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_control_plane_trigger_enforcement.py
View on GitHub"""
test_control_plane_trigger_enforcement.py — SQLite trigger-based transition enforcement (issue-523)
=====================================================================================================
Purpose:
Proves that context/control_plane.db enforces legal state transitions at the SQLite engine
level, independent of the Python application layer — closing the raw-sqlite3-bypass gap
described in GitHub Issue #523. Exercises the AFTER UPDATE and AFTER INSERT triggers, the
valid_transitions registry-sync table, and the transition_violations audit log directly via
raw sqlite3 connections (not through ControlPlane/SqlitePersistenceAdapter), since the whole
point of this feature is protecting against callers that skip the Python layer entirely.
The UPDATE trigger reverts (coerces state+updated_at back to OLD.*, never destroys). The
INSERT trigger deletes the illegally-inserted row instead — a deliberate asymmetry (see
docs/plans/issue-523-trigger-enforcement-spec.md guardrail table): a fresh illegal INSERT
has no accumulated transitions/receipts/decisions to protect, so delete loses nothing a
legitimate caller would have accrued, and it avoids a cross-trigger cascade where a
corrective UPDATE would itself fire enforce_valid_transition (e.g. DONE -> INTAKE isn't a
legal edge) and revert the fix right back.
Key Input Dependencies:
- Temporary SQLite databases via pytest's tmp_path fixture
- control_plane.registry.TransitionRegistry.load_default() (for the parity assertion)
Key Functions:
- test_valid_transitions_table_matches_registry()
- test_illegal_update_transition_via_raw_sql_is_reverted_and_logged()
- test_legal_transition_via_raw_sql_is_unaffected_and_unlogged()
- test_illegal_insert_initial_state_via_raw_sql_is_deleted_and_logged()
- test_delete_then_reinsert_bypass_is_caught_by_insert_trigger()
- test_insert_trigger_cascade_does_not_fire_update_trigger()
- test_trigger_survives_schema_rebuild()
- test_recursive_triggers_on_hits_recursion_limit_but_no_illegal_write_persists()
"""
import sqlite3
import sys
from pathlib import Path
import pytest
SCRIPTS_DIR = Path(__file__).resolve().parent.parent / "scripts"
if str(SCRIPTS_DIR) not in sys.path:
sys.path.insert(0, str(SCRIPTS_DIR))
from control_plane.adapters import SqlitePersistenceAdapter, FilesystemAdapter, SCHEMA_SQL, _split_schema_sql_statements
from control_plane.registry import TransitionRegistry
from control_plane.constants import (
STATE_INTAKE, STATE_INTERVIEW, STATE_IN_WORKTREE, STATE_WORKTREE_REVIEW, STATE_DONE,
)
# INTAKE -> IN_WORKTREE is not a legal edge in ALLOWED_TRANSITIONS (unlike INTAKE -> DONE,
# which IS a legal short-circuit edge) — this is the illegal-UPDATE example used throughout
# issue-523's design discussion. Using DONE here would silently test nothing.
ILLEGAL_UPDATE_TARGET = STATE_IN_WORKTREE
def _make_adapter(tmp_path):
db_path = tmp_path / "control_plane.db"
adapter = SqlitePersistenceAdapter(db_path, FilesystemAdapter())
adapter.ensure_schema()
return adapter, db_path
def _insert_legal_task(conn, task_id, state):
"""Inserts a task row bypassing Python (raw SQL), at a state legal as an initial state."""
conn.execute(
"INSERT INTO tasks (task_id, title, state, runtime_tool) VALUES (?, ?, ?, ?)",
(task_id, "Test Task", state, "claude"),
)
conn.commit()
def test_valid_transitions_table_matches_registry(tmp_path):
"""valid_transitions (as materialized in a live DB) matches TransitionRegistry.get_all_edges()
exactly — the automated drift-detection test requested during os-control-intake."""
adapter, db_path = _make_adapter(tmp_path)
conn = sqlite3.connect(str(db_path))
try:
db_edges = {
(row[0], row[1])
for row in conn.execute("SELECT from_state, to_state FROM valid_transitions WHERE from_state IS NOT NULL")
}
finally:
conn.close()
registry_edges = TransitionRegistry.load_default().get_all_edges()
assert db_edges == registry_edges
def test_illegal_update_transition_via_raw_sql_is_reverted_and_logged(tmp_path):
"""A raw sqlite3 UPDATE attempting an illegal (OLD.state, NEW.state) transition is reverted
(both state and updated_at restored — finding #7) and logged to transition_violations."""
adapter, db_path = _make_adapter(tmp_path)
conn = sqlite3.connect(str(db_path))
try:
_insert_legal_task(conn, "task-1", STATE_INTAKE)
before = conn.execute("SELECT state, updated_at FROM tasks WHERE task_id='task-1'").fetchone()
conn.execute(f"UPDATE tasks SET state='{ILLEGAL_UPDATE_TARGET}' WHERE task_id='task-1'")
conn.commit()
after = conn.execute("SELECT state, updated_at FROM tasks WHERE task_id='task-1'").fetchone()
assert after[0] == STATE_INTAKE
assert after[1] == before[1]
violations = conn.execute(
"SELECT task_id, attempted_from_state, attempted_to_state FROM transition_violations WHERE task_id='task-1'"
).fetchall()
assert violations == [("task-1", STATE_INTAKE, ILLEGAL_UPDATE_TARGET)]
finally:
conn.close()
def test_legal_transition_via_raw_sql_is_unaffected_and_unlogged(tmp_path):
"""A raw sqlite3 UPDATE performing a legal transition is unaffected by the trigger and
produces no transition_violations row."""
adapter, db_path = _make_adapter(tmp_path)
conn = sqlite3.connect(str(db_path))
try:
_insert_legal_task(conn, "task-2", STATE_INTAKE)
conn.execute("UPDATE tasks SET state='INTERVIEW' WHERE task_id='task-2'")
conn.commit()
after = conn.execute("SELECT state FROM tasks WHERE task_id='task-2'").fetchone()
assert after[0] == STATE_INTERVIEW
count = conn.execute(
"SELECT COUNT(*) FROM transition_violations WHERE task_id='task-2'"
).fetchone()[0]
assert count == 0
finally:
conn.close()
def test_illegal_insert_initial_state_via_raw_sql_is_deleted_and_logged(tmp_path):
"""A raw sqlite3 INSERT creating a new task already in an illegal initial state (e.g. DONE)
is deleted (revised: not coerced — see module docstring) and logged to
transition_violations with attempted_from_state=NULL (finding #5)."""
adapter, db_path = _make_adapter(tmp_path)
conn = sqlite3.connect(str(db_path))
try:
conn.execute(
"INSERT INTO tasks (task_id, title, state, runtime_tool) VALUES (?, ?, ?, ?)",
("task-3", "Test Task", STATE_DONE, "claude"),
)
conn.commit()
after = conn.execute("SELECT * FROM tasks WHERE task_id='task-3'").fetchone()
assert after is None
violations = conn.execute(
"SELECT task_id, attempted_from_state, attempted_to_state FROM transition_violations WHERE task_id='task-3'"
).fetchall()
assert violations == [("task-3", None, STATE_DONE)]
finally:
conn.close()
def test_delete_then_reinsert_bypass_is_caught_by_insert_trigger(tmp_path):
"""DELETE + re-INSERT (an UPDATE-trigger bypass, finding #5) is caught by the INSERT-side
trigger on the re-INSERT half, since the re-INSERT is itself checked against legal initial
states — the re-inserted row is deleted, leaving the task nonexistent."""
adapter, db_path = _make_adapter(tmp_path)
conn = sqlite3.connect(str(db_path))
try:
_insert_legal_task(conn, "task-4", STATE_INTAKE)
conn.execute("DELETE FROM tasks WHERE task_id='task-4'")
conn.execute(
"INSERT INTO tasks (task_id, title, state, runtime_tool) VALUES (?, ?, ?, ?)",
("task-4", "Test Task", STATE_DONE, "claude"),
)
conn.commit()
after = conn.execute("SELECT * FROM tasks WHERE task_id='task-4'").fetchone()
assert after is None
violations = conn.execute(
"SELECT task_id, attempted_from_state, attempted_to_state FROM transition_violations WHERE task_id='task-4'"
).fetchall()
assert violations == [("task-4", None, STATE_DONE)]
finally:
conn.close()
def test_insert_trigger_cascade_does_not_fire_update_trigger(tmp_path):
"""Regression test for the cross-trigger cascade bug found during implementation: the
INSERT trigger's original coerce-to-INTAKE corrective UPDATE fired enforce_valid_transition
as a side effect (since e.g. DONE -> INTAKE isn't a legal edge), which reverted the
coercion right back and logged a spurious second violation. Deleting instead of coercing
fixes this — proves exactly one violation row is logged, not two, and the row is gone."""
adapter, db_path = _make_adapter(tmp_path)
conn = sqlite3.connect(str(db_path))
try:
conn.execute(
"INSERT INTO tasks (task_id, title, state, runtime_tool) VALUES (?, ?, ?, ?)",
("task-7", "Test Task", STATE_DONE, "claude"),
)
conn.commit()
count = conn.execute(
"SELECT COUNT(*) FROM transition_violations WHERE task_id='task-7'"
).fetchone()[0]
assert count == 1
after = conn.execute("SELECT * FROM tasks WHERE task_id='task-7'").fetchone()
assert after is None
finally:
conn.close()
def test_trigger_survives_schema_rebuild(tmp_path):
"""The enforcement triggers are re-declared as part of SCHEMA_SQL, so they survive
_rebuild_schema_transactional()'s rename->recreate->copy path, not just a fresh
ensure_schema() call."""
adapter, db_path = _make_adapter(tmp_path)
conn = adapter.get_connection()
try:
adapter._rebuild_schema_transactional(conn)
finally:
conn.close()
conn = sqlite3.connect(str(db_path))
try:
_insert_legal_task(conn, "task-5", STATE_INTAKE)
conn.execute(f"UPDATE tasks SET state='{ILLEGAL_UPDATE_TARGET}' WHERE task_id='task-5'")
conn.commit()
after = conn.execute("SELECT state FROM tasks WHERE task_id='task-5'").fetchone()
assert after[0] == STATE_INTAKE
finally:
conn.close()
def test_recursive_triggers_on_hits_recursion_limit_but_no_illegal_write_persists(tmp_path):
"""With PRAGMA recursive_triggers = ON explicitly set (finding #2), the corrective UPDATE
inside the UPDATE trigger's body re-fires the same trigger (the correction itself looks
like a fresh, non-adjacent transition), oscillating until SQLite's built-in recursion-depth
cap aborts the entire statement — rolling back the illegal write AND the violation-log
INSERT together, the same way RAISE(ABORT) does (see spec guardrail table).
Documented, accepted residual risk (decided 2026-09-07, not a security regression): the
illegal write never persists — this asserts state is unaffected — but the violation goes
unlogged in this non-default pragma mode, and the caller gets an unhandled
sqlite3.OperationalError instead of a clean revert. recursive_triggers is OFF by default in
SQLite; this only matters if something else touching this DB explicitly turns it on.
Uses WORKTREE_REVIEW -> DRAFT_PLAN (not INTAKE-sourced) as the illegal pair: the
control-plane-reset-transition feature added a reset_to_intake edge from every
non-INTAKE state back to INTAKE, so a corrective revert-to-INTAKE is now always
legal and would no longer recurse — WORKTREE_REVIEW/DRAFT_PLAN remain illegal in
both directions, preserving this test's original premise."""
adapter, db_path = _make_adapter(tmp_path)
conn = sqlite3.connect(str(db_path))
try:
conn.execute("PRAGMA recursive_triggers = ON;")
_insert_legal_task(conn, "task-6", STATE_INTAKE)
# Seed at WORKTREE_REVIEW via a trigger-safe raw set (not a legal initial
# state for _insert_legal_task's INSERT path).
conn.execute("DROP TRIGGER IF EXISTS enforce_valid_transition;")
conn.execute("UPDATE tasks SET state = 'WORKTREE_REVIEW' WHERE task_id = 'task-6'")
conn.commit()
for stmt in _split_schema_sql_statements(SCHEMA_SQL):
if "CREATE TRIGGER" in stmt.upper():
conn.execute(stmt)
conn.commit()
with pytest.raises(sqlite3.OperationalError, match="too many levels of trigger recursion"):
conn.execute("UPDATE tasks SET state='DRAFT_PLAN' WHERE task_id='task-6'")
conn.commit()
after = conn.execute("SELECT state FROM tasks WHERE task_id='task-6'").fetchone()
assert after[0] == STATE_WORKTREE_REVIEW
count = conn.execute(
"SELECT COUNT(*) FROM transition_violations WHERE task_id='task-6'"
).fetchone()[0]
assert count == 0
finally:
conn.close()
def test_sync_valid_transitions_is_atomic_not_left_empty_on_failure(tmp_path):
"""Regression test for a code-review finding: the original _sync_valid_transitions()
issued a bare DELETE followed by a bare executemany(INSERT) with no transaction wrapper,
so a failure between the two (or a concurrent reader on the shared multi-worktree DB)
could observe valid_transitions as empty or partially-populated — during which every
legitimate transition would be misclassified as illegal and reverted/deleted. Proves
atomicity by forcing a real constraint violation partway through the INSERT batch (a
NOT NULL violation on to_state, injected via a patched registry) and asserting the table
retains its prior full set afterward, never empty or partial."""
import unittest.mock
adapter, db_path = _make_adapter(tmp_path)
conn = sqlite3.connect(str(db_path))
try:
before_count = conn.execute("SELECT COUNT(*) FROM valid_transitions").fetchone()[0]
assert before_count > 0
from control_plane.registry import TransitionRegistry
real_edges = TransitionRegistry.load_default().get_all_edges_with_proof()
fake_registry = unittest.mock.MagicMock()
# Good rows first, then one NOT NULL-violating row (to_state=None) to force a
# mid-batch failure rather than an immediate one. _sync_valid_transitions reads
# get_all_edges_with_proof() (auth-ciba-increment-b) -- 4-tuples
# (from, to, authorized_actor, requires_proof) -- not the earlier 2-/3-tuple accessors.
fake_registry.get_all_edges_with_proof.return_value = real_edges + [(STATE_INTAKE, None, "agent_or_human", 0)]
with unittest.mock.patch.object(TransitionRegistry, "load_default", return_value=fake_registry):
with pytest.raises(sqlite3.IntegrityError):
adapter._sync_valid_transitions(conn)
after_count = conn.execute("SELECT COUNT(*) FROM valid_transitions").fetchone()[0]
assert after_count == before_count
finally:
conn.close()
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 | Verified | 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 |