@equinor/fusion-framework-cli-plugin-copilot 1.0.0

package/dist/types/commands/app/tools/go-back.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the back navigation tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for going back in browser history
+ */
+export declare function createGoBackTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/hover.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the hover interaction tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for hovering a page element
+ */
+export declare function createHoverTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/index.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, AgentBrowserToolList, DefineTool } from './types.js';
+/**
+ * Creates the full set of Copilot browser tools for an eval session.
+ *
+ * @param context - Shared execution context for all browser tool wrappers
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Registered browser tools in the order they should be exposed to the model
+ */
+export declare function createAgentBrowserTools(context: AgentBrowserToolContext, defineTool: DefineTool): AgentBrowserToolList;

package/dist/types/commands/app/tools/navigate.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the browser navigation tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for opening a URL
+ */
+export declare function createNavigateTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/press-key.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the keyboard press tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for pressing a single key
+ */
+export declare function createPressKeyTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/read-file.d.ts

@@ -0,0 +1,13 @@
+import type { DefineTool } from './types.js';
+/**
+ * Creates the read-file tool so the judge agent can inspect evidence artifacts.
+ *
+ * For image files the tool returns a `ToolBinaryResult` with base64-encoded
+ * data that multimodal models can view directly. Text files are returned
+ * as plain text. Directories return a listing of their contents.
+ *
+ * @param baseDir - The run output directory that scopes all reads
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for reading evidence files
+ */
+export declare function createReadFileTool(baseDir: string, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/registry.d.ts

@@ -0,0 +1,28 @@
+import type { AgentBrowserToolList } from './types.js';
+/**
+ * Ordered collection of agent-browser tool registrations for a Copilot session.
+ *
+ * Tools are exposed to the model in insertion order. Use {@link addTools} to
+ * register multiple tools at once during session setup.
+ */
+export declare class AgentBrowserToolRegistry {
+    private readonly tools;
+    /**
+     * Adds a tool to the registry.
+     *
+     * @param tool - Tool definition created with the Copilot SDK `defineTool` helper
+     */
+    addTool(tool: AgentBrowserToolList[number]): void;
+    /**
+     * Adds multiple tools to the registry in order.
+     *
+     * @param tools - Tool definitions created with the Copilot SDK `defineTool` helper
+     */
+    addTools(...tools: AgentBrowserToolList): void;
+    /**
+     * Returns a shallow copy of the registered tools in insertion order.
+     *
+     * @returns The registered Copilot SDK tool list
+     */
+    getTools(): AgentBrowserToolList;
+}

package/dist/types/commands/app/tools/reload.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the page reload tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for reloading the current page
+ */
+export declare function createReloadTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/screenshot.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the screenshot capture tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for saving an image screenshot and returning it to the model
+ */
+export declare function createScreenshotTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/scroll.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the scroll tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for page scrolling or scrolling an element into view
+ */
+export declare function createScrollTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/select.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the select interaction tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for choosing an option in a select element
+ */
+export declare function createSelectTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/snapshot.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the accessibility snapshot tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for capturing the current page snapshot
+ */
+export declare function createSnapshotTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/type-text.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the text typing tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for typing text into the focused element or a target
+ */
+export declare function createTypeTextTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/types.d.ts

@@ -0,0 +1,12 @@
+import type { SessionConfig, defineTool } from '@github/copilot-sdk';
+/** Shared runtime context passed to every agent-browser tool factory. */
+export interface AgentBrowserToolContext {
+    /** Directory where evidence artifacts for the current eval run are stored. */
+    outDir: string;
+    /** Executes an `agent-browser` command and returns its stdout. */
+    invoke: (args: string[], timeoutMs?: number) => string;
+}
+/** Type-only alias for the Copilot SDK `defineTool` helper. */
+export type DefineTool = typeof defineTool;
+/** Copilot SDK tool list type used by the registry. */
+export type AgentBrowserToolList = NonNullable<SessionConfig['tools']>;

package/dist/types/commands/app/tools/wait.d.ts

@@ -0,0 +1,9 @@
+import type { AgentBrowserToolContext, DefineTool } from './types.js';
+/**
+ * Creates the wait tool.
+ *
+ * @param context - Shared browser tool execution context
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for waiting on load state, text, elements, or time
+ */
+export declare function createWaitTool(context: AgentBrowserToolContext, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/tools/write-file.d.ts

@@ -0,0 +1,19 @@
+import type { DefineTool } from './types.js';
+/**
+ * Creates the write-file tool so the agent can save artifacts
+ * (plan.json, verdict.json, etc.) to the run output directory.
+ *
+ * @param baseDir - The run output directory that scopes all writes
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for writing a file
+ */
+export declare function createWriteFileTool(baseDir: string, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;
+/**
+ * Creates the append-file tool so the agent can append lines
+ * to artifacts like executions.jsonl.
+ *
+ * @param baseDir - The run output directory that scopes all writes
+ * @param defineTool - Copilot SDK helper used to declare tools
+ * @returns Copilot tool definition for appending to a file
+ */
+export declare function createAppendFileTool(baseDir: string, defineTool: DefineTool): import("@github/copilot-sdk").Tool<unknown>;

package/dist/types/commands/app/types.d.ts

@@ -0,0 +1,68 @@
+import type { Observable } from 'rxjs';
+export interface RuntimeExecutionContext {
+    outDir: string;
+    url: string;
+}
+/** CLI option set for the `copilot app eval` command. */
+export interface CopilotEvalOptions {
+    /** Port for the local app dev server. */
+    port: string;
+    /** Host address for the local app dev server. */
+    host: string;
+    /** Skip server start and use an already-running URL. */
+    url?: string;
+    /** Show detailed output from the dev server and agent-browser. */
+    verbose: boolean;
+    /** LLM model override (e.g. "claude-sonnet-4"). */
+    model?: string;
+    /** Reasoning effort: low, medium, high, xhigh. */
+    reasoning?: 'low' | 'medium' | 'high' | 'xhigh';
+    /** Output directory override for run artifacts. */
+    output?: string;
+    /** Open a headed browser for interactive MSAL login. */
+    login: boolean;
+    /** Specific eval to run (name or file path). */
+    eval?: string;
+    /** Observable that emits on SIGINT for graceful shutdown. */
+    shutdown$?: Observable<string>;
+}
+/** A logical test step that groups related criteria under one scenario. */
+export interface PlanStep {
+    /** Short scenario description (e.g. "Open the landing page"). */
+    scenario: string;
+    /** Acceptance criteria to verify (e.g. ["Header shows Hello Fusion", ...]). */
+    criteria: string[];
+    /** What constitutes a pass for each criterion (same order as criteria). */
+    pass: string[];
+    /** What constitutes a fail for each criterion (same order as criteria). */
+    fail: string[];
+}
+/** Structured plan produced by Phase 1. */
+export interface Plan {
+    /** Brief summary of the feature under test. */
+    summary: string;
+    /** Ordered acceptance criteria to execute. */
+    steps: PlanStep[];
+}
+/** Result of executing a single {@link PlanStep}. */
+export interface StepResult {
+    /** The criterion that was tested. */
+    criterion: string;
+    /** Whether the criterion passed. */
+    ok: boolean;
+    /** Agent's observation or explanation. */
+    note: string;
+    /** Evidence artifact filenames collected during this step. */
+    evidence: string[];
+}
+/** Structured verdict produced by the Copilot agent at the end of an eval session. */
+export interface Verdict {
+    /** Whether all acceptance criteria passed. */
+    pass: boolean;
+    /** Overall summary explaining the pass/fail outcome. */
+    reasoning: string;
+    /** Per-criterion results with evidence references. */
+    steps: StepResult[];
+    /** Optional UX observations and improvement suggestions based on what was seen. */
+    ux?: string[];
+}

package/dist/types/eval-resolve.d.ts

@@ -0,0 +1,31 @@
+/**
+ * Normalized positional input for the `copilot app eval` command.
+ *
+ * When the user passes an application directory, `evalFile` is undefined.
+ * When the user passes a direct `eval/*.md` file path, `appPath` is derived
+ * from the file location and `evalFile` points to the selected markdown file.
+ */
+export interface EvalCommandInput {
+    /** Absolute path to the Fusion application directory. */
+    appPath: string;
+    /** Optional absolute path to the directly selected eval markdown file. */
+    evalFile?: string;
+}
+/**
+ * Resolves the positional `copilot app eval` argument.
+ *
+ * Supports both application directory paths and direct paths to markdown files
+ * inside an application's `eval/` directory.
+ *
+ * @param inputPath - Positional CLI argument passed to the eval command.
+ * @returns Normalized application path and optional direct eval file path.
+ */
+export declare function resolveEvalCommandInput(inputPath: string): EvalCommandInput;
+/**
+ * Resolves eval markdown files for a Fusion application.
+ *
+ * @param appPath - Path to the Fusion application directory
+ * @param evalOpt - Value of the `--eval` CLI option (name, path, or `undefined`)
+ * @returns Array of absolute paths to eval markdown files
+ */
+export declare function resolveEvalFiles(appPath: string, evalOpt?: string): string[];

package/dist/types/index.d.ts

@@ -0,0 +1,15 @@
+import type { Command } from 'commander';
+/**
+ * Registers the `copilot` CLI plugin with the Fusion Framework CLI.
+ *
+ * Attaches the top-level `copilot` command group, providing Copilot SDK
+ * powered evaluation tools with agent-browser.
+ *
+ * Usage:
+ *   ffc copilot app eval ./cookbooks/app-react
+ *   ffc copilot app eval . --eval smoke --model claude-sonnet-4
+ *
+ * @param program - The root Commander program instance
+ */
+export declare function registerCopilotPlugin(program: Command): void;
+export default registerCopilotPlugin;

package/dist/types/utils/agent-browser.d.ts

@@ -0,0 +1,26 @@
+/** Persistent Chrome profile directory used across agent-browser sessions. */
+export declare const PROFILE_DIR: string;
+/**
+ * Runs an agent-browser command with the persistent Chrome profile.
+ *
+ * Calls the `agent-browser` binary resolved from `PATH` (homebrew or npm global).
+ * Every invocation automatically injects `--profile <PROFILE_DIR>` so browser
+ * state (cookies, session storage, MSAL tokens) persists across calls.
+ *
+ * @param args - Subcommand and arguments passed directly to `agent-browser`
+ * @param timeoutMs - Maximum execution time in milliseconds (default: 30 000)
+ * @returns Trimmed stdout from the command
+ * @throws {Error} When `agent-browser` exits with a non-zero code or times out
+ */
+export declare function ab(args: string[], timeoutMs?: number): string;
+/**
+ * Extracts the user-facing error message from an agent-browser failure.
+ *
+ * Strips ANSI escape codes from stderr and returns only the diagnostic
+ * lines prefixed with `✗` or `⚠`. Falls back to the first 200 characters
+ * of the error message when no diagnostic lines are found.
+ *
+ * @param err - The caught error (typically from {@link ab})
+ * @returns A single-line summary suitable for console output
+ */
+export declare function abErrorMessage(err: unknown): string;

package/dist/types/utils/browser-poll.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Polls the browser's console log output until a predicate returns `true`
+ * or the timeout expires.
+ *
+ * @param predicate - Callback receiving the raw console log text; return `true` to stop
+ * @param timeoutMs - Maximum polling duration in milliseconds
+ * @returns `true` if the predicate matched before timeout, `false` otherwise
+ */
+export declare function pollConsole(predicate: (logs: string) => boolean, timeoutMs: number): Promise<boolean>;

package/dist/types/utils/daemon.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Kills any running agent-browser daemon and its Chrome children,
+ * then removes stale state files so the next run starts clean.
+ *
+ * Cleanup sequence:
+ * 1. Graceful `agent-browser close` (5 s timeout)
+ * 2. Force-kill the daemon PID read from {@link DAEMON_PID_FILE}
+ * 3. `pkill` leftover headless Chrome processes
+ * 4. Delete Session Storage to clear stuck MSAL `interaction.status` flags
+ */
+export declare function resetDaemon(): void;
+/**
+ * Clears the MSAL `interaction.status` flag from `sessionStorage` in the
+ * running browser via `agent-browser eval`.
+ */
+export declare function clearMsalInteraction(): void;

package/dist/types/utils/index.d.ts

@@ -0,0 +1,5 @@
+export { ab, abErrorMessage, PROFILE_DIR } from './agent-browser.js';
+export { resetDaemon, clearMsalInteraction } from './daemon.js';
+export { cleanup, sleep, stripAnsi } from './process.js';
+export { createRunDir, resolveAppKey, waitForServer } from './server.js';
+export { pollConsole } from './browser-poll.js';

package/dist/types/utils/process.d.ts

@@ -0,0 +1,21 @@
+import type { ChildProcess } from 'node:child_process';
+/**
+ * Sends `SIGTERM` to a child process if it is still running.
+ *
+ * @param proc - The child process to terminate (no-op when `undefined` or already killed)
+ */
+export declare function cleanup(proc?: ChildProcess): void;
+/**
+ * Returns a promise that resolves after the given delay.
+ *
+ * @param ms - Delay in milliseconds
+ * @returns A promise that resolves after `ms` milliseconds
+ */
+export declare function sleep(ms: number): Promise<void>;
+/**
+ * Strips ANSI escape codes (color / formatting) from a string.
+ *
+ * @param s - The string potentially containing ANSI sequences
+ * @returns The input string with all ANSI escape sequences removed
+ */
+export declare function stripAnsi(s: string): string;

package/dist/types/utils/server.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Creates a timestamped, run-specific output directory for eval artifacts.
+ *
+ * @param title - Human label used as the directory name prefix (slugified)
+ * @param baseDir - Optional parent directory override
+ * @returns Absolute path to the newly created directory
+ */
+export declare function createRunDir(title: string, baseDir?: string): string;
+/**
+ * Resolves the Fusion `appKey` by running `ffc app manifest --silent`
+ * in the given application directory and parsing the JSON output.
+ *
+ * @param appDir - Absolute path to the Fusion application
+ * @returns The `appKey` string from the manifest
+ * @throws {Error} When `ffc app manifest` fails or returns invalid JSON
+ */
+export declare function resolveAppKey(appDir: string): string;
+/**
+ * Polls a server URL until it responds with a non-5xx status or the
+ * timeout is reached.
+ *
+ * @param url - The URL to poll (e.g. `http://localhost:3000`)
+ * @param timeoutSeconds - Maximum number of seconds to wait
+ * @returns `true` if the server responded before timeout, `false` otherwise
+ */
+export declare function waitForServer(url: string, timeoutSeconds: number): Promise<boolean>;

package/dist/types/version.d.ts

@@ -0,0 +1 @@
+export declare const version = "1.0.0";

package/package.json

@@ -0,0 +1,57 @@
+{
+  "name": "@equinor/fusion-framework-cli-plugin-copilot",
+  "version": "1.0.0",
+  "description": "Copilot SDK powered evaluation plugin for Fusion Framework CLI",
+  "type": "module",
+  "main": "dist/esm/index.js",
+  "types": "dist/types/index.d.ts",
+  "exports": {
+    ".": {
+      "import": "./dist/esm/index.js",
+      "types": "./dist/types/index.d.ts"
+    }
+  },
+  "typesVersions": {
+    "*": {
+      ".": [
+        "dist/types/index.d.ts"
+      ]
+    }
+  },
+  "keywords": [
+    "fusion-framework",
+    "cli",
+    "plugin",
+    "copilot",
+    "copilot-sdk",
+    "agent-browser"
+  ],
+  "author": "",
+  "license": "ISC",
+  "publishConfig": {
+    "access": "public"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/equinor/fusion-framework.git",
+    "directory": "packages/cli-plugins/copilot"
+  },
+  "dependencies": {
+    "@github/copilot-sdk": "^0.1.32",
+    "chalk": "^5.6.2",
+    "commander": "^14.0.1",
+    "ora": "^9.3.0",
+    "rxjs": "^7.8.2"
+  },
+  "peerDependencies": {
+    "@equinor/fusion-framework-cli": "^14.0.0"
+  },
+  "devDependencies": {
+    "typescript": "^5.9.3",
+    "@equinor/fusion-framework-cli": "^14.0.0"
+  },
+  "scripts": {
+    "build": "tsc -b",
+    "build:types": "tsc -b"
+  }
+}