akiiwa-games/AgentCompass / skills/agent-compass
agent-compass
FlaggedChoose owner, model and effort, outbound authorization, advisor use, and continuation for delegated work. Reads local usage and effective account policy, profiles the task with the configured System-One scorer, then allocates a compatible worker. Use for independent worker or advisor work; ordinary local execution needs no routing.
scripts/test_launch_values.py
View on GitHub"""Rule (router-internals §5, Launch values): no launch passes a value through cmd.exe. A Windows batch launcher (an npm
`.cmd` shim) is started as the program its launch line runs (launch_values.program), so a value holding cmd.exe syntax
(`&`, `|`, `^`, `%VAR%`, `>`) reaches the program as one argument and runs nothing; a launcher that cannot be read that
way refuses the launch, stated. The fake shims and the recording program live in a temporary directory."""
import argparse
import json
import os
import sys
import tempfile
import unittest
from pathlib import Path
from unittest.mock import patch
import host_herdr
import launch_values
import public_data
import roll
import scorer_client
import subscription_cli
RECORDER = """import json, os, sys
from pathlib import Path
with open(Path(__file__).with_name('argv.jsonl'), 'a', encoding='utf-8') as out:
out.write(json.dumps({'argv': sys.argv[1:], 'shim_env': os.environ.get('SHIM_SET')}) + '\\n')
print('{}')
"""
# The shape npm's cmd-shim writes (the program in `_prog`, the script beside `%dp0%`), with a variable of its own.
NPM_SHIM = """@ECHO off\r
GOTO start\r
:find_dp0\r
SET dp0=%~dp0\r
EXIT /b\r
:start\r
SETLOCAL\r
CALL :find_dp0\r
IF EXIST "%dp0%\\node.exe" (\r
SET "_prog=%dp0%\\node.exe"\r
) ELSE (\r
SET "_prog={python}"\r
SET PATHEXT=%PATHEXT:;.JS;=;%\r
)\r
\r
endLocal & goto #_undefined_# 2>NUL || title %COMSPEC% & "%_prog%" "%dp0%\\record.py" %*\r
"""
DESKTOP_SHIM = """@echo off\r
setlocal\r
set "SCRIPT_DIR=%~dp0"\r
set "SHIM_SET=from-the-launcher"\r
"{python}" "%SCRIPT_DIR%record.py" node-script %*\r
exit /b %errorlevel%\r
"""
# The reviewer's probe shim: it hands %* to cmd.exe itself, so it cannot be read as a program and its arguments.
UNREADABLE_SHIM = '@echo off\r\necho %* > "%~dp0args-%~n0.txt"\r\n'
class BatchLauncherTests(unittest.TestCase):
def setUp(self):
self.folder = Path(tempfile.mkdtemp(prefix='launch-values-'))
self.addCleanup(lambda: __import__('shutil').rmtree(self.folder, ignore_errors=True))
(self.folder / 'record.py').write_text(RECORDER, encoding='utf-8')
self.marker = self.folder / 'INJECTED.txt'
self.hostile = [f'm&echo.pwned>{self.marker}', f'm|echo.pwned>{self.marker}', f'm^&echo.pwned>{self.marker}',
'm%PWN%', f'm>{self.marker}']
env = patch.dict(os.environ, {'PATH': str(self.folder) + os.pathsep + os.environ.get('PATH', ''),
'PWN': 'EXPANDED'})
env.start()
self.addCleanup(env.stop)
def shim(self, name, text):
path = self.folder / f'{name}.CMD'
path.write_bytes(text.replace('{python}', sys.executable).encode('ascii'))
return path
def recorded(self):
path = self.folder / 'argv.jsonl'
return [json.loads(line) for line in path.read_text(encoding='utf-8').splitlines()] if path.exists() else []
def assert_nothing_injected(self):
self.assertFalse(self.marker.exists(), 'a value ran as a command')
self.assertFalse(any(p.name.startswith('pwned') for p in self.folder.iterdir()))
def test_a_shim_is_read_as_its_program_and_script(self):
shim = self.shim('pi', NPM_SHIM)
argv, env = launch_values.program(shim)
self.assertEqual([os.path.normcase(a) for a in argv],
[os.path.normcase(sys.executable), os.path.normcase(str(self.folder / 'record.py'))])
self.assertEqual(env, {}) # dp0 and _prog are consumed; PATHEXT's substring edit is not a value
argv, env = launch_values.program(self.shim('paseo', DESKTOP_SHIM))
self.assertEqual(argv[2], 'node-script')
self.assertEqual(env, {'SHIM_SET': 'from-the-launcher'})
self.assertEqual(launch_values.program(self.folder / 'tool.exe'), ([str(self.folder / 'tool.exe')], {}))
def test_the_cli_worker_passes_every_value_as_one_argument(self):
"""roll.cli for the npm-shim CLIs (pi, opencode): the model and the prompt reach the program unchanged."""
for family in ('pi', 'opencode'):
self.shim(family, NPM_SHIM)
started = []
real = roll.subprocess.Popen
def popen(*args, **kwargs):
started.append(real(*args, **kwargs))
return started[-1]
out = {'parent_id': 'p', 'parent_host': 'paseo', 'parent_address': 'p'}
with patch.object(roll, 'provider_definition', lambda provider, bundle=None: (provider, {})), \
patch.object(roll.host_placement, 'place', lambda *a, **k: {'cwd': str(self.folder)}), \
patch.object(roll, 'CLI_RUNS_DIR', self.folder / 'runs'), \
patch.object(roll.subprocess, 'Popen', popen):
for family in ('pi', 'opencode'):
for value in self.hostile:
result, code = roll.cli(argparse.Namespace(workspace=None), out,
{'provider': family, 'model': value}, f'task {value}')
self.assertEqual((result['launched'], code), (True, 0), result)
started[-1].wait(30)
seen = [line['argv'] for line in self.recorded()]
self.assertEqual(len(seen), 2 * len(self.hostile))
for value in self.hostile:
self.assertTrue(any(any(arg.endswith(value) for arg in argv) for argv in seen), value)
self.assertNotIn('EXPANDED', json.dumps(seen))
self.assert_nothing_injected()
def test_the_scorer_and_cli_worker_command_passes_values_unchanged(self):
"""subscription_cli.cli_prefix (the scorer's command and roll's CLI worker command) on an npm shim."""
self.shim('claude', NPM_SHIM)
prefix = subscription_cli.cli_prefix('claude')
for value in self.hostile:
code, _out, _err = subscription_cli.run_process([*prefix, '--model', value], str(self.folder), prompt='x',
timeout=30)
self.assertEqual(code, 0)
self.assertEqual([line['argv'][-1] for line in self.recorded()], self.hostile)
self.assert_nothing_injected()
def test_herdr_core_runs_as_its_program(self):
self.shim('hc', NPM_SHIM)
for value in self.hostile:
host_herdr.hc('terminal', 'spawn', '--command', value)
self.assertEqual([line['argv'][-1] for line in self.recorded()], self.hostile)
self.assert_nothing_injected()
def test_a_launcher_that_cannot_be_read_refuses_the_launch_and_runs_nothing(self):
for family in ('pi', 'opencode', 'claude'):
self.shim(family, UNREADABLE_SHIM)
with self.assertRaisesRegex(launch_values.UnreadableLauncher, 'goes on after %\\*'):
subscription_cli.cli_prefix('claude')
out = {'parent_id': 'p', 'parent_host': 'paseo', 'parent_address': 'p'}
before = public_data._EFFECTS[0]
with patch.object(roll, 'provider_definition', lambda provider, bundle=None: (provider, {})), \
patch.object(roll.host_placement, 'place', lambda *a, **k: {'cwd': str(self.folder)}), \
patch.object(roll, 'CLI_RUNS_DIR', self.folder / 'runs'), \
patch.object(roll.subprocess, 'Popen', side_effect=AssertionError('started')):
result, code = roll.cli(argparse.Namespace(workspace=None), out, {'provider': 'pi', 'model': self.hostile[0]},
'task')
self.assertEqual((result['launched'], code), (False, 1))
self.assertIn('cannot be started without cmd.exe', result['next'])
self.assertEqual(public_data._EFFECTS[0], before) # refused before the launch's effect
with patch.dict(scorer_client.SCORER, {'provider': 'claude'}), \
self.assertRaisesRegex(scorer_client.ScorerUnavailable, 'cannot be started without cmd.exe'):
scorer_client.scorer_prefix()
self.assertEqual(self.recorded(), [])
self.assertFalse(list(self.folder.glob('args-*.txt')))
self.assert_nothing_injected()
if __name__ == '__main__':
unittest.main()
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 |