with open(os.path.join(folder, "s-4.json"), "w") as f: json.dump({"session_id": "s-4", "transcript_path": self.transcript, "files": {self.transcript: old}}, f) hook = {"session_id": "s-4", "transcript_path": self.transcript, "cwd": self.dir} r = run("ledger.py", ["stop"], payload=hook, env=self.env) self.assertEqual(r.stderr, "") with open(os.path.join(folder, "s-4.json")) as f: self.assertEqual(json.load(f)["summary"]["calls"], 1) def test_eval_rows_compare_arms_per_case(self): ledger = os.path.join(self.dir, "estimates.jsonl") bad = {"kind": "eval", "task": "e1", "case": "review"} # arm 없음 self.assertEqual(run("ledger.py", ["append", "--ledger", ledger], payload=bad, env=self.env).returncode, 1) rows = [] for arm, mins, findings, start in (("v1", 20, 3, 70000), ("v1", 30, 5, 70000), ("v2", 10, 6, 75000), ("v2", 20, 8, 75000)): rows.append({"kind": "eval", "task": "e1", "case": "review", "arm": arm, "active_minutes": mins, "calls": mins * 2, "output_tokens": mins * 1000, "start_context": start, "max_context": start * 2, "findings": findings, "false_findings": 1}) for row in rows: self.assertEqual(run("ledger.py", ["append", "--ledger", ledger], payload=row, env=self.env).returncode, 0) self.write_ledger(ledger, self.pair("plain", 5)) r = run("ledger.py", ["compare", "--task", "e1", "--ledger", ledger], env=self.env) self.assertEqual(r.returncode, 0, r.stderr) self.assertIn("\n|---|---|---|---|---|---|---|---|---|---|\n", r.stdout) # 마크다운 표 구분선 self.assertIn("| review | v1 | 2 | 25 | 50 | 25000 | 70000 | 140000 | 4 | 1 |", r.stdout) self.assertIn("| review | v2 | 2 | 15 | 30 | 15000 | 75000 | 150000 | 7 | 1 |", r.stdout) self.assertIn("v2/v1 중앙값: 작업 중 분 0.60 · 호출 0.60 · 출력 0.60 · 시작 컨텍스트 1.07 · 발견 1.75", r.stdout) self.assertNotIn("| e1 |", run("ledger.py", ["precedents", "--ledger", ledger], env=self.env).stdout) self.assertIn("평가 줄 없음", run("ledger.py", ["compare", "--task", "none", "--ledger", ledger], env=self.env).stdout) self.write_transcript([self.assistant("m1", "2026-11-07T10:00:00Z", 10)]) r = run("ledger.py", ["eval", "--task", "e1", "--arm", "v2", "--case", "review", self.transcript], env=self.env) draft = json.loads(r.stdout) self.assertEqual((draft["kind"], draft["arm"], draft["case"], draft["calls"]), ("eval", "v2", "review", 1)) self.assertIn("findings", draft) self.assertIn("false_findings", draft) def test_nudge_prints_one_line_once_per_repo_on_startup(self): repo = os.path.join(self.dir, "repo") os.makedirs(repo) subprocess.run(["git", "init", "-q", repo], check=True) hook = {"session_id": "s", "cwd": repo, "source": "startup"} self.assertEqual(run("ledger.py", ["nudge"], payload=hook, env=self.env).stdout, "") # 장부 없음 ledger = os.path.join(repo, "docs", "ledger", "estimates.jsonl") self.write_ledger(ledger, self.pair("in", 25)) self.assertEqual(run("ledger.py", ["nudge"], payload=hook, env=self.env).stdout, "") # 빗나간 것 없음 self.write_ledger(ledger, self.pair("under", 3)) self.assertEqual(run("ledger.py", ["nudge"], payload={**hook, "source": "resume"}, env=self.env).stdout, "") r = run("ledger.py", ["nudge"], payload=hook, env=self.env) self.assertEqual(r.returncode, 0) self.assertEqual(len(r.stdout.splitlines()), 1) self.assertIn("회고 안 한 빗나간 예측 1건", r.stdout) self.assertEqual(run("ledger.py", ["nudge"], payload=hook, env=self.env).stdout, "") # 하루 한 번 r = run("ledger.py", ["nudge"], payload="{broken", env=self.env) self.assertEqual((r.returncode, r.stdout), (0, "")) class TechdebtTest(unittest.TestCase): def test_lists_only_added_markers(self): with tempfile.TemporaryDirectory() as d: def git(*a): subprocess.run(["git", "-C", d, *a], check=True, capture_output=True) git("init", "-q", "-b", "main") git("config", "user.email", "t@example.com") git("config", "user.name", "t") git("config", "core.hooksPath", "/dev/null") with open(os.path.join(d, "a.py"), "w") as f: f.write("# TODO old\n") git("add", ".") git("commit", "-q", "-m", "base") with open(os.path.join(d, "a.py"), "a") as f: f.write("x = 1 # noqa\n# FIXME new\n") r = run("techdebt.py", ["--base", "HEAD"], cwd=d) self.assertEqual(r.returncode, 0, r.stderr) self.assertIn("a.py:3 # FIXME new", r.stdout) self.assertIn("a.py:2 x = 1 # noqa", r.stdout) self.assertNotIn("TODO old", r.stdout) if __name__ == "__main__": unittest.main()