writing-design

Writing Design

Craft design for a new piece of writing — applied before or while you draft, so the structure, momentum, and clarity are right the first time. Provenance lives in skill.json; this file is runtime routing only.

Produces: an intent-specific plan — outline-plan.md (structure), draft-scaffold.md (draft), or persuasion-plan.md (persuade) — naming the target reader, the heuristics applied, the guards honored, and the grounding sources.

Boundaries

Do NOT use to revise, copyedit, or diagnose an EXISTING draft (use writing-audit), to design a documentation system's IA, retrieval, or telemetry (use docs-design), or to gather and validate source material (use research).

Core principle

Decide what the reader needs before you polish how it reads. Most weak writing is a structure or audience problem wearing a sentence-level disguise: name the governing point and the target reader first, then draft.

Activation

• Bare invocation ("use writing-design", "help me write this", "start"): show a compact menu: mode choice (guided / autopilot / grill me?) and numbered intents from the router. Wait. No file inspection, no network calls, no writes.
• Concrete invocation with intent and genre inferable: skip to step 3.
• Ambiguous invocation: ask one — e.g., "Do you need structure (outline), a first draft, or a persuasive arc?" or "What genre — argument memo, technical doc, talk pitch, narrative, or general prose?"

Workflow

1. Pick intent. Load references/intent-router.csv. Match the prompt to structure, draft, or persuade. Ambiguous → ask once. (Revising or critiquing an existing draft instead? That is writing-audit.)
2. Pick genre. Load the intent's CSV from references/intents/<intent>.csv. Match to one genre — argument-memo, technical-doc, talk-pitch, narrative, or general-prose. Not every genre is a target for every intent; the intent's CSV lists the genres it serves. If the genre you'd pick isn't a row, it isn't a target for that intent — route to the nearest served pairing and name it (e.g., making a how-to compelling is the narrative genre, not persuade). Ambiguous → ask once with the CSV menu.
3. Load grounded context. Load only the chosen row's playbook from references/playbooks/<genre>.md plus its core_refs. Do not load other playbooks.
4. Name the target reader using references/core/audience-frame.md — the piece is for a specific reader with a specific knowledge state and goal.
5. Apply the playbook heuristics tagged for this intent. Produce the concrete shape — an outline, a drafted pass, or a persuasive spine — not abstract advice. For a wide space, optionally dispatch reader-lens sketches (see "Subagent dispatch") and synthesize the strongest.
6. Honor the guards. When the work adds momentum or persuasion, apply references/core/narrative-honesty-guard.md — shape only where the tension is real. When it tightens prose, keep the writer's voice.
7. Emit output. Write the intent's template: templates/outline-plan.md, templates/draft-scaffold.md, or templates/persuasion-plan.md.

Wrong direction? If the user says this isn't what they meant, go back to Understand (step 1) — do not patch in the wrong direction. Restate the corrected understanding and re-plan.

Modes

Guided Draft (default), Autopilot, Grill Me — see references/modes.md. Offer the mode at bare invocation; default to Guided Draft on concrete invocations.

Output requirements

Every output names the target reader, the genre playbook(s) applied, the intent's load-bearing section (the outline / the drafted pass / the persuasive spine), the guards honored, and the grounding sources from skill.json.inspired_by.

Subagent dispatch

Optional for a wide persuade or structure space: dispatch reader-lens sketches in parallel — see references/subagent-dispatch.md — then synthesize the strongest. Skip for a single tightly-scoped piece.

Reference map

• references/intent-router.csv — level-1 router (structure / draft / persuade).
• references/intents/<intent>.csv — level-2 router (genre) per intent.
• references/playbooks/<genre>.md — genre playbooks (shared with writing-audit).
• references/core/*.md — shared method rubrics, plus the voice and narrative-honesty guards.
• references/subagent-dispatch.md — reader-lens dispatch and synthesis.
• references/modes.md — Guided Draft / Autopilot / Grill Me (shared).
• references/starter-scenarios.csv — named worked examples for bare invocation.
• templates/*.md — outline / draft / persuasion outputs.
• evals/ — activation cases, static checks, trigger evals.
• skill.json — provenance, grounding sources, version, status.