richfrem/agent-plugins-skills / plugins/agent-agentic-os/.apm/skills/running-control-plane
running-control-plane
FlaggedCRITICAL INTAKE GATEWAY: use at the start of any non-trivial engineering task, feature, refactor, multi-file bugfix, work package or new idea, including requests to explore, brainstorm, build, fix or change something, before planning or code changes. Enforces read-only discovery, 1-3 structured questions with defaults, control-plane registration and the immutable four-pillar TASK_SPEC.md. Also runs the rest of the control plane: health ("run health check", "check os metrics", "system monitor", event bus, locks, memory, "is my signing identity ready" status check), transition simulation ("dry run transition", "simulate transition", "check edge X -> Y"), worktree selection, and gating on/off ("disable the control plane", "turn off the push guard", "stop gating my commits"). Not for project setup or key creation (use setting-up-agentic-os) or improving skills (use improving-skills-in-lab).
tests/test_gate3_code_acceptance.py
View on GitHub"""
tests/test_gate3_code_acceptance.py
===================================
Purpose:
Failing-first acceptance tests for the hardening of Gate 3 (auth-ciba-increment-b, human decision 2026-09-20):
BOTH edges into VERIFY_EXIT -- WORKTREE_REVIEW -> VERIFY_EXIT (direct code acceptance) and
MULTI_AGENT_CODE_REVIEW -> VERIFY_EXIT (post-review code acceptance) -- are the moment human authority
accepts the agent's code into the release pipeline, so they need the same cryptographic proof as Gate 1: the
coordinator halts with HUMAN_PROOF_REQUIRED and a request_id, the challenge binds the task, the edge, the
commit SHA, the diff hash and the untracked-file hash (a plain `git diff` would miss new files), and only a
verified `ssh-keygen -Y sign` signature consumed in the commit transaction advances the state. The soft
`--skip-review` / `--skip-reason` flags are gone. Real git repository, real SQLite, real ssh-keygen; the
identity is injected, nothing is mocked.
Key Input Dependencies:
- control_plane/proof_edges.py, snapshot.py (code_acceptance_snapshot), ssh_signing.py (EDGE_ASSERTIONS),
coordinator.py (HumanProofRequired), adapters.py, gate1_approval.py, transition_templates.yaml, agent_control.py
- tests/helpers/gate1_fixtures.py, tests/test_approve_transition.py (key helpers)
Key Functions (test cases):
- test_both_edges_halt_for_a_signature_and_change_nothing
- test_challenge_binds_head_diff_and_untracked
- test_a_valid_signature_commits_verify_exit (both edges)
- test_a_changed_tracked_file_or_new_untracked_file_refuses_and_audits
- test_a_forged_signature_refuses_and_audits
- test_staged_human_decisions_are_refused_without_a_proof
- test_skip_flags_are_gone_and_skip_is_refused_on_proof_edges
- test_registry_has_no_skip_semantics_on_the_gate3_edges
"""
import io
import os
import pwd
import shutil
import sqlite3
import subprocess
import sys
from pathlib import Path
import pytest
_scripts_dir = str(Path(__file__).resolve().parent.parent / "scripts")
if _scripts_dir not in sys.path:
sys.path.insert(0, _scripts_dir)
import agent_control
from helpers.human_signer import OPERATOR_PRINCIPAL, get_test_human
from control_plane.coordinator import HumanProofRequired, TransitionCoordinator, TransitionCoordinatorError
from control_plane.gate1_approval import GateApprovalError, approve_transition, show_challenge
from control_plane.identity_layout import default_layout
from control_plane.ports import PersistenceInvariantViolation, TransitionCommitRequest, TransitionDecision
from control_plane.registry import TransitionRegistry
from control_plane.ssh_signing import SIGN_NAMESPACE
from helpers.gate1_fixtures import init_git_worktree, reach_multi_agent_code_review, reach_worktree_review
TASK = "gate3-task-001"
DIRECT = ("WORKTREE_REVIEW", "VERIFY_EXIT")
POST_REVIEW = ("MULTI_AGENT_CODE_REVIEW", "VERIFY_EXIT")
_FOUND = shutil.which("ssh-keygen")
SSH_KEYGEN = _FOUND or "ssh-keygen"
pytestmark = [pytest.mark.no_auto_signer, pytest.mark.skipif(_FOUND is None or shutil.which("git") is None, reason="ssh-keygen and git are required")]
def _other_uid() -> int:
for name in ("nobody", "daemon"):
try:
uid = pwd.getpwnam(name).pw_uid
if uid != os.geteuid():
return uid
except KeyError:
continue
return os.geteuid() + 4242
class CodeGate:
"""A task sitting at WORKTREE_REVIEW (or MULTI_AGENT_CODE_REVIEW) with a real git worktree and a real key."""
def __init__(self, tmp_path: Path, edge):
self.tmp = tmp_path.resolve()
self.edge = edge
self.worktree = init_git_worktree(self.tmp / "worktree")
self.sim = reach_worktree_review(self.tmp, TASK, self.worktree)
if edge == POST_REVIEW:
reach_multi_agent_code_review(self.sim, TASK)
self.cp = self.sim.control_plane
self.repo = Path(self.cp.repo_root)
(self.worktree / "module.py").write_text("VALUE = 2\n") # the agent's uncommitted work
(self.worktree / "new_file.py").write_text("NEW = 1\n") # and a brand-new untracked file
self.layout = default_layout(self.repo)
for d in (self.layout.root, self.layout.challenge_dir):
d.mkdir(parents=True, exist_ok=True)
os.chmod(d, 0o700)
# One approver per pipeline: this regular task's Gate 1 was signed with the stand-in human's operator key,
# so Gate 3 must be signed with that same key (key continuity, not just a matching principal name).
human = get_test_human()
human.ensure_identity(self.repo)
self.key = human.operator_key
self.principal = OPERATOR_PRINCIPAL
self.identity = {"agent_name": "agentic-os-local-agent", "agent_uid": _other_uid(), "agent_gids": set()}
@staticmethod
def keygen(path: Path) -> None:
subprocess.run([SSH_KEYGEN, "-q", "-t", "ed25519", "-N", "", "-C", "op", "-f", str(path)], check=True, capture_output=True)
def coordinator(self):
return TransitionCoordinator(
self.cp, registry=self.sim.registry, output_stream=io.StringIO(),
)
def request(self) -> int:
with pytest.raises(HumanProofRequired) as excinfo:
self.coordinator().coordinate_transition(task_id=TASK, to_state="VERIFY_EXIT", actor="agent", reason="request", interactive=False)
self.request_id = excinfo.value.remediation["request_id"]
return self.request_id
def show(self):
out = io.StringIO()
path = show_challenge(self.cp, self.request_id, layout=self.layout, key_hint=str(self.key), agent_identity=self.identity, out=out)
return path, out.getvalue()
def sign(self, path: Path, key: Path = None) -> None:
sig = Path(str(path) + ".sig")
if sig.exists():
sig.unlink()
subprocess.run([SSH_KEYGEN, "-Y", "sign", "-f", str(key or self.key), "-n", SIGN_NAMESPACE, str(path)], check=True, capture_output=True)
def approve(self):
return approve_transition(self.cp, self.request_id, layout=self.layout, principal=self.principal, agent_identity=self.identity, out=io.StringIO())
def db(self):
return sqlite3.connect(self.cp.db_path)
def state(self):
return self.db().execute("SELECT state FROM tasks WHERE task_id = ?", (TASK,)).fetchone()[0]
def violations(self):
return self.db().execute("SELECT kind, detail FROM transition_violations WHERE task_id = ?", (TASK,)).fetchall()
@pytest.fixture(params=[DIRECT, POST_REVIEW], ids=["direct", "post-review"])
def gate(request, tmp_path):
return CodeGate(tmp_path, request.param)
def test_both_edges_halt_for_a_signature_and_change_nothing(gate):
with pytest.raises(HumanProofRequired) as excinfo:
gate.coordinator().coordinate_transition(task_id=TASK, to_state="VERIFY_EXIT", actor="agent", reason="try", interactive=False)
assert excinfo.value.remediation["request_id"]
assert "HUMAN_PROOF_REQUIRED" in str(excinfo.value)
assert gate.state() == gate.edge[0]
assert gate.db().execute("SELECT COUNT(*) FROM transition_decisions WHERE task_id = ? AND to_state = 'VERIFY_EXIT'", (TASK,)).fetchone()[0] == 0
def test_challenge_binds_head_diff_and_untracked(gate):
gate.request()
path, printed = gate.show()
head = subprocess.run(["git", "-C", str(gate.worktree), "rev-parse", "HEAD"], capture_output=True, text=True).stdout.strip()
assert head in printed and "tracked_diff:" in printed and "untracked:" in printed
assert f"transition: {gate.edge[0]} -> VERIFY_EXIT" in printed and f"task: {TASK}" in printed
assert "nonce:" in printed and "expires:" in printed
def test_a_valid_signature_commits_verify_exit(gate):
gate.request()
path, _ = gate.show()
gate.sign(path)
record = gate.approve()
assert record.to_state == "VERIFY_EXIT" and gate.state() == "VERIFY_EXIT"
assert gate.db().execute("SELECT status FROM transition_request WHERE request_id = ?", (gate.request_id,)).fetchone()[0] == "CONSUMED"
# no decision rows, no actor strings: the consumed, signature-verified request is the authority
assert gate.db().execute("SELECT COUNT(*) FROM transition_decisions WHERE task_id = ? AND to_state = 'VERIFY_EXIT'", (TASK,)).fetchone()[0] == 0
@pytest.mark.parametrize("tamper", ["tracked", "untracked-new", "untracked-edit"])
def test_a_changed_tracked_file_or_new_untracked_file_refuses_and_audits(gate, tamper):
gate.request()
path, _ = gate.show()
gate.sign(path)
if tamper == "tracked":
(gate.worktree / "module.py").write_text("VALUE = 3\n")
elif tamper == "untracked-new":
(gate.worktree / "sneaky.py").write_text("import os\n")
else:
(gate.worktree / "new_file.py").write_text("NEW = 999\n")
with pytest.raises(GateApprovalError):
gate.approve()
assert gate.state() == gate.edge[0]
assert gate.violations(), "the refused attempt must leave an audit row"
assert {v[0] for v in gate.violations()} == {"REJECTED_ATTEMPT"}
def test_a_forged_signature_refuses_and_audits(gate, tmp_path):
attacker = tmp_path / "attacker"
attacker.mkdir()
CodeGate.keygen(attacker / "id")
gate.request()
path, _ = gate.show()
gate.sign(path, key=attacker / "id")
with pytest.raises(GateApprovalError):
gate.approve()
assert gate.state() == gate.edge[0] and gate.violations()
def test_staged_human_decisions_are_refused_without_a_proof(gate):
"""No soft `actor='human'` decision rows: the commit path itself refuses them on a proof edge."""
conn = gate.db()
occupancy = conn.execute("SELECT MAX(transition_id) FROM task_transitions WHERE task_id = ?", (TASK,)).fetchone()[0]
template = gate.sim.registry.get_template(*gate.edge[:1], "VERIFY_EXIT")
decisions = [
TransitionDecision(
task_id=TASK, source_occupancy_transition_id=occupancy, from_state=gate.edge[0], to_state="VERIFY_EXIT",
question_id=q["question_id"], answer=q["accepted_answers"][0], decision_type="ANSWER", actor="human", recorded_at=1.0,
)
for q in template.human_questions
]
request = TransitionCommitRequest(
task_id=TASK, expected_from_state=gate.edge[0], to_state="VERIFY_EXIT", source_occupancy_transition_id=occupancy,
template_id=template.transition_id, actor="human", reason="soft", staged_decisions=decisions, staged_receipts=[],
)
with pytest.raises(PersistenceInvariantViolation):
gate.cp._persistence.apply_transition_with_receipts(request)
assert gate.state() == gate.edge[0]
def test_skip_flags_are_gone_and_skip_is_refused_on_proof_edges(gate):
parser = agent_control._build_parser()
for verb in ("coordinate-transition", "transition"):
with pytest.raises(SystemExit):
parser.parse_args([verb, "--task-id", TASK, "--to", "VERIFY_EXIT", "--skip-review", "--skip-reason", "x", "--human-confirmed", "HUMAN-CONFIRMED: x"])
with pytest.raises((TransitionCoordinatorError, HumanProofRequired)):
gate.coordinator().coordinate_transition(
task_id=TASK, to_state="VERIFY_EXIT", actor="human", reason="skip", interactive=True, skip_review=True, skip_reason="x",
)
assert gate.state() == gate.edge[0]
def test_registry_has_no_skip_semantics_on_the_gate3_edges():
registry = TransitionRegistry.load_default()
for edge in (DIRECT, POST_REVIEW):
template = registry.get_template(*edge)
assert template.requires_cryptographic_proof is True
assert template.skip.get("allowed") is False
assert "code_review_or_skip" not in template.deterministic_checks
assert template.human_questions == []
assert "skip" not in template.next_steps_hint.lower(), edge
def _write_lock_is_free(db_path) -> bool:
"""True if another connection can take the write lock right now (0.3s budget)."""
probe = sqlite3.connect(db_path, timeout=0.3, isolation_level=None)
try:
probe.execute("BEGIN IMMEDIATE;")
probe.execute("ROLLBACK;")
return True
except sqlite3.OperationalError:
return False
finally:
probe.close()
def test_no_write_lock_is_held_during_snapshot_or_signature_verification(gate, monkeypatch):
"""The commit's BEGIN IMMEDIATE must not span external work (git snapshot hashing, the
`ssh-keygen -Y verify` subprocess): both run BEFORE the write lock is taken, or a concurrent
writer gets `database is locked` for the full duration of the subprocess."""
from control_plane import gate1_approval, transition_request
observed = {"snapshot": [], "verify": []}
real_snapshot, real_verify = gate1_approval.snapshot_for_edge, transition_request.verify_signature
def snapshot_probe(*args, **kwargs):
observed["snapshot"].append(_write_lock_is_free(gate.cp.db_path))
return real_snapshot(*args, **kwargs)
def verify_probe(*args, **kwargs):
observed["verify"].append(_write_lock_is_free(gate.cp.db_path))
return real_verify(*args, **kwargs)
monkeypatch.setattr(gate1_approval, "snapshot_for_edge", snapshot_probe)
monkeypatch.setattr(transition_request, "verify_signature", verify_probe)
gate.request()
path, _ = gate.show()
gate.sign(path)
observed["snapshot"].clear() # show/challenge-time snapshots are not under test
gate.approve()
assert gate.state() == "VERIFY_EXIT"
assert observed["verify"] and all(observed["verify"]), f"write lock held during ssh-keygen verify: {observed['verify']}"
assert observed["snapshot"] and all(observed["snapshot"]), f"write lock held during live snapshot: {observed['snapshot']}"
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Unverified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |