MongLong0214/frontier-simplify / skills/frontier-simplify-review
frontier-simplify-review
UnverifiedReview a change for concrete defects, check them with a refuter and a reproducer, then check repairs and their regressions against the preserved review. Use for substantial PR reviews or repeated review rounds. Stops automatic review after at most three attempts per PR and names the calling agent's next action; does not guarantee safe merge convergence.
scripts/selftest-scope-runner.py
View on GitHub#!/usr/bin/env python3
"""Actual review wrapper installs exact, immutable phase-specific scope input."""
import json
import os
from pathlib import Path
import shutil
import subprocess
import sys
import tempfile
SCRIPTS = Path(__file__).resolve().parent
sys.path.insert(0, str(SCRIPTS / 'lib'))
import ledger
from protocol import Rejected, digest
def git(repo, *args):
return subprocess.check_output(['git', '-C', str(repo), *args], stderr=subprocess.PIPE).decode().strip()
def check(label, good):
print(('ok ' if good else 'NOT OK ') + label)
return good
with tempfile.TemporaryDirectory(prefix='review-scope-runner-') as temporary:
t = Path(temporary)
repo = t / 'repo'
repo.mkdir()
git(repo, 'init', '-q')
git(repo, 'config', 'user.name', 'test')
git(repo, 'config', 'user.email', 'test@example.invalid')
(repo / 'foo.py').write_text('before\n')
(repo / 'foo.test.py').write_text('before test\n')
git(repo, 'add', '.')
git(repo, 'commit', '-qm', 'base')
base = git(repo, 'rev-parse', 'HEAD')
(repo / 'foo.py').write_text('reviewed\n')
(repo / 'foo.test.py').write_text('reviewed test\n')
git(repo, 'commit', '-qam', 'reviewed')
head = git(repo, 'rev-parse', 'HEAD')
binpath = t / 'bin'
binpath.mkdir()
fake = binpath / 'codex'
fake.write_text('#!' + sys.executable + '\n' + '''
import json, os, subprocess, sys
from pathlib import Path
scope_bytes = Path('SCOPE.json').read_bytes()
scope = json.loads(scope_bytes)
names = set(subprocess.check_output(['git', 'diff', '--no-renames', '--name-only',
'-z', scope['from_sha'], scope['to_sha']]).decode().strip('\\0').split('\\0'))
assert scope['from_sha'] == os.environ['TEST_FROM']
assert scope['to_sha'] == os.environ['TEST_TO']
assert scope['input_kind'] == os.environ['TEST_KIND']
assert {item['path'] for item in scope['files']} == (names if names != {''} else set())
assert 'DIFF.patch' in os.listdir('.') and 'CHANGED.txt' in os.listdir('.')
if scope['input_kind'] == 'remediation':
assert all(Path(name).exists() for name in ('ROUND1_INVENTORY.md', 'PREVIOUS_REVIEW.md',
'IMPLEMENTER_RESPONSE.md', 'REMEDIATION.patch', 'REMEDIATION_CHANGED.txt',
'REMEDIATION_HUNKS.md'))
if not Path('CANDIDATE_REVIEW.md').exists(): # the finder's call; the validators run after it
Path(os.environ['TEST_LOG']).write_text(json.dumps({'scope': scope, 'scope_bytes': scope_bytes.decode(),
'remediation_patch': Path('REMEDIATION.patch').read_text() if scope['input_kind'] == 'remediation' else None,
'prompt': sys.stdin.buffer.read().decode()}))
if os.environ.get('TEST_TAMPER'):
Path('SCOPE.json').write_text('tampered')
for row in [
{'type':'item.completed','item':{'type':'command_execution','command':'cat SCOPE.json'}},
{'type':'item.completed','item':{'type':'agent_message','text':'Reviewed frozen scope.'}},
{'type':'turn.completed'}]:
print(json.dumps(row))
''')
fake.chmod(0o755)
env = {key: value for key, value in os.environ.items() if not key.startswith('REVIEW_')}
env.update(PATH=str(binpath) + os.pathsep + env['PATH'],
REVIEW_ARTIFACTS=str(t / 'artifacts'), PYTHONDONTWRITEBYTECODE='1')
def host(pr, phase, to_sha, from_sha, kind, response=None, tamper=False):
log = t / (pr + '-' + phase + '.json')
local = dict(env, TEST_FROM=from_sha, TEST_TO=to_sha, TEST_KIND=kind,
TEST_LOG=str(log))
if response:
local['REVIEW_RESPONSE'] = str(response)
if tamper:
local['TEST_TAMPER'] = '1'
result = subprocess.run([str(SCRIPTS / 'review-round.sh'), phase, str(repo), to_sha,
pr, base, 'codex'], env=local, capture_output=True, text=True)
return result, log
def root(pr):
return next((t / 'artifacts').glob('*/' + pr))
results = []
first, log = host('phase-one', '1', head, base, 'change')
results.append(check('phase1-fake-executor-read-exact-frozen-scope',
first.returncode == 10 and log.exists() and
set(json.loads(log.read_text())['scope']['files'][n]['path'] for n in (0, 1)) ==
{'foo.py', 'foo.test.py'}))
if first.returncode == 10:
starts, ends, _ = ledger.audit(root('phase-one'), repo)
data = json.loads(log.read_text())
results.append(check('phase1-started-prepared-copy-and-prompt-agree',
starts[1].get('scope_schema_version') == 1 and
ledger.hashes(root('phase-one') / 'round-0001', ['SCOPE.json'])['SCOPE.json'] ==
starts[1]['inputs']['SCOPE.json'] and ends[1]['recorded'] and
ends[1]['checkout_observations']['input_copies']['status'] == 'unchanged' and
'SCOPE.json' in data['prompt']))
legacy = dict(starts[1], inputs=dict(starts[1]['inputs']))
legacy.pop('scope_schema_version')
legacy['inputs'].pop('SCOPE.json')
old_checks = ledger.recompute(repo, root('phase-one') / 'round-0001', legacy)
missing = dict(starts[1], inputs=dict(starts[1]['inputs']))
missing['inputs'].pop('SCOPE.json')
missing_checks = ledger.recompute(repo, root('phase-one') / 'round-0001', missing)
results.append(check('undeclared-legacy-is-readable-but-declared-missing-is-not',
next(c for c in old_checks if c['guard'] == 'input-digests')['ok'] and
not next(c for c in missing_checks if c['guard'] == 'input-digests')['ok']))
def copied_history(name, transform):
copy = t / name
shutil.copytree(root('phase-one'), copy)
rows = ledger.read(copy)
transform(rows)
(copy / 'ledger.jsonl').unlink()
for row in rows:
ledger.append(copy, row)
return copy
historic = copied_history('initial-v3-no-declaration', lambda rows: (
rows[0].pop('scope_schema_version'), rows[1]['inputs'].pop('SCOPE.json')))
results.append(check('initial-v3-without-declaration-remains-readable',
bool(ledger.audit(historic, repo)[2])))
absent = copied_history('declared-prepared-missing', lambda rows: (
rows[1]['inputs'].pop('SCOPE.json'), rows.pop()))
try:
ledger.audit(absent, repo)
except Rejected as error:
absent_refused = 'scope input digest' in str(error)
else:
absent_refused = False
results.append(check('declared-prepared-missing-fails-even-before-execution', absent_refused))
tampered = t / 'frozen-scope-tamper'
shutil.copytree(root('phase-one'), tampered)
(tampered / 'round-0001/SCOPE.json').write_text('{}')
try:
ledger.audit(tampered, repo)
except Rejected:
tamper_refused = True
else:
tamper_refused = False
results.append(check('frozen-scope-hash-tamper-fails-readback', tamper_refused))
malformed = t / 'scope-json-array'
shutil.copytree(root('phase-one') / 'round-0001', malformed)
(malformed / 'SCOPE.json').write_bytes(b'[]')
malformed_start = dict(starts[1], inputs=dict(starts[1]['inputs']))
malformed_start['inputs']['SCOPE.json'] = digest(b'[]')
malformed_check = next(c for c in ledger.recompute(repo, malformed, malformed_start)
if c['guard'] == 'input-digests')
results.append(check('rehashed-nonobject-scope-refused-without-reader-crash',
not malformed_check['ok'] and '[scope-target]' in malformed_check['reason']))
for version in (0, 2):
try:
ledger.scope_version(dict(starts[1], scope_schema_version=version))
except Rejected:
refused = True
else:
refused = False
results.append(check(f'unsupported-declared-version-{version}-refused', refused))
(repo / 'foo.py').write_text('repaired\n')
git(repo, 'commit', '-qam', 'repair')
repaired = git(repo, 'rev-parse', 'HEAD')
response = t / 'response.txt'
response.write_text('Finding addressed with changed implementation.\n')
followup, followup_log = host('phase-one', '2', repaired, head, 'remediation', response)
followup_data = json.loads(followup_log.read_text()) if followup_log.exists() else {}
results.append(check('phase2-reads-original-and-whole-remediation-scope',
followup.returncode == 10 and
followup_data.get('scope', {}).get('from_sha') == head and
followup_data.get('scope', {}).get('to_sha') == repaired and
'repaired' in followup_data.get('remediation_patch', '') and
'SCOPE.json' in followup_data.get('prompt', '') and
(root('phase-one') / 'round-0002/PREVIOUS_REVIEW.md').read_bytes() ==
(root('phase-one') / 'round-0001/ARTIFACT.md').read_bytes() and
(root('phase-one') / 'round-0002/IMPLEMENTER_RESPONSE.md').read_bytes() ==
response.read_bytes()))
phase_starts, _, _ = ledger.audit(root('phase-one'), repo)
results.append(check('phase2-reserves-exact-original-and-previous-round',
phase_starts[2].get('original_round') == 1 and
phase_starts[2].get('previous_round') == 1))
wrong_original = copied_history('wrong-original-round',
lambda rows: rows[3].update(original_round=99))
try:
ledger.audit(wrong_original, repo)
except Rejected as error:
wrong_refused = 'original' in str(error)
else:
wrong_refused = False
results.append(check('rehashed-original-round-substitution-refused', wrong_refused))
old_phase2 = copied_history('older-phase2-without-original-round',
lambda rows: rows[3].pop('original_round', None))
results.append(check('older-phase2-binding-remains-readable',
bool(ledger.audit(old_phase2, repo)[2])))
later_response = t / 'later-response.txt'
later_response.write_text('Later evidence for the same repair.\n')
third, _ = host('phase-one', '2', repaired, head, 'remediation', later_response)
complete_starts, _, _ = ledger.audit(root('phase-one'), repo)
results.append(check('third-attempt-binds-latest-usable-review',
third.returncode == 11 and complete_starts[3].get('previous_round') == 2))
rebound = t / 'rebound-previous-review'
shutil.copytree(root('phase-one'), rebound)
stale_previous = (rebound / 'round-0001/ARTIFACT.md').read_bytes()
(rebound / 'round-0003/PREVIOUS_REVIEW.md').write_bytes(stale_previous)
rows = ledger.read(rebound)
rows[6]['previous_round'] = 1
rows[7]['inputs']['PREVIOUS_REVIEW.md'] = digest(stale_previous)
(rebound / 'ledger.jsonl').unlink()
for row in rows:
ledger.append(rebound, row)
try:
ledger.audit(rebound, repo)
except Rejected as error:
rebound_refused = 'previous' in str(error)
else:
rebound_refused = False
results.append(check('rehashed-older-previous-review-substitution-refused', rebound_refused))
response_start, response_log = host('response-only', '1', head, base, 'change')
response_followup, response_followup_log = host('response-only', 'auto', head, head,
'remediation', response)
response_data = json.loads(response_followup_log.read_text()) if response_followup_log.exists() else {}
response_rows = (root('response-only') / 'ledger.jsonl').read_bytes()
if response_followup_log.exists():
response_followup_log.unlink()
response_cache, _ = host('response-only', 'auto', head, head, 'remediation', response)
results.append(check('response-only-empty-remediation-keeps-review-and-cache',
response_start.returncode == 10 and response_followup.returncode == 10
and response_data.get('scope', {}).get('files') == []
and response_data.get('remediation_patch') == ''
and response_cache.returncode == 10
and (root('response-only') / 'ledger.jsonl').read_bytes() == response_rows
and not response_followup_log.exists()))
empty, empty_log = host('empty-target', '1', base, base, 'change')
results.append(check('empty-phase1-refused-before-executor',
empty.returncode == 5 and not empty_log.exists() and
not (root('empty-target') / 'ledger.jsonl').exists()))
tamper, _ = host('copied-tamper', '1', head, base, 'change', tamper=True)
tamper_end = ledger.read(root('copied-tamper'))[-1]
results.append(check('copied-scope-tamper-is-observed-and-not-recorded',
tamper.returncode == 5 and not tamper_end.get('recorded') and
tamper_end.get('checkout_observations', {}).get('input_copies', {}).get('status') ==
'changed'))
sys.exit(0 if all(results) else 1)
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: 470k | 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 |