fix(droid): harden exec passthrough and add integration coverage

This commit is contained in:
Tam Nhu Tran
2026-02-26 23:28:36 +07:00
parent 5963ba973e
commit ce44d2ece0
6 changed files with 441 additions and 42 deletions
+4
View File
@@ -172,11 +172,15 @@ CCS supports structural Droid command passthrough after profile selection:
```bash
ccsd codex exec --skip-permissions-unsafe "fix failing tests"
ccsd codex --skip-permissions-unsafe "fix failing tests" # auto-routed to: droid exec ...
ccsd codex -m custom:gpt-5.3-codex "fix failing tests" # short exec flags auto-routed too
```
If you pass exec-only flags without a prompt (for example `--skip-permissions-unsafe`),
Droid `exec` will return its native "No prompt provided" usage guidance.
If multiple reasoning flags are provided in Droid exec mode, CCS keeps the first
flag and warns about duplicates.
Dashboard parity: `ccs config` -> `Factory Droid`
### Per-Profile Target Defaults
+26 -16
View File
@@ -62,7 +62,7 @@ import {
DroidReasoningFlagError,
resolveDroidReasoningRuntime,
} from './targets/droid-reasoning-runtime';
import { routeDroidCommandArgs } from './targets/droid-command-router';
import { DroidCommandRouterError, routeDroidCommandArgs } from './targets/droid-command-router';
// Version and Update check utilities
import { getVersion } from './utils/version';
@@ -719,11 +719,11 @@ async function main(): Promise<void> {
let targetRemainingArgs = remainingArgs;
let droidReasoningOverride: string | number | undefined;
if (resolvedTarget === 'droid') {
const droidRoute = routeDroidCommandArgs(remainingArgs);
targetRemainingArgs = droidRoute.argsForDroid;
try {
const droidRoute = routeDroidCommandArgs(remainingArgs);
targetRemainingArgs = droidRoute.argsForDroid;
if (droidRoute.mode === 'interactive') {
try {
if (droidRoute.mode === 'interactive') {
const runtime = resolveDroidReasoningRuntime(remainingArgs, process.env.CCS_THINKING);
targetRemainingArgs = runtime.argsWithoutReasoningFlags;
droidReasoningOverride = runtime.reasoningOverride;
@@ -735,19 +735,29 @@ async function main(): Promise<void> {
)
);
}
} catch (error) {
if (error instanceof DroidReasoningFlagError) {
console.error(fail(error.message));
console.error(' Examples: --thinking low, --thinking 8192, --thinking off');
console.error(' Codex alias: --effort medium|high|xhigh');
process.exit(1);
} else {
if (droidRoute.duplicateReasoningDisplays.length > 0) {
console.error(
warn(
`[!] Multiple reasoning flags detected. Using first occurrence: ${droidRoute.reasoningSourceDisplay || '<first-flag>'}`
)
);
}
if (droidRoute.autoPrependedExec && process.stdout.isTTY) {
console.error(
info('Detected Droid exec-only flags. Routing as: droid exec <flags> [prompt]')
);
}
throw error;
}
} else if (droidRoute.autoPrependedExec && process.stdout.isTTY) {
console.error(
info('Detected Droid exec-only flags. Routing as: droid exec <flags> [prompt]')
);
} catch (error) {
if (error instanceof DroidReasoningFlagError || error instanceof DroidCommandRouterError) {
console.error(fail(error.message));
console.error(' Examples: --thinking low, --thinking 8192, --thinking off');
console.error(' Codex alias: --effort medium|high|xhigh');
console.error(' Droid exec: --reasoning-effort high');
process.exit(1);
}
throw error;
}
}
+2
View File
@@ -342,6 +342,7 @@ Run ${color('ccs config', 'command')} for web dashboard`.trim();
'ccsd codex exec --skip-permissions-unsafe "fix failing tests"',
'Pass through Droid exec mode',
],
['ccsd codex -m custom:gpt-5.3-codex "fix failing tests"', 'Auto-routes short exec flags'],
[
'ccsd codex --skip-permissions-unsafe "fix failing tests"',
'Auto-routes to Droid exec when exec-only flags are detected',
@@ -404,6 +405,7 @@ Run ${color('ccs config', 'command')} for web dashboard`.trim();
['Droid exec:', 'Use native Droid flag: --reasoning-effort <level>'],
['', 'CCS auto-maps --thinking/--effort to --reasoning-effort in droid exec mode.'],
['', 'For interactive droid sessions, CCS applies reasoning via Droid BYOK model config.'],
['', 'When multiple reasoning flags are provided, the first flag wins.'],
['', ''],
['Note:', 'Extended thinking allocates compute for step-by-step reasoning'],
['', 'before responding.'],
+148 -18
View File
@@ -15,6 +15,20 @@ export interface DroidCommandRoute {
argsForDroid: string[];
command?: string;
autoPrependedExec: boolean;
reasoningSourceDisplay?: string;
duplicateReasoningDisplays: string[];
}
type DroidReasoningFlag = '--reasoning-effort' | '-r' | '--effort' | '--thinking';
export class DroidCommandRouterError extends Error {
constructor(
message: string,
public readonly flag: DroidReasoningFlag
) {
super(message);
this.name = 'DroidCommandRouterError';
}
}
const DROID_SUBCOMMANDS = new Set([
@@ -48,6 +62,19 @@ const DROID_EXEC_ONLY_LONG_FLAGS = new Set([
'--list-tools',
]);
const DROID_EXEC_ONLY_SHORT_FLAGS = new Set(['-o', '-f', '-s', '-m']);
const DROID_REASONING_EFFORT_VALUES = new Set([
'none',
'off',
'minimal',
'low',
'medium',
'high',
'max',
'xhigh',
'auto',
]);
function getLongFlagToken(arg: string): string {
const eqIndex = arg.indexOf('=');
return eqIndex >= 0 ? arg.slice(0, eqIndex) : arg;
@@ -57,6 +84,11 @@ function isExplicitSubcommand(arg: string | undefined): boolean {
return !!arg && DROID_SUBCOMMANDS.has(arg);
}
function isLikelyReasoningEffortValue(value: string | undefined): boolean {
if (!value || value.startsWith('-')) return false;
return DROID_REASONING_EFFORT_VALUES.has(value.toLowerCase());
}
function hasExecOnlyFlagsAtFront(args: string[]): boolean {
for (let i = 0; i < args.length; i++) {
const arg = args[i];
@@ -75,7 +107,19 @@ function hasExecOnlyFlagsAtFront(args: string[]): boolean {
}
if (!arg.startsWith('-')) return false;
if (!arg.startsWith('--')) continue; // short flags are ambiguous at root (`-r` is resume)
if (!arg.startsWith('--')) {
// Short flags:
// - `-r` is ambiguous (root resume vs exec reasoning-effort), so only route
// when value looks like a reasoning effort level.
if (DROID_EXEC_ONLY_SHORT_FLAGS.has(arg)) {
return true;
}
if (arg === '-r') {
const value = args[i + 1];
return isLikelyReasoningEffortValue(value);
}
continue;
}
const flagToken = getLongFlagToken(arg);
if (DROID_EXEC_ONLY_LONG_FLAGS.has(flagToken)) {
@@ -86,38 +130,112 @@ function hasExecOnlyFlagsAtFront(args: string[]): boolean {
return false;
}
function normalizeLegacyReasoningAliasesForExec(args: string[]): string[] {
const normalized: string[] = [];
interface ExecReasoningNormalizationResult {
args: string[];
sourceDisplay?: string;
duplicateDisplays: string[];
}
for (let i = 0; i < args.length; i++) {
function normalizeExecReasoningFlags(args: string[]): ExecReasoningNormalizationResult {
const normalized: string[] = [];
const duplicateDisplays: string[] = [];
let sourceDisplay: string | undefined;
let hasReasoning = false;
const applyReasoning = (value: string, display: string): void => {
if (!hasReasoning) {
normalized.push('--reasoning-effort', value);
hasReasoning = true;
sourceDisplay = display;
return;
}
duplicateDisplays.push(display);
};
const handleMissingValue = (
flag: DroidReasoningFlag,
missingDisplay: string
): never | undefined => {
if (!hasReasoning) {
throw new DroidCommandRouterError(`${flag} requires a value`, flag);
}
duplicateDisplays.push(missingDisplay);
return undefined;
};
// Preserve leading command token for explicit auto-prepended command mode.
const startsWithExec = args[0] === 'exec';
let startIndex = 0;
if (startsWithExec) {
normalized.push('exec');
startIndex = 1;
}
for (let i = startIndex; i < args.length; i++) {
const arg = args[i];
if (arg === '--effort' || arg === '--thinking') {
if (arg === '--') {
normalized.push(...args.slice(i));
break;
}
if (
arg === '--reasoning-effort' ||
arg === '--effort' ||
arg === '--thinking' ||
arg === '-r'
) {
const value = args[i + 1];
if (value && !value.startsWith('-')) {
normalized.push('--reasoning-effort', value);
i += 1;
} else {
// Keep invalid/missing-value form so Droid can surface native validation.
normalized.push(arg);
if (!value || value.startsWith('-')) {
handleMissingValue(arg as DroidReasoningFlag, `${arg} <missing-value>`);
continue;
}
applyReasoning(value, `${arg} ${value}`);
i += 1;
continue;
}
if (arg.startsWith('--reasoning-effort=')) {
const value = arg.slice('--reasoning-effort='.length);
if (!value) {
handleMissingValue('--reasoning-effort', '--reasoning-effort=<missing-value>');
continue;
}
applyReasoning(value, `--reasoning-effort=${value}`);
continue;
}
if (arg.startsWith('--effort=')) {
normalized.push(`--reasoning-effort=${arg.slice('--effort='.length)}`);
const value = arg.slice('--effort='.length);
if (!value) {
handleMissingValue('--effort', '--effort=<missing-value>');
continue;
}
applyReasoning(value, `--effort=${value}`);
continue;
}
if (arg.startsWith('--thinking=')) {
normalized.push(`--reasoning-effort=${arg.slice('--thinking='.length)}`);
const value = arg.slice('--thinking='.length);
if (!value) {
handleMissingValue('--thinking', '--thinking=<missing-value>');
continue;
}
applyReasoning(value, `--thinking=${value}`);
continue;
}
normalized.push(arg);
}
return normalized;
return {
args: normalized,
sourceDisplay,
duplicateDisplays,
};
}
export function routeDroidCommandArgs(args: string[]): DroidCommandRoute {
@@ -126,28 +244,39 @@ export function routeDroidCommandArgs(args: string[]): DroidCommandRoute {
mode: 'interactive',
argsForDroid: [],
autoPrependedExec: false,
duplicateReasoningDisplays: [],
};
}
if (isExplicitSubcommand(args[0])) {
const command = args[0];
const argsForDroid =
command === 'exec' ? normalizeLegacyReasoningAliasesForExec(args) : [...args];
const normalized =
command === 'exec'
? normalizeExecReasoningFlags(args)
: {
args: [...args],
duplicateDisplays: [],
};
return {
mode: 'command',
command,
argsForDroid,
argsForDroid: normalized.args,
autoPrependedExec: false,
reasoningSourceDisplay: normalized.sourceDisplay,
duplicateReasoningDisplays: normalized.duplicateDisplays,
};
}
if (hasExecOnlyFlagsAtFront(args)) {
const argsWithExec = ['exec', ...args];
const normalized = normalizeExecReasoningFlags(argsWithExec);
return {
mode: 'command',
command: 'exec',
argsForDroid: normalizeLegacyReasoningAliasesForExec(argsWithExec),
argsForDroid: normalized.args,
autoPrependedExec: true,
reasoningSourceDisplay: normalized.sourceDisplay,
duplicateReasoningDisplays: normalized.duplicateDisplays,
};
}
@@ -155,5 +284,6 @@ export function routeDroidCommandArgs(args: string[]): DroidCommandRoute {
mode: 'interactive',
argsForDroid: [...args],
autoPrependedExec: false,
duplicateReasoningDisplays: [],
};
}
@@ -1,5 +1,8 @@
import { describe, expect, it } from 'bun:test';
import { routeDroidCommandArgs } from '../../../src/targets/droid-command-router';
import {
DroidCommandRouterError,
routeDroidCommandArgs,
} from '../../../src/targets/droid-command-router';
describe('droid-command-router', () => {
it('keeps interactive mode for plain profile launches', () => {
@@ -8,6 +11,7 @@ describe('droid-command-router', () => {
expect(route.mode).toBe('interactive');
expect(route.argsForDroid).toEqual([]);
expect(route.autoPrependedExec).toBe(false);
expect(route.duplicateReasoningDisplays).toEqual([]);
});
it('keeps explicit droid subcommands untouched', () => {
@@ -17,6 +21,7 @@ describe('droid-command-router', () => {
expect(route.command).toBe('mcp');
expect(route.argsForDroid).toEqual(['mcp', '--help']);
expect(route.autoPrependedExec).toBe(false);
expect(route.duplicateReasoningDisplays).toEqual([]);
});
it('auto-prepends exec for exec-only flags provided after profile', () => {
@@ -26,6 +31,7 @@ describe('droid-command-router', () => {
expect(route.command).toBe('exec');
expect(route.argsForDroid).toEqual(['exec', '--skip-permissions-unsafe']);
expect(route.autoPrependedExec).toBe(true);
expect(route.duplicateReasoningDisplays).toEqual([]);
});
it('does not auto-prepend exec for root help flag', () => {
@@ -34,6 +40,7 @@ describe('droid-command-router', () => {
expect(route.mode).toBe('interactive');
expect(route.argsForDroid).toEqual(['--help']);
expect(route.autoPrependedExec).toBe(false);
expect(route.duplicateReasoningDisplays).toEqual([]);
});
it('normalizes --effort alias to --reasoning-effort for explicit exec', () => {
@@ -41,12 +48,8 @@ describe('droid-command-router', () => {
expect(route.mode).toBe('command');
expect(route.command).toBe('exec');
expect(route.argsForDroid).toEqual([
'exec',
'--reasoning-effort',
'xhigh',
'fix test flake',
]);
expect(route.argsForDroid).toEqual(['exec', '--reasoning-effort', 'xhigh', 'fix test flake']);
expect(route.reasoningSourceDisplay).toBe('--effort xhigh');
});
it('normalizes --thinking alias when exec is auto-prepended', () => {
@@ -58,10 +61,12 @@ describe('droid-command-router', () => {
'exec',
'--auto',
'high',
'--reasoning-effort=medium',
'--reasoning-effort',
'medium',
'summarize logs',
]);
expect(route.autoPrependedExec).toBe(true);
expect(route.reasoningSourceDisplay).toBe('--thinking=medium');
});
it('still auto-prepends exec when --effort appears before exec-only flags', () => {
@@ -82,5 +87,80 @@ describe('droid-command-router', () => {
'fix flaky test',
]);
expect(route.autoPrependedExec).toBe(true);
expect(route.reasoningSourceDisplay).toBe('--effort xhigh');
});
it('auto-prepends exec for non-ambiguous short exec flags', () => {
const route = routeDroidCommandArgs(['-m', 'custom:gpt-5.3-codex', 'fix flaky test']);
expect(route.mode).toBe('command');
expect(route.command).toBe('exec');
expect(route.argsForDroid).toEqual(['exec', '-m', 'custom:gpt-5.3-codex', 'fix flaky test']);
expect(route.autoPrependedExec).toBe(true);
});
it('routes -r to exec when value matches reasoning effort level', () => {
const route = routeDroidCommandArgs(['-r', 'high', 'summarize logs']);
expect(route.mode).toBe('command');
expect(route.command).toBe('exec');
expect(route.argsForDroid).toEqual(['exec', '--reasoning-effort', 'high', 'summarize logs']);
expect(route.autoPrependedExec).toBe(true);
});
it('keeps interactive mode for ambiguous -r resume-style usage', () => {
const route = routeDroidCommandArgs(['-r', 'session-1234']);
expect(route.mode).toBe('interactive');
expect(route.argsForDroid).toEqual(['-r', 'session-1234']);
expect(route.autoPrependedExec).toBe(false);
});
it('dedupes mixed reasoning flags with first occurrence precedence', () => {
const route = routeDroidCommandArgs([
'exec',
'--reasoning-effort',
'high',
'--thinking',
'low',
'--effort=xhigh',
'summarize logs',
]);
expect(route.mode).toBe('command');
expect(route.command).toBe('exec');
expect(route.argsForDroid).toEqual(['exec', '--reasoning-effort', 'high', 'summarize logs']);
expect(route.reasoningSourceDisplay).toBe('--reasoning-effort high');
expect(route.duplicateReasoningDisplays).toEqual(['--thinking low', '--effort=xhigh']);
});
it('throws for missing reasoning value in command mode (alias)', () => {
expect(() => routeDroidCommandArgs(['exec', '--effort'])).toThrow(DroidCommandRouterError);
});
it('throws for missing reasoning value in command mode (native)', () => {
expect(() => routeDroidCommandArgs(['exec', '--reasoning-effort'])).toThrow(
DroidCommandRouterError
);
});
it('records malformed duplicate reasoning flags when first value is already selected', () => {
const route = routeDroidCommandArgs([
'exec',
'--thinking',
'medium',
'--reasoning-effort',
'--skip-permissions-unsafe',
'summarize logs',
]);
expect(route.argsForDroid).toEqual([
'exec',
'--reasoning-effort',
'medium',
'--skip-permissions-unsafe',
'summarize logs',
]);
expect(route.duplicateReasoningDisplays).toEqual(['--reasoning-effort <missing-value>']);
});
});
@@ -0,0 +1,173 @@
import { afterEach, beforeEach, describe, expect, it } from 'bun:test';
import { spawnSync } from 'child_process';
import * as fs from 'fs';
import * as os from 'os';
import * as path from 'path';
interface RunResult {
status: number | null;
stdout: string;
stderr: string;
}
function runCcs(args: string[], env: NodeJS.ProcessEnv): RunResult {
const ccsEntry = path.join(process.cwd(), 'src', 'ccs.ts');
const result = spawnSync(process.execPath, [ccsEntry, ...args], {
encoding: 'utf8',
env,
timeout: 20000,
});
return {
status: result.status,
stdout: result.stdout || '',
stderr: result.stderr || '',
};
}
describe('droid command routing integration', () => {
let tmpHome: string;
let ccsDir: string;
let settingsPath: string;
let configPath: string;
let fakeDroidPath: string;
let droidArgsLogPath: string;
let baseEnv: NodeJS.ProcessEnv;
beforeEach(() => {
if (process.platform === 'win32') {
return;
}
tmpHome = fs.mkdtempSync(path.join(os.tmpdir(), 'ccs-droid-route-it-'));
ccsDir = path.join(tmpHome, '.ccs');
fs.mkdirSync(ccsDir, { recursive: true });
settingsPath = path.join(ccsDir, 'myglm.settings.json');
configPath = path.join(ccsDir, 'config.json');
fakeDroidPath = path.join(tmpHome, 'fake-droid.js');
droidArgsLogPath = path.join(tmpHome, 'droid-args.json');
fs.writeFileSync(
settingsPath,
JSON.stringify(
{
env: {
ANTHROPIC_BASE_URL: 'https://example.invalid/anthropic',
ANTHROPIC_AUTH_TOKEN: 'test-token',
ANTHROPIC_MODEL: 'gpt-5.3-codex',
CCS_DROID_PROVIDER: 'openai',
},
},
null,
2
)
);
fs.writeFileSync(
configPath,
JSON.stringify(
{
profiles: {
myglm: settingsPath,
},
},
null,
2
)
);
fs.writeFileSync(
fakeDroidPath,
`#!/usr/bin/env node
const fs = require('fs');
const out = process.env.CCS_TEST_DROID_ARGS_OUT;
if (!out) process.exit(2);
fs.writeFileSync(out, JSON.stringify(process.argv.slice(2)));
process.exit(0);
`,
{ encoding: 'utf8', mode: 0o755 }
);
fs.chmodSync(fakeDroidPath, 0o755);
baseEnv = {
...process.env,
CI: '1',
NO_COLOR: '1',
CCS_HOME: tmpHome,
CCS_DROID_PATH: fakeDroidPath,
CCS_TEST_DROID_ARGS_OUT: droidArgsLogPath,
};
});
afterEach(() => {
if (process.platform === 'win32') {
return;
}
fs.rmSync(tmpHome, { recursive: true, force: true });
});
it('auto-routes exec-only long flags to droid exec from main ccs flow', () => {
if (process.platform === 'win32') return;
const result = runCcs(
['myglm', '--target', 'droid', '--skip-permissions-unsafe', 'fix failing tests'],
baseEnv
);
expect(result.status).toBe(0);
const routedArgs = JSON.parse(fs.readFileSync(droidArgsLogPath, 'utf8')) as string[];
expect(routedArgs).toEqual(['exec', '--skip-permissions-unsafe', 'fix failing tests']);
});
it('auto-routes non-ambiguous short exec flags', () => {
if (process.platform === 'win32') return;
const result = runCcs(
['myglm', '--target', 'droid', '-m', 'custom:gpt-5.3-codex', 'fix failing tests'],
baseEnv
);
expect(result.status).toBe(0);
const routedArgs = JSON.parse(fs.readFileSync(droidArgsLogPath, 'utf8')) as string[];
expect(routedArgs).toEqual(['exec', '-m', 'custom:gpt-5.3-codex', 'fix failing tests']);
});
it('dedupes reasoning flags with first occurrence precedence in exec mode', () => {
if (process.platform === 'win32') return;
const result = runCcs(
[
'myglm',
'--target',
'droid',
'exec',
'--reasoning-effort',
'high',
'--thinking',
'low',
'summarize logs',
],
baseEnv
);
expect(result.status).toBe(0);
expect(result.stderr).toContain('Multiple reasoning flags detected');
const routedArgs = JSON.parse(fs.readFileSync(droidArgsLogPath, 'utf8')) as string[];
expect(routedArgs).toEqual(['exec', '--reasoning-effort', 'high', 'summarize logs']);
});
it('fails fast for malformed reasoning alias in command mode', () => {
if (process.platform === 'win32') return;
const result = runCcs(['myglm', '--target', 'droid', 'exec', '--effort'], baseEnv);
expect(result.status).toBe(1);
expect(result.stderr).toContain('--effort requires a value');
expect(fs.existsSync(droidArgsLogPath)).toBe(true);
const probeArgs = JSON.parse(fs.readFileSync(droidArgsLogPath, 'utf8')) as string[];
// Droid binary is still invoked once for version preflight (`--version`) before routing.
expect(probeArgs).toEqual(['--version']);
});
});