An agentic browser harness: an AI side panel & orb in your own logged-in Chrome, Edge, Arc, Brave or even Firefox run by the agent/subscription you already use (Claude Code, Codex, Antigravity). Free, open source, no API key. Also an MCP server.
TypeScript agentic-browseragentic-browsingagy-cliaiai-agentai-browser-extensionai-toolsbrave-extensionbrowser-agentbrowser-automationbrowser-extensionbrowser-usebrowserskillchrome-extensionclaude-codecodexfirefox-addonmcp-servernanobrowserweb-automation
/**
* Describing the policy to the panel that renders it.
*
* Everything here is derived from `DEFAULT_RULES` and the live config, so a rule added
* to the policy appears in the settings screen with no second edit — and a rule whose
* title or reason changes says the new thing in both places.
*
* `fallback` is the effect this install would use with nothing overridden, which is not
* always the shipped constant: `form-submission` takes its default from the legacy
* `requireApproval` key. Computing it by re-running `policyFrom` without the rule
* overrides is the only way to get that right without restating the derivation.
*/
import {
FENCE_SETTING,
UNATTENDED_SETTING,
isRuleEffect,
type GuardrailSettings,
type GuardrailToggle,
type GuardrailValue,
type RuleEffect,
} from '@/lib/settings/guardrails';
import { DEFAULT_RULES, policyFrom, type GuardrailConfig } from './policy';
/**
* Not preferences. Allowing `javascript:` navigation, letting a page call an internal
* verb, waving through a URL that resolves to no destination, or letting a released
* credential travel in a query string have no use worth a switch someone can hit by
* accident. Hand-editing config.json still works.
*/
const LOCKED = new Set(['reserved-action', 'non-http-navigation', 'unreadable-navigation', 'secret-in-url']);
const RULE_IDS = new Set(DEFAULT_RULES.map((rule) => rule.id));
export function guardrailSettings(
config: GuardrailConfig,
requireApproval: readonly string[],
configPath: string,
): GuardrailSettings {
const overrides = config.rules ?? {};
const baseline = policyFrom({ ...config, rules: {} }, requireApproval);
const rules: GuardrailToggle[] = baseline.rules.map((rule) => {
const override = overrides[rule.id];
return {
id: rule.id,
title: rule.title,
reason: rule.reason,
fallback: rule.effect,
...(isRuleEffect(override) ? { override } : {}),
...(LOCKED.has(rule.id) ? { locked: true } : {}),
};
});
return {
rules,
fence: { enabled: config.fence !== false, overridden: config.fence !== undefined },
unattended: {
effect: config.unattended === 'allow' ? 'allow' : 'deny',
overridden: config.unattended !== undefined,
},
hosts: config.hosts ?? [],
configPath,
};
}
/** Whether the panel may write this setting at all, and whether the value fits it. */
export function settingWritable(setting: string, value: GuardrailValue): boolean {
if (setting === FENCE_SETTING) return value === null || typeof value === 'boolean';
if (setting === UNATTENDED_SETTING) return value === null || value === 'allow' || value === 'deny';
if (!RULE_IDS.has(setting) || LOCKED.has(setting)) return false;
return value === null || isRuleEffect(value);
}
export function ruleEffect(value: GuardrailValue): RuleEffect | null {
return isRuleEffect(value) ? value : null;
}