essay-draft-from-research

Essay Draft from Research

Interview the user to produce a focused essay draft grounded in their research directory. The output is a single markdown file the user (or an editor) can read end-to-end as a first draft — not an outline, not a research summary — with enough of the user's voice and angle baked in that the next pass is revision, not generation.

Why this skill exists

Drafting from a directory of notes usually fails in one of two directions: either the draft regurgitates the sources without finding an angle, or it asserts an angle the research doesn't actually support. Both happen because the draft gets written before the writer has interrogated the corpus with their own thesis in hand. The interview is what fixes that — it pulls the angle out of the writer, anchored on what's actually in the documents, before any prose gets written. The invariant: read the research first, sharpen the angle in conversation, then draft.

Workflow

The sequence matters. Do not jump ahead.

1. Confirm the seed intent in one sentence

Before opening any document, get the user's intent in one sentence: what essay do they want to draft, and where is the research?

If they already named both clearly in the prompt, paraphrase and confirm — don't make them repeat themselves. If the topic is vague ("something about my notes on AI") or the research path is missing, ask one targeted question to nail down (a) the working topic or angle and (b) the path to the research directory. Reading the wrong directory is wasted work.

2. Strategic research-directory scan to seed context

The grounding read is steered by the seed, not exhaustive. Skim what the topic makes relevant; ignore what it doesn't. The goal is to know enough to ask sharp anchored questions in step 3 — not to internalize every document. Deep reading of specific sources for quoting and evidence happens after the interview locks the angle (see step 7).

Specifically:

• Run ls (or Glob **/*.{md,txt,pdf,docx}) on the research directory to inventory what's there. Note file counts, formats, and any obvious grouping (subdirs, naming patterns).
• Read 3–5 documents end-to-end — the ones whose titles or filenames look most central to the seed topic. If filenames are uninformative, read the 3–5 most recently modified.
• Skim the rest — first paragraph and any headers — to register what kinds of material exist (interviews, articles, raw quotes, the user's own notes, primary sources).
• Extract signals: what claims or threads recur across documents; where sources contradict each other; what's a strong quote vs. paraphrasable; the user's own voice if their notes are present; obvious gaps the seed topic might require.
• If the directory is large (>30 files) or unfamiliar, spawn an Explore subagent in parallel rather than reading serially. Tell it: "Inventory the research directory at <path> for an essay on <topic>. Identify (a) the 5 most central documents, (b) recurring themes, (c) contradictions or tensions across sources, (d) the strongest quotable passages, (e) obvious gaps. Report under 400 words."