feat(build-cli): add flub ai command and AI-enabled (Insiders) devcontainer

[tylerbutler]

Summary

• Adds a flub ai command that launches an interactive Copilot SDK session to help users pick the right agent alias and MCP server configuration
• Creates a new AI-enabled (Insiders) devcontainer profile that builds the flub CLI from source during prebuild, making flub ai available in the codespace
• The base AI-enabled profile is unchanged — insiders differs by only 4 lines in devcontainer.json (name, onCreateCommand chains install-flub.sh, openFiles path)
• Shared files like launcher-prompt.md live in ai-agent/ only; flub ai searches ai-agent-insiders/ first then falls back to ai-agent/
• Adds FLUB_PERF=1 env var support for oclif performance diagnostics

Test plan

• Create a Codespace using the AI-enabled (Insiders) profile
• Verify flub ai is available and launches the interactive session
• Verify the base AI-enabled profile still works without flub ai
• Run flub ai -v and confirm verbose logging shows file resolution and model source
• Run DEBUG=oclif:perf FLUB_PERF=1 flub ai and confirm performance output

[tylerbutler]

Unfortunately devcontainers don't support inheritance, so we have to duplicate 99% of this file.

[tylerbutler]

😆 I did not request this, but... it's a little too cheesy. Prob should remove it. But it's hilarious to me that this UX trend influenced Claude.

[noencke]

Being the one that checked it in... I vote to keep it :D. I love it!

[tylerbutler]

Oh I didn't realize this wasn't new - my bad; in that case all statements retracted! :)

[tylerbutler]

Closing thread so claude doesn't get confused when reviewing feedback.

[github-actions]

🔗 No broken links found! ✅

Your attention to detail is admirable.

linkcheck output

> fluid-framework-docs-site@0.0.0 ci:check-links /home/runner/work/FluidFramework/FluidFramework/docs
> start-server-and-test "npm run serve -- --no-open" 3000 check-links

1: starting server using command "npm run serve -- --no-open"
and when url "[ 'http://127.0.0.1:3000' ]" is responding with HTTP status code 200
running tests using command "npm run check-links"


> fluid-framework-docs-site@0.0.0 serve
> docusaurus serve --no-open

[SUCCESS] Serving "build" directory at: http://localhost:3000/

> fluid-framework-docs-site@0.0.0 check-links
> linkcheck http://localhost:3000 --skip-file skipped-urls.txt

Crawling...

Stats:
  279037 links
    1880 destination URLs
    2126 URLs ignored
       0 warnings
       0 errors

[tylerbutler]

Make this a warning.

[tylerbutler]

Worth a formal type?

[Copilot]

Pull request overview

Adds an AI-assisted launcher experience to the flub CLI and introduces an “AI-enabled (Insiders)” devcontainer profile that prebuilds and links flub from source so flub ai is available in Codespaces.

Changes:

• Add new flub ai oclif command powered by the GitHub Copilot SDK, plus unit tests and command docs.
• Add “AI-enabled (Insiders)” devcontainer profile and an install-flub.sh helper to build/link flub during onCreateCommand.
• Add FLUB_PERF=1 opt-in to enable oclif performance diagnostics in bin/run.js and bin/dev.js.

Reviewed changes

Copilot reviewed 13 out of 16 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
scripts/codespace-setup/install-flub.sh	New script to build build-tools from source and npm link the flub CLI in devcontainers.
build-tools/pnpm-lock.yaml	Lockfile updates to include @github/copilot-sdk and its transitive deps.
build-tools/packages/build-cli/src/test/commands/ai.test.ts	Unit tests for alias validation and prompt-answer normalization helpers.
build-tools/packages/build-cli/src/library/ai/copilotSession.ts	Copilot SDK session wrapper (tool definition, preflight, streaming output, cleanup).
build-tools/packages/build-cli/src/commands/ai.ts	New flub ai command: resolves prompt/aliases, runs session, confirms, launches alias via bash.
build-tools/packages/build-cli/package.json	Add Copilot SDK dependency and register an ai topic.
build-tools/packages/build-cli/docs/ai.md	Generated/added docs page for flub ai.
build-tools/packages/build-cli/bin/run.js	Enable oclif perf diagnostics when FLUB_PERF=1.
build-tools/packages/build-cli/bin/dev.js	Same perf diagnostics toggle for dev runner.
build-tools/packages/build-cli/README.md	Adds flub ai to command topic list.
.devcontainer/ai-agent/launcher-prompt.md	New shared launcher prompt template (with model frontmatter).
.devcontainer/ai-agent/devcontainer.json	Adjust AI-enabled profile env/hook configuration (and removes prior postAttach open behavior).
.devcontainer/ai-agent-insiders/first-run-notice.txt	Insiders welcome notice mentioning flub ai.
.devcontainer/ai-agent-insiders/devcontainer.json	New insiders profile that chains install-flub.sh in onCreateCommand.
.devcontainer/ai-agent-insiders/GETTING_STARTED.md	Insiders getting-started guide with flub ai callout and alias explanations.
.devcontainer/README.md	Documents the new insiders profile and the shared/override file fallback strategy.

Files not reviewed (1)

• build-tools/pnpm-lock.yaml: Language not supported

[Copilot]

The header comment says this script follows the same install pattern as CI, but CI’s include-install-build-tools step runs pnpm i --frozen-lockfile without --ignore-scripts. Using --ignore-scripts here could skip required workspace lifecycle scripts and make pnpm build:compile/npm link fail or behave differently than CI. Consider aligning this install step with the pipeline template (or update the comment and document why scripts must be disabled here, plus what guarantees the build still works).

Suggested change

	# CI=true suppresses pnpm's interactive "purge modules" prompt in non-TTY environments.
	# --ignore-scripts skips postinstall hooks (which include nested pnpm installs that
	# conflict in a non-TTY environment).
	CI=true pnpm -C "$REPO_ROOT/build-tools" install --frozen-lockfile --reporter=default \
	--ignore-scripts
	# CI=true suppresses pnpm's interactive "purge modules" prompt in non-TTY environments
	# while keeping the install behavior aligned with CI.
	CI=true pnpm -C "$REPO_ROOT/build-tools" install --frozen-lockfile --reporter=default

[Copilot]

This change removes the previous post-attach behavior that opened the Getting Started guide in VS Code (postAttachCommand), and now leaves postAttachCommand empty. codespaces.openFiles only applies in Codespaces, so users attaching via VS Code/SSH may no longer get the guide automatically. If that’s unintended, restore an appropriate post-attach open (or add a VS Code customization that applies outside Codespaces).

Suggested change

	"postAttachCommand": {},
	"postAttachCommand": "if [ -z \"${CODESPACES}\" ] && command -v code >/dev/null 2>&1; then code -r .devcontainer/ai-agent/GETTING_STARTED.md; fi",

[Copilot]

PR description notes the base AI-enabled profile is unchanged, but this file is modified (env vars, lifecycle hook shape, and the postAttach behavior). If the intent is truly “insiders-only changes”, consider reverting the base profile edits or updating the PR summary so reviewers/users understand the base profile behavior has changed.

[Copilot]

When launching the selected alias, the execa("bash", ...) error handler exits with a derived exit code but does not log the failure reason. If bash is missing (e.g. Windows) or the shell command fails before producing output, the user will see "Launching ..." and then an unexplained exit. Consider surfacing the error (at least for spawn failures) via this.error(...)/this.errorLog(...) before exiting, while still preserving the underlying exit code.

Suggested change

	: 1;
	: 1;
	const errorMessage =
	error !== null &&
	typeof error === "object" &&
	"shortMessage" in error &&
	typeof (error as { shortMessage?: unknown }).shortMessage === "string"
	? (error as { shortMessage: string }).shortMessage
	: error instanceof Error
	? error.message
	: undefined;
	this.errorLog(
	errorMessage === undefined || errorMessage.length === 0
	? `Failed to launch ${proposal.alias}.`
	: `Failed to launch ${proposal.alias}: ${errorMessage}`,
	);