Self-hosted workflows for coding-agent delivery. Orchestrate CodeBuddy, Codex and Claude Code with scripts and human approval gates. Includes Local workflows for coding assistants.
JavaScript agent-orchestrationagentic-workflowai-agentsai-codingclaude-codecode-reviewcodebuddycodexcoding-agentscursordeveloper-toolshuman-on-the-loopllmmcpmulti-agentsdlcself-hostedsopworkflow-automationworkflow-engine
"""E2E: drive the installed Portable workflow end to end with a fake host Agent.
Each test installs the packaged CLI into a sample project, then runs
``fake_agent.py``, which follows the Skill's ``resume``-driven protocol exactly
as a host Agent would, using recorded stage outputs instead of an LLM.
"""
import json
import subprocess
import sys
import unittest
import harness as h
AGENT_STAGES = ("DESIGN", "IMPLEMENT", "REVIEW", "TEST")
STAGE_ORDER = ["REQUIREMENT", "DESIGN", "IMPLEMENT", "REVIEW", "TEST", "SUMMARY"]
def installed_adapter(project, host):
from fake_agent import HOST_SKILL_DIRS
path = project / HOST_SKILL_DIRS[host] / "devflow/adapters" / host / "manifest.json"
return json.loads(path.read_text(encoding="utf-8"))
def events(project, kind):
return [item for item in h.transcript(project) if item["kind"] == kind]
class PortableWorkflowE2E(unittest.TestCase):
def install(self, host="claude"):
project = h.new_project(self)
h.loopforge("skills", "install", host, "--project-root", project)
return project
def assert_delivered(self, project):
state = h.workflow_state(project)
self.assertEqual("completed", state["status"], h.transcript_text(project))
for stage in STAGE_ORDER:
self.assertEqual("completed", state["stages"][stage]["status"], stage)
self.assertEqual("skipped", state["stages"]["KNOWLEDGE"]["status"])
verify = subprocess.run(
[sys.executable, "-m", "unittest", "check_slugify"], cwd=str(project),
capture_output=True, text=True,
)
self.assertEqual(0, verify.returncode, verify.stderr)
self.assertIn("OK", h.loopforge_validate(project))
self.assertEqual("user file that must survive\n",
(project / "USER_NOTES.md").read_text(encoding="utf-8"))
return state
def test_medium_task_runs_every_stage_on_every_host(self):
for host in h.hosts_for("portable"):
with self.subTest(host=host):
project = self.install(host)
h.fake_agent(host, project, check_rc=0)
state = self.assert_delivered(project)
finished = [e["stage"] for e in events(project, "finish") if e["result"] == "completed"]
self.assertEqual(STAGE_ORDER, finished)
prompts = events(project, "prompt")
self.assertEqual(STAGE_ORDER, [p["stage"] for p in prompts])
self.assertTrue(all(p["mentions_role"] for p in prompts), prompts)
adapter = installed_adapter(project, host)
spawned = {e["stage"]: e for e in events(project, "spawned")}
self.assertEqual(set(AGENT_STAGES), set(spawned))
ids = set()
for stage in AGENT_STAGES:
recorded = state["stages"][stage]
self.assertEqual(adapter["stage_executor"], recorded["executor_type"])
self.assertEqual(adapter["dispatch_tool"], recorded["dispatch_tool"])
self.assertEqual(spawned[stage]["executor_id"], recorded["executor_id"])
ids.add(recorded["executor_id"])
# Independent review: every agent stage has its own executor.
self.assertEqual(len(AGENT_STAGES), len(ids))
for stage in ("REQUIREMENT", "SUMMARY"):
self.assertEqual("coordinator", state["stages"][stage]["executor_type"])
def test_resume_after_crash_inside_a_stage(self):
for host in ("claude", "codebuddy"):
with self.subTest(host=host):
project = self.install(host)
h.fake_agent(host, project, "--crash-in", "IMPLEMENT", check_rc=75)
state = h.workflow_state(project)
self.assertEqual("in_progress", state["stages"]["IMPLEMENT"]["status"])
self.assertEqual("in_progress", state["status"])
h.fake_agent(host, project, "--resume", check_rc=0)
self.assert_delivered(project)
sessions = []
for item in h.transcript(project):
if item["session"] not in sessions:
sessions.append(item["session"])
self.assertEqual(2, len(sessions))
second = [e for e in events(project, "action") if e["session"] == sessions[1]]
if host == "codebuddy":
# A new CodeBuddy session must drop the old Team and re-dispatch.
self.assertEqual("rotate-team", second[0]["action"])
self.assertEqual(1, len(events(project, "team-rotated")))
else:
self.assertEqual(("continue-executor", "IMPLEMENT"),
(second[0]["action"], second[0]["stage"]))
self.assertFalse(any(e["kind"] == "init" and e["session"] == sessions[1]
for e in h.transcript(project)))
def test_resume_after_interruption_between_actions(self):
project = self.install("codex")
h.fake_agent("codex", project, "--stop-after", "7", check_rc=75)
self.assertNotEqual("completed", h.workflow_state(project)["status"])
h.fake_agent("codex", project, "--resume", check_rc=0)
self.assert_delivered(project)
finished = [e["stage"] for e in events(project, "finish") if e["result"] == "completed"]
self.assertEqual(STAGE_ORDER, finished, "no stage may run twice after resume")
def test_requirement_gate_waits_for_explicit_user_confirmation(self):
project = self.install("cursor")
h.fake_agent("cursor", project, "--no-user-confirmation", check_rc=3)
state = h.workflow_state(project)
self.assertEqual("awaiting_approval", state["status"])
self.assertEqual("REQUIREMENT", state["awaiting_approval"])
self.assertIsNone(state["stages"]["DESIGN"]["prepared_at"])
self.assertFalse(events(project, "spawned"), "no DESIGN executor before approval")
# The gate must be enforced by the CLI itself, not only by the Agent loop.
state_path = project / "artifacts/e2e-task/workflow-state.json"
premature = h.run_workflow(
project, "prepare", "--state", state_path, "--stage", "DESIGN", check=False
)
self.assertNotEqual(0, premature.returncode)
self.assertIn("approval required for REQUIREMENT", premature.stderr)
unconfirmed = h.run_workflow(
project, "approve", "--state", state_path, "--stage", "REQUIREMENT", check=False
)
self.assertNotEqual(0, unconfirmed.returncode)
self.assertIn("requires explicit user confirmation", unconfirmed.stderr)
resume = json.loads(h.run_workflow(project, "resume", "--state", state_path).stdout)
self.assertEqual("approve", resume["action"])
self.assertTrue(resume["requires_user_confirmation"])
self.assertIn("--user-confirmed", resume["command"])
h.fake_agent("cursor", project, "--resume", check_rc=0)
self.assert_delivered(project)
def test_rejected_review_sends_work_back_to_implement(self):
project = self.install("claude")
h.fake_agent("claude", project, "--review-reject", "1", check_rc=0)
self.assert_delivered(project)
outcomes = [(e["stage"], e["result"]) for e in events(project, "finish")
if e["stage"] in ("IMPLEMENT", "REVIEW")]
self.assertEqual([("IMPLEMENT", "completed"), ("REVIEW", "failed"),
("IMPLEMENT", "completed"), ("REVIEW", "completed")], outcomes)
def test_unfilled_artifact_is_rejected_and_workflow_does_not_advance(self):
project = self.install("claude")
result = h.fake_agent("claude", project, "--leave-placeholders", "DESIGN", check_rc=2)
self.assertIn("artifact", result.stderr)
state = h.workflow_state(project)
self.assertEqual("in_progress", state["stages"]["DESIGN"]["status"])
self.assertEqual("pending", state["stages"]["IMPLEMENT"]["status"])
def test_repeated_stage_failures_block_the_workflow(self):
project = self.install("pi")
h.fake_agent("pi", project, "--fail-stage", "TEST:2", check_rc=4)
state = h.workflow_state(project)
self.assertEqual("blocked", state["status"])
self.assertEqual("failed", state["stages"]["TEST"]["status"])
self.assertEqual(2, state["stages"]["TEST"]["retry_count"])
if __name__ == "__main__":
unittest.main()