---
name: "skill-message-outline"
type: skill
description: Turns an approved message brief into an outline whose headings carry the argument when read alone, choosing the framework by purpose and reader from convention-comms-frameworks. Three passes — Select, Outline, Check. Pass 3 reads a finished draft back against the brief and names where it departs. Refuses to start without an approved brief and says which of its fields are missing rather than inventing them. Names no format, so the same outline becomes an email, a document or a deck without the argument changing. Hands on an approved outline carrying the framework and the reason for it, headings written as claims, and a takeaway and named evidence on every line, which the format wrappers assemble and any reviewer checks against. Read before any writing starts, and before reviewing a draft someone else wrote. Surfaced by APP-2331; the brief it works from is Atlas's, APP-2332.
license: CC BY-NC-SA 4.0
---

# narrative

The argument before the prose. Written output has been drafted straight from a request, so the case it makes is discovered during the writing and changes with each revision — this skill is the step that stops that. It takes an approved message brief, picks the framework the purpose calls for, turns it into an outline whose headings carry the argument on their own, and can read a finished draft back against the brief to say where it departed.

The frameworks themselves live in `convention-comms-frameworks`; this skill is the behaviour that selects, applies and checks them. It is **Layer 1** and format-agnostic: it names no format, and the same outline it produces is assembled into an email, a document or a deck by a thin wrapper downstream without the argument changing (APP-2331).

## Trigger

Three ways in:

* **Before any writing starts**, on anything with a reader and a purpose. This is the standing case. An approved brief and no outline is the normal state this skill is called from.
* **On a finished draft** whose structure is in question — one that is not landing, one written elsewhere, one being reviewed before it goes out.
* **Called by a format wrapper**, which delegates the argument's shape here rather than re-deriving it, and takes the approved outline as its input.

## Behaviour

Three passes. Pass 1 and Pass 2 run together when authoring; Pass 3 runs alone when reviewing.

### Inputs — what it refuses to start without

**One input: an approved message brief** (`APP-2332`, Atlas's). The skill does not start without it, and does not write one for itself. The brief carries the point in one sentence, the reader, the ask, why now, the most obvious objection, the two or three strongest supports with their evidence, and any bad news with its owner and plan.

Where the brief is missing, or a field in it is empty, **say which field and stop.** Do not infer the reader, do not assume the ask, do not guess the evidence. A guessed reader produces a piece that is structurally sound and aimed at nobody, which is worse than an obviously unfinished one because it survives review. **No approved brief, no outline** — and the thing to hand back is the named gap, not a draft with the gap papered over.

Three fields are load-bearing enough to name individually:

* **The point in one sentence.** If it cannot be written, there is no piece yet, and no framework will supply one. A sentence a reasonable person could not disagree with is a topic, not a point.
* **The reader.** One named person, what they are measured on, what they already know, how they take information, what they usually push back on. This is what sets the altitude.
* **The ask.** What they do, who decides, by when, and what happens if nobody answers. An empty ask on a piece that needs one is a gap in the brief, not something to fill in while outlining.

### Pass 1 — Select

1. **Name the purpose** from the brief. The point usually names it: an ask is a decision brief, a number is an update, a choice is Options → Trade-offs → Recommendation, a finding is What / So What / Now What, something going wrong now is SBAR.
2. **Apply AAA first.** Set the **altitude** to the level the reader can act at — consequence, sequence or mechanism — and hold it for the whole piece. Confirm the **audience** is one named reader, not a category. Confirm the **action** is executable as written. AAA sits above the chosen structure rather than beside it, and everything below is written at the altitude set here.
3. **Match the purpose to a framework** using the selection table in `convention-comms-frameworks`. The table is held there, not here, so that adding a framework is a one-file change and the rationale sits beside the definition.
4. **Where the piece serves two purposes**, say so rather than blending. It is two pieces, or one piece with a stated primary purpose and the secondary demoted to an annexe. A blended structure serves neither.
5. **Check the framework's *where it falls short* field before committing to it**, not after. Every entry names what it has no rule for; if what it has no rule for is the thing this piece most needs, reselect.
6. **State the framework and the reason in one line at the top of the outline.** One line, plain, so the Owner can see which framework was chosen and why without reading the outline. Pass 3 and any reviewer read this line; an unstated framework cannot be checked, only guessed at.

### Pass 2 — Outline

1. **Read the selected framework's full entry** — steps, what it suits, where it falls short, overlap.
2. **Lay out the steps in the entry's order.** Never drop a step to fit the length. A dropped step is what every entry's *where it falls short* field describes, and the dropped one is nearly always the interpretive middle: SCQA's Question, What / So What / Now What's So What, the pyramid's evidence under each support.
3. **Hold the proportions, not a length.** Each step gets weight in proportion to the work it does for this reader. Scaling a piece up or down changes the proportions of the steps, never the set of them.
4. **Write every heading as a claim.** This is the discipline the whole outline is judged on, and it is universal across frameworks. A heading states what the section argues ("Hiring is the risk to Q4"), never what it is about ("Q4 update"). **The test: reading the headings top to bottom, and nothing else, must carry the whole argument.** If they read as a table of contents, the piece has a shape and no argument.
5. **Give every outline line a takeaway and named evidence.** Beside each heading, state what the reader should leave that section with, and the specific figure, quote, source or document that earns it — named, never "evidence to follow". Where a line has no evidence, **that is the finding**: either the claim is cut, or the evidence is commissioned before the gate. An outline of headings alone cannot be checked against a finished draft, which is how a piece comes back with the right spine and unsupported paragraphs.
6. **Place the most obvious objection** from the brief immediately after the claim it threatens. Not in an annexe, and not softened into a version that is easier to answer.
7. **Place the decision line** wherever a decision is asked for — decider, deadline, default if no answer — and check the default is one that will actually happen.
8. **Place any bad news plainly**, with its owner and its plan, at the altitude the reader can act on it. No framework here forces this, which is exactly why it is a step.
9. **Size the supporting group honestly.** Three is the usual shape; two or four is the right answer when three would mean padding or merging two distinct points into one.
10. **Cut.** Read the outline end to end as one argument. No figure, claim or phrase is carried in more than one section unless the repeat is a deliberate, stated recap. Report what was cut.
11. **Name no format.** No slide counts, no subject lines, no page breaks, no section lengths expressed in a format's units. If the outline cannot be read without knowing what it will become, it is not finished.

### The gate — outline approved before any writing

**No prose is written until the outline is approved. Two stages, always: argument first, words second.** The outline goes up with the framework named, the reason for it in one line, the headings as claims, and each line's takeaway and named evidence.

The gate sits here rather than at draft because an outline is cheap to argue with and a draft is not. Once the prose exists the conversation becomes about the prose, and a structural objection arrives too late to act on without discarding work. Writing ahead of approval is the failure this skill exists to prevent, and it stays a failure however finished the thinking feels.

### Pass 3 — Check (read-only)

A structural review of a finished draft against the brief and the framework. Read-only, plain language, never a rewrite — the same discipline as `skill-review-pr` and `skill-deck-parity`.

1. **Detect the framework**, from the step order and the opening and closing pattern. Where the piece names its own framework, verify it rather than accept it.
2. **Check the purpose against the framework actually used**, and flag any mismatch. A mismatch is a reselection, not a fix.
3. **Check the framework is carried to the end.** A structure abandoned two-thirds through is the normal way this breaks — the opening is deliberate and the close is whatever was left. Name the entry's *where it falls short* failure and say whether it happened.
4. **Check the draft against the brief, field by field**, and name where it departs:
   * Is the brief's point the piece's governing claim, or has it drifted?
   * Is it written at the reader's altitude, or at the writer's?
   * Is the ask present, executable, and carrying a decider, a deadline and a default?
   * Is the most obvious objection raised in the reader's own terms, and answered or conceded?
   * Did each support's named evidence survive into the draft, or is it asserted now?
   * Is the bad news still there, still plain, still with an owner and a plan?
5. **Check the headings still carry the argument alone**, and that nothing is repeated across sections without a stated recap.
6. **Write the verdict** in the format below.

## Verdict format

Reuses the `skill-review-pr` shape rather than inventing one — same two modes, same plain-language bar, same review-only boundary.

* **Sound** — a plain statement of the framework detected, that it fits the purpose, that it is carried to the end, and that the draft matches the brief. Name what was checked, so the review is visibly real. Note anything the framework does not cover.
* **Departures** — each one with: where it departs (the section), what the brief or the framework entry says, how the draft differs, and the **specific fix**. Not a rewrite; this pass reports and the author fixes.

Every verdict states four things: **framework detected, fit to purpose, where it departs, specific fix.** Plain language throughout — the Owner acts on it without opening the draft.

## What it hands on

One artefact: **the approved outline.** It carries

* the framework and the one-line reason it was chosen;
* the altitude, stated;
* the headings, written as claims, in the framework's step order;
* each line's takeaway and its named evidence;
* the decision line, where one is needed;
* the most obvious objection, placed;
* any bad news, with owner and plan.

That outline is the input a format wrapper assembles from, and the thing any reviewer checks the finished piece against. It names no format, so one approved outline serves an email, a document and a deck without the argument being re-decided per format.

## Reads

`convention-comms-frameworks` (the framework definitions and the selection table), the approved message brief (`APP-2332`), `convention-comms-owner` where the reader is the Owner, and the domain's own voice convention — `convention-mrp-voice` for Mr P, a venture's brand voice or `skill-content-voice` elsewhere. Structure comes from here; register comes from there, and the two compose without touching.

## Ownership and who reads it

**Skills carry no owner.** This one is read by the writing agent (`APP-2328`), by Atlas when checking that a brief is sufficient to outline from, and by any reviewer running Pass 3. Pixel is downstream and handles layout only where design is needed.

It is one skill rather than one per format, because the behaviour — pick a framework, apply it, check against the brief — is identical whichever format the piece ends up in. What differs is the framework selected and the voice it is written in, and both are already inputs: the framework by purpose via `convention-comms-frameworks`, the voice by domain. Splitting it by format would duplicate the loop to vary two things that are already parameterised, and would put the argument back inside the format, which is the arrangement APP-2331 exists to end.

## Guardrails

* **Never start without an approved brief.** Name the missing field and stop.
* **Never write the brief for the author.** Assessing whether a brief is sufficient is in scope; supplying its contents is not.
* **Never name a format.** No slides, no email, no document, anywhere in the outline or the verdict.
* **Never write prose before the outline is approved**, however finished the thinking feels.
* **Never invent evidence.** A line with no evidence is reported as a finding, not filled in.
* **Never soften the objection.** Answering a weaker objection than the real one signals that the real one was avoided.
* **Pass 3 reports; it does not rewrite.** The author fixes.
* **Never bluff a default.** If no honest default exists for the decision line, say the work stops and name what stops.
