---
alwaysApply: false
globs: ["**/*"]
description: "The only skill that speaks about depth. Use whenever you address the person, explain a gate, or summarise an artifact."
---

# Depth Projection

Depth changes the **envelope** around your work. It never changes the artifact
and never changes which gates apply. One artifact, one gate set, three
envelopes.

**Resolve the depth level once.** Read `user_skill_level` from the project's
BMAD config. Map it: `beginner` to A1, `intermediate` or `expert` to A2,
`non-technical` to A3. Anything absent or unrecognised is A2. Never ask twice,
and never block on it. If nothing is recorded and the turn does not tell you,
ask exactly one question, phrased by what the person *does*: "Do you write
software yourself, or should I do the technical work and explain the decisions
in plain words?" Never use the words depth, level, tier, mode or persona with
the person.

**Fill only these four slots.** They are the whole envelope. You may fill a
slot; you may never rename one, add one, or invent a fifth.

- `gate-preamble` — before a gate applies. Exactly three parts: what you are
  about to do, restated in the person's world; what goes wrong if you skip it,
  as a concrete consequence with a horizon; what you need from them, as a
  decision with a proposed default or an acknowledgement.
- `artifact-summary` — after an artifact exists. What it is, what decision it
  locks in, what changes if it turns out wrong.
- `decision-frame` — when a choice needs the person. Restate the decision
  inside their own domain where you can. Where you cannot, recommend a default,
  state the consequence, and record the choice as theirs.
- `narration` — the skills you loaded, before you act.

**Which slots each depth fills.**

- **A1** fills `gate-preamble` short, plus `artifact-summary`. They can read the
  artifact, so teach the *why* once and move on.
- **A2** fills `narration` and `artifact-summary`. They know engineering, so
  give them the process; they still need to hear what the work produced.
- **A3** fills all four, every time, and none of them is optional.

**`artifact-summary` is never skipped.** Every level fills it, because a
report on finished work is a summary at every level. The level changes how
much you explain, never whether you report.

**A3 register.** Do not restate plain-language rules here. Route every A3
envelope through the installed plain-language and readability skills, and treat
their measured targets as the acceptance criterion for the wording.

**A3 narration.** Never list skill slugs. Give one sentence per loaded skill
saying what that rule will make happen, in the person's terms, then put the
slugs on a single trailing line so the work stays auditable. The sentence count
must equal the slug count, so a dropped sentence is visible.

**A3 gate denials.** A refusal reaches the person as one plain sentence that
says you stopped yourself, that nothing is broken, and what you need. The
machine-facing reason never reaches their screen.

**What depth must never do.** It must never skip a gate, reorder a phase, or
change an acceptance criterion. If reducing friction seems to require that, the
answer is no: gate bypass is measured at every depth and must stay at zero.