rein-aharness/tests/test_runner.py
tegwick 4144eba160 feat: instance manifest, tool profiles, metrics, and budget enforcement
Land HARNESS-WP-0001 T01/T02/T04/T05: extend ADR-005 schedule.yml with
harness fields, named tool-profile registry, ADR-004 metrics writes, and
BudgetTracker wiring. CLI gains validate/profiles; task-file path kept.
2026-07-17 23:49:03 +02:00

208 lines
6 KiB
Python

from __future__ import annotations
import json
import subprocess
from pathlib import Path
import pytest
import yaml
from agent_harness.manifest import HARNESS_MAJOR
from agent_harness.runner import RunResult, run_task
from agent_harness.taskspec import TaskSpec, TaskSpecError
def _make_repo(tmp_path: Path) -> Path:
repo = tmp_path / "sandbox"
repo.mkdir()
subprocess.run(["git", "init", "-q"], cwd=repo, check=True)
(repo / "README.md").write_text("sandbox\n")
subprocess.run(["git", "add", "."], cwd=repo, check=True)
subprocess.run(
["git", "-c", "user.email=t@t", "-c", "user.name=t", "commit", "-qm", "init"],
cwd=repo,
check=True,
)
return repo
def _write_manifest(repo: Path, agents: dict) -> None:
kaizen = repo / ".kaizen"
kaizen.mkdir(exist_ok=True)
data = {"version": "1", "harness": HARNESS_MAJOR, "agents": agents}
(kaizen / "schedule.yml").write_text(
yaml.safe_dump(data, sort_keys=False), encoding="utf-8"
)
class CommittingAdapter:
"""Fake adapter that simulates a session which commits."""
def __init__(self, repo: Path):
self.repo = repo
self.prompts: list[str] = []
self.configs: list = []
def execute_prompt(self, prompt, config):
self.prompts.append(prompt)
self.configs.append(config)
(self.repo / "HELLO.md").write_text("hello\n")
subprocess.run(["git", "add", "."], cwd=self.repo, check=True)
subprocess.run(
["git", "-c", "user.email=t@t", "-c", "user.name=t", "commit", "-qm", "task done"],
cwd=self.repo,
check=True,
)
from llm_connect.models import LLMResponse
return LLMResponse(content="done", model="fake", usage={}, finish_reason="stop")
class IdleAdapter:
def execute_prompt(self, prompt, config):
from llm_connect.models import LLMResponse
return LLMResponse(content="nothing to do", model="fake", usage={}, finish_reason="stop")
class BudgetBlowingAdapter:
def execute_prompt(self, prompt, config):
from llm_connect.exceptions import LLMBudgetExceededError
if config.budget_tracker is not None:
# Simulate preflight/exhaustion the way adapters do.
config.budget_tracker.consume(config.budget_tracker.total)
config.budget_tracker.consume(1)
raise LLMBudgetExceededError(
"Token budget exceeded",
total=1,
spent=1,
requested=1,
)
def _spec(repo: Path, agent: str = "coach") -> TaskSpec:
return TaskSpec(
title="write hello",
description="create HELLO.md",
target_repo=repo,
agent=agent,
)
def test_run_task_success_when_session_commits(tmp_path) -> None:
repo = _make_repo(tmp_path)
adapter = CommittingAdapter(repo)
result = run_task(
_spec(repo), adapter=adapter, report_to_hub=False, write_metrics=True
)
assert isinstance(result, RunResult)
assert result.ok is True
assert result.committed is True
assert result.head_before != result.head_after
assert result.tool_profile == "green-commit-only"
assert "write hello" in adapter.prompts[0]
assert "Never push" in adapter.prompts[0]
assert "Tool profile for this run: green-commit-only" in adapter.prompts[0]
metrics_path = repo / ".kaizen" / "metrics" / "coach" / "executions.jsonl"
assert metrics_path.is_file()
rec = json.loads(metrics_path.read_text().strip().splitlines()[-1])
assert rec["success"] is True
assert rec["committed"] is True
def test_run_task_fails_without_commit(tmp_path) -> None:
repo = _make_repo(tmp_path)
result = run_task(
_spec(repo), adapter=IdleAdapter(), report_to_hub=False, write_metrics=False
)
assert result.ok is False
assert result.committed is False
assert result.reason == "session completed without committing"
def test_run_task_refuses_unknown_tool_profile(tmp_path) -> None:
repo = _make_repo(tmp_path)
_write_manifest(
repo,
{
"coach": {
"cadence": "daily",
"tool_profile": "not-a-real-profile",
}
},
)
result = run_task(
_spec(repo), adapter=IdleAdapter(), report_to_hub=False, write_metrics=False
)
assert result.ok is False
assert result.reason.startswith("refused:")
assert "not-a-real-profile" in result.reason
def test_run_task_resolves_manifest_profile_and_budget(tmp_path) -> None:
repo = _make_repo(tmp_path)
_write_manifest(
repo,
{
"coach": {
"cadence": "daily",
"lane": "blue",
"tool_profile": "blue-mail-triage",
"budget": 25000,
}
},
)
adapter = CommittingAdapter(repo)
result = run_task(
_spec(repo), adapter=adapter, report_to_hub=False, write_metrics=False
)
assert result.ok is True
assert result.tool_profile == "blue-mail-triage"
assert result.budget_tokens == 25000
assert adapter.configs[0].budget_tracker is not None
assert adapter.configs[0].budget_tracker.total == 25000
assert "blue-mail-triage" in adapter.prompts[0]
def test_run_task_budget_exhaustion_fails(tmp_path) -> None:
repo = _make_repo(tmp_path)
_write_manifest(
repo,
{
"coach": {
"cadence": "daily",
"tool_profile": "green-commit-only",
"budget": 10,
}
},
)
result = run_task(
_spec(repo),
adapter=BudgetBlowingAdapter(),
report_to_hub=False,
write_metrics=False,
)
assert result.ok is False
assert "session failed" in result.reason
assert result.budget_tokens == 10
def test_taskspec_rejects_non_repo(tmp_path) -> None:
spec_file = tmp_path / "task.json"
spec_file.write_text(
'{"title": "x", "description": "y", "target_repo": "%s"}' % tmp_path
)
with pytest.raises(TaskSpecError, match="not a git repository"):
TaskSpec.from_file(spec_file)