deep-reading-analyst

Wing-1 deep-thinking skill for articles, papers, opinion pieces, case studies, and long-form decision documents. Routes the user's content through 8 thinking frameworks at four depth levels (Quick / Standard / Deep / Research) and returns insight tied to the user's goal, not framework completion.

When to use

• User pastes an article URL, paper, or long text and wants depth ("analyze", "deep dive", "extract insights", "help me understand").
• User asks for a specific framework ("apply SCQA to this", "use inversion thinking", "give me the mental models lens").
• User is making a decision and wants pre-mortem / multi-lens analysis on a written proposal.
• User is studying or note-taking on dense material (research papers, strategy memos, books).

Do NOT use when:

• User wants a 3-bullet TL;DR — use agent-docs-writing or write a direct summary.
• Content is code or a diff — route to judge-bug-hunter, judge-code-quality, or adversarial-review.
• User wants risk analysis on a code change — route to adversarial-review (diff-bound) or threat-modeling.
• User wants debugging or incident analysis — route to bug-analyzer or systematic-debugging.

Framework Arsenal

Depth	Time	Frameworks	Reference module
L1 — Quick	~15 min	SCQA, 5W2H	scqa-framework, 5w2h-analysis
L2 — Standard	~30 min	L1 + Critical Thinking, Inversion	+ critical-thinking, inversion-thinking
L3 — Deep	~60 min	L2 + Mental Models, First Principles, Systems Thinking, Six Hats	+ mental-models, first-principles, systems-thinking, six-hats
L4 — Research	120 min+	L3 + Cross-source comparison via web search	+ comparison-matrix

Every framework module is a project-local guideline under docs/guidelines/agent-infra/. Nothing is loaded from upstream at runtime.

Procedure: deep-reading-analyst

Step 0: Inspect

1. Detect content type — article, paper, opinion piece, case study, how-to, strategy memo. Drives auto-suggested frameworks (Step 1).
2. Detect goal signal — problem-solving, learning, writing reference, decision-making, curiosity. Drives Step 4 output shape.
3. Skip if mismatched — see "Do NOT use when" above; route to the named skill.

Step 1: Initialize Analysis

Open with ONE decision point (per ask-when-uncertain — a single numbered-options block answerable with a single number is one question; never three separate asks):

1. Standard read — L2 depth, auto-selected frameworks (default).
2. Quick skim — L1 depth, auto-selected frameworks.
3. Deep dive — L3/L4 depth, auto-selected frameworks.
4. Custom — name your goal (problem-solving · learning · writing · decision-making · curiosity), depth level, and/or specific frameworks in your reply.

Depth levels are scope descriptors, not wall-clock promises (per direct-answers — no duration estimates): L1 skim · L2 standard · L3 deep · L4 exhaustive-research.

If the user does not answer, default to L2 Standard with auto-selected frameworks.

Auto-suggest matrix:

Content type	Default frameworks
Strategy / business article	SCQA + Mental Models + Inversion
Research paper	5W2H + Critical Thinking + Systems Thinking
How-to guide	SCQA + 5W2H + First Principles
Opinion piece	Critical Thinking + Inversion + Six Hats
Case study	SCQA + Mental Models + Systems Thinking

Step 2: Structural Understanding (always run)

Regardless of depth, open with two short blocks:

2A — Basic structure.

📄 Type: [article/paper/report]
🎯 Core thesis: [one sentence]
Structure:
├─ Argument 1 → [key support]
├─ Argument 2 → [key support]
└─ Argument 3 → [key support]
Key concepts: [3–5 terms with one-line definitions]

2B — SCQA breakdown. Apply the four-element decomposition from scqa-framework:

**S (Situation)**: [context the article establishes]
**C (Complication)**: [problem identified]
**Q (Question)**: [core question]
**A (Answer)**: [main solution / conclusion]
Structure quality — clarity / logic / completeness: [★★★☆☆]

2C — 5W2H gap check (L1+). Quick scan: which of What, Why, Who, When, Where, How, How much are well-covered, partial, or missing? Flag the 1–2 most critical gaps.

Step 3: Apply Frameworks (depth-gated)

Load only the frameworks the user's depth bought. Each framework follows the same pattern: load reference module → apply lens → produce one fixed-shape block.

L2 additions:

• Critical Thinking (critical-thinking) — argument strength score (X/10), strengths, weaknesses, logical fallacies detected.
• Inversion (inversion-thinking) — pre-mortem on the article's recommendation: "how would this fail?", missing risk factors, mitigations.

L3 additions:

• Mental Models (mental-models) — pick 3–5 models from different disciplines (physics, biology, psychology, economics, math), apply each lens, surface cross-model patterns.
• First Principles (first-principles) — strip to fundamental truths, validate each core assumption, rebuild from base.
• Systems Thinking (systems-thinking) — map relationships, feedback loops, leverage points.
• Six Hats (six-hats) — White (facts), Red (feelings), Black (cautions), Yellow (benefits), Green (creativity), Blue (process).

L4 addition:

• Cross-source comparison (comparison-matrix) — web-search 2–3 related sources, compare SCQA across sources, identify consensus vs divergence, synthesize integrated perspective.
• Disconfirmation + per-part grounding (pre-synthesis gate) — run at least one search that tries to falsify the leading takeaway (rule alternatives in/out, not only confirm), and ground each part of the reading goal in a retrieved source before writing it. Distinct from verify-before-complete (completion evidence).

Step 4: Synthesize by Goal

Output shape is driven by Step 0 goal, not by frameworks applied. Pick exactly one of the four blocks below.

For problem-solving: Applicable solutions (2–3 from content) → Application plan with timed action steps → Success metrics → Risk mitigations from Inversion.

For learning: Core concepts (definition + example) → Mental models gained → Connections to prior knowledge → First Principles fundamental question → 3 verification questions (understanding / application / evaluation).

For writing reference: Key arguments + evidence (with paragraph citations) → Quotable insights with context → Critical analysis (strengths for citing, limitations for balanced discussion) → Alternative perspectives from Mental Models → Gaps and counterfactuals.

For decision-making: Options presented → Multi-model evaluation (economic + risk + systems lens) → Six Hats decision analysis → Scenario analysis (best / worst / most likely) → Synthesized recommendation.

Step 5: Knowledge Activation (always end here)

Regardless of goal, close with three fixed blocks:

## 🎯 Top 3 takeaways
1. **[Insight]** — Why it matters: [...] · One action: [specific, time-bound]
2. **[Insight]** — Why it matters: [...] · One action: [specific, time-bound]
3. **[Insight]** — Why it matters: [...] · One action: [specific, time-bound]

## 💡 Quick win — one tiny, specific action for the next 24 hours.

## 🧭 Frameworks used
✅ SCQA  ✅ 5W2H  ✅ Critical  ✅ Inversion
□ Mental Models  □ First Principles  □ Systems  □ Six Hats

Step 6: Validate

1. Every claim is faithful to the source — no misrepresentation, facts distinguished from opinions.
2. Frameworks applied purposefully, not force-fit — drop a framework that adds no insight rather than padding the output.
3. Output ends with concrete, actionable steps — no analysis-without-application.
4. Specific citations (paragraph numbers, quotes) where the source supports them.

Output format

1. Structural block (Step 2) — type / thesis / structure tree / key concepts / SCQA / 5W2H gaps.
2. Framework blocks (Step 3) — one fixed-shape block per framework the depth bought.
3. Goal-shaped synthesis (Step 4) — problem-solving / learning / writing / decision-making.
4. Knowledge activation (Step 5) — top 3 takeaways · quick win · frameworks-used checkboxes.

Gotcha

• The model tends to apply every framework even at L1 — respect the depth budget; skip frameworks the user did not buy.
• The model tends to summarize instead of analyze when the user pastes long text — go deep on 1–3 points, not shallow on all of them.
• Inversion drifts into adversarial code review — this skill targets decisions and arguments, not diffs. Route diff stress-tests to adversarial-review / judge-bug-hunter.
• Mental Models drifts into name-dropping — pick 3–5, apply each lens concretely to the article's claims, drop models that yield no new insight.
• L4 cross-source comparison drifts into a literature review — keep it to 2–3 sources, focus on consensus / divergence / unique value.
• Output without action steps is a failure mode — every Step-4 synthesis must end with timed, concrete actions tied to the user's goal.

Do NOT

• Do NOT force-apply all frameworks at the user's chosen depth — drop ones that add no insight.
• Do NOT copy text verbatim from the source — always reword for the user's understanding; any exact quote follows content-quoting-floor (≤15 words, one quote per source).
• Do NOT restate a fresh-lookup-class claim (current status, version, price, date) from model memory — verify against a live source or flag it as unverified, per asking-and-brevity-examples § Volatile-fact freshness.
• Do NOT use academic jargon without one-line definitions in the "key concepts" block.
• Do NOT skip Step 5 — the takeaways + quick win are the load-bearing output, not optional decoration.
• Do NOT route code reviews, diff stress-tests, or incident debugging through this skill.

Reference modules

All 8 framework modules are project-local guidelines under docs/guidelines/agent-infra/ (full text adopted under the Reference-Guideline Sunset Policy). Each carries an ## ADOPT citation footer pinning its MIT upstream origin.

• scqa-framework — situation / complication / question / answer decomposition.
• 5w2h-analysis — completeness check (7 questions).
• critical-thinking — argument quality / fallacy detection.
• inversion-thinking — pre-mortem on decisions, distinct from adversarial-review.
• mental-models — Munger's multi-discipline toolkit.
• first-principles — fundamental-truth extraction.
• systems-thinking — feedback loops + leverage points.
• six-hats — White / Red / Black / Yellow / Green / Blue protocol.
• comparison-matrix — cross-source synthesis (L4 only).

Deprecated — on-demand upstream loading. Earlier revisions of this skill described loading some modules "on demand from a SHA-pinned URL", backed by an external_source frontmatter pin. That capability was never wired (no fetcher ever consumed the pin; it was schema-only metadata), so the pin and its refresh_trigger / sunset_criterion bundle have been removed — every module is local, no runtime fetch is needed. If a future revision genuinely needs live upstream content, build an explicit fetch step; do not reintroduce a non-functional metadata pin.