pre-reading-briefing

Use when entering the review/revision phase of a draft. Generates a pre-reading briefing for each section—prerequisites, main claims, and anticipated objections—before the author or reviewers read thr

SKILL.md

GitHub

Pre-Reading Briefing Skill

Purpose

"Writing" a draft and "reading & verifying" it impose different cognitive loads. Even when drafting is accelerated, the author or reviewers need separate effort to read the finished manuscript and understand its terminology, assumptions, and line of argument. This skill generates a per-section "how to read this" guide before entering the review phase (PH-REV), providing reading-side scaffolding.

In cognitive-scaffolding terms, this skill primarily extends M1 (Mirroring) to the comprehension side together with M4 (ZPD prompting): it reflects the reader's current position and incrementally presents the prerequisites for the next section to be read.

Trigger Conditions

  • A chapter or section draft is complete, before the author's own read-through/verification
  • As a reader's guide before sharing a draft with reviewers or collaborators
  • After a long writing hiatus, when one's own draft reads like someone else's text
  • Before post-hoc audits such as claim-evidence-gate, to narrow the scope of human verification

Prerequisites

  • The target draft files (chapter/section) exist
  • Ideally, domain concept definitions (glossary) exist under design/ (used for concept grounding)

Workflow

Step 1: Determine scope

  1. Agree with the user on the briefing target (chapter, sections, paragraph groups).
  2. Confirm the reader model (Step 2 below).

Step 2: Select the reader model

Depth of explanation and handling of assumed knowledge depend on the reader role.

Reader model Assumption Briefing policy
Author (re-reading) Time has passed since writing; details forgotten Re-present "why this structure" and "original intent"
First-time reviewer Domain expert, but unfamiliar with this paper's setup State definitions and argument structure upfront
Novice / practitioner Limited background knowledge Explain prerequisites plainly; mark expert sections as "skippable"

If the reader model is unclear, always ask the user one question before generating (do not guess).

Step 3: Generate per-section briefing

For each section, present the following before reading:

Element Content
Purpose of this section One sentence; show the mapping to the research question (RQ)
Prerequisite concepts & definitions Up to 3 technical terms used in the text; reference domain concept definitions if available
Main claim The single most important point of the section, in 1–2 sentences
Anticipated objection / caveat One point a reader is likely to question; indicate whether a counter-argument exists
Comprehension check question Typed (see below). Do not repeat what the Purpose/Claim lines already say.

Target 5–10 lines per section. Do not produce a detailed summary (this is a reading scaffold, not a copy of the text).

Question types and who answers (v1.1.0)

The agent tags questions at generation time. If ambiguous, default to TERM. The author may retag at any time; do not treat retagging as an extra decision. Typed questions apply only when the reader model is Author (re-reading).

Tag Asks Who fills it
TERM Meaning of a term or concept Author; one question at a time (Capture→Mirror→Confirm)
VALID Manuscript vs an external source of truth (version, protocol) Agent drafts; author adopt/rejects
MAP Alignment between internal artifacts (table↔section, definition↔observation) Same
GAP Optional MAP subtype: gap vs chapter/observation Same

Intent questions ("why cite this source?") are not a fourth type. If the answer is in a design doc, tag MAP; otherwise TERM.

VALID / MAP draft rules (P0)

  • Drafts may be generated in one pass. Confirm 2–3 rows per turn. Do not offer accept-all.
  • Every cell must cite the join evidence (table row, section number, short excerpt). Do not emit a draft without evidence.
  • MAP values: match / mismatch / unmatched (missing from table or section).

TERM cap: 3 questions per session. Prefer concepts the author marked unknown. Park the rest via session-research-handoff. Not answering every question is not a failure. Unknown TERM → explain; unknown MAP → draft.

Step 4: Placement & delivery

  • Confirm with the user whether to insert the briefing as a "Pre-Reading Briefing" block at the head of the chapter, or to output it as a separate file (e.g., docs/<paper-id>/design/reading-brief-chapterN.md).
  • If inserted into the manuscript, mark it clearly as a temporary review artifact; recommended practice is to remove it from the final/submission version.

Step 5: Post-reading feedback

Do not ask the user to answer every question. Route by type.

  • TERM: at most 3 unknowns the author named; explain then Confirm.
  • VALID / MAP: emit evidenced drafts; adopt/reject 2–3 rows per turn.
  • Record outcomes (answered / explained / draft-adopted / rejected) in the briefing answer column.
  • Feed adopt/reject and remaining unknowns into the next revision. Exhaustive audit belongs to claim-evidence-gate (this skill's VALID is an author-facing prompt, not a gate).

Separating production/comprehension metrics (recommended)

When operating this skill, record writing time (production side) and briefing generation + read-through + check-question time (comprehension side) as separate metrics. This makes the "writable but not readable" asymmetry quantitatively observable.

Outputs

  • docs/<paper-id>/design/reading-brief-chapterN.md (when managed as a separate file)
  • Or a temporary block inserted at the chapter head (removed from the final manuscript)

Output template

## Pre-Reading Briefing: Chapter N — Title
**Target reader**: Author (re-reading)
**Estimated read time**: ~X minutes

### §N.1 Section title
- **Purpose**: For RQ1, establishes that ...
- **Prerequisites**: Term A (= definition ...), Term B
- **Main claim**: ...
- **Anticipated objection**: "But what about ...?" → addressed in §N.2
- **Check question** `[TERM]`: Can you state this section's conclusion in one sentence?

Relations to other skills

  • session-research-handoff: Combines well with warm-up questions when resuming a session
  • claim-evidence-gate: VALID here is an author-facing reading aid. The gate is the exhaustive machine audit. Do not make the author fill the same table twice.
  • scholarly-concept-modeling: Source of prerequisite concept definitions
  • academic-paper-translation: Applicable as comprehension support for foreign-language literature

Cognitive scaffolding

All interactions under this skill follow the Cognitive Scaffolding Principles (S1–S4). In particular, S1 (mirroring) is extended to the comprehension side: always reflect where the reader is now and what prerequisites they need next.

Disclaimer

The generated briefing is a secondary artifact for comprehension support; it does not guarantee the correctness of the paper. Final verification of content must always be performed by the author against the manuscript itself.