Plain mode

Plain mode lets someone who does not write code direct the Alfred fleet in plain language. A teammate can describe a change the way they would say it out loud, for example “make the signup button on the welcome screen green”, and the planning assistant turns that into the same structured work item the fleet always runs on. The person never sees a spec, a repository name, a readiness score, or a pull request. They describe an outcome, answer at most a question or two, and approve a preview.

Source: lib/intake_profiles.py and the full note in docs/PLAIN_MODE.md.

The idea

The agent does spec-driven development for you. You approve outcomes, not code.

Alfred’s quality bar is unchanged: every piece of work still becomes a structured draft with a problem statement, desired behavior, scope, checks, and a verification plan, and it still passes every existing gate before anything ships. Plain mode only changes who has to think in those terms. In the default (technical) mode, you do. In plain mode, the assistant does that work invisibly and talks to the person entirely in plain language.

Turn it on

Set one environment variable on whatever surface is talking to the person:

export ALFRED_INTAKE_PROFILE=plain

• Unset (or any unrecognized value) keeps the original technical behavior, unchanged. A typo never silently downgrades the default technical mode into plain mode.
• plain switches the planning assistant’s conversational surface to the non-technical front door.

The variable only affects two things:

1. The clarifying-question persona. When an LLM refiner is enabled, plain mode gives it a friendly, no-jargon persona. It asks at most one or two short, plain questions (“Which screen is this on?”, “What color did you have in mind?”) and never uses words like spec, acceptance criteria, repository, readiness, pull request, or diff.

2. The user-facing summary. Instead of an amendment count and a readiness verdict, the person sees a short plan:

   Here’s what I’ll do:

   • Make the signup button on the welcome screen green so it stands out.

   I’ll put this together and show you a preview to look over before anything goes live.

   OK to go ahead?

Everything else, the structured draft, readiness scoring, the GitHub-ready issue body, the development spec, and the whole downstream bridge and fleet, is identical in both modes. Plain mode is a thin strategy seam, not a separate code path.

Where it fits

Plain mode is the friendly entrance; the existing pieces do the rest.

• Slack listener. A teammate DMs Alfred describing what they want. With plain mode on, Alfred replies with plain questions and a plain plan instead of internal commands and readiness scores.
• Alfred Desktop. The Compose box becomes a plain-language intake.
• The fleet. Receives the exact same structured work it always has, and keeps every gate (claim-lock, spend caps, review, approval) intact.

The non-technical user approves an outcome and later reviews a preview. They never touch code, specs, or GitHub.