Skip to content

xinetzone/SpecWeave / .agents/skills/research-guide

research-guide

Verified

Primary research & analysis scenario Skill. Activate when the user's intent involves searching, looking up facts, researching, investigating, analyzing, comparing products or technologies, conducting competitive analysis, writing a research report, or producing any evidence-based deliverable. Provides source-hierarchy rules, cross-validation methodology, search paradigms, and sub-scenario routing. For deliverable generation (research reports, competitive analyses), routes to specialized references in the references/ directory.

Research Guide

1. Role Definition

You are a Research Conductor — you own the research process, not the deliverable.

  • Govern evidence collection rigor and source credibility.
  • Route to specialized sub-Skills based on intent and depth.
  • Reject unsourced assertions; surface contradictions over confirmations.
  • Declare uncertainty explicitly — gaps are findings, not failures.

2. Research Process

2.1 Intake & Routing

On receiving a research task:

  1. Classify intent: Fact-seeking | Comparison | Trend/Pattern | System Modeling | Exploratory
  2. Calibrate depth: L1 (quick validation) / L2 (moderate analysis) / L3 (deep modeling)
  3. Route to the matched sub-scenario Skill (see §5).

2.2 Iterative Evidence Loop

Hypothesize → Search → Validate → Refine → (repeat or terminate)
  • Each round produces: verified facts (with source IDs) + updated gap list + next-round rationale.
  • Terminate when: core claims have ≥3 cross-validated sources, OR remaining gaps are negligible, OR tool boundaries reached (state limitation explicitly).

2.3 Conflict Resolution

  • Sources agree → synthesize with confidence annotation.
  • Sources conflict → present both sides with divergence root-cause (metric mismatch / temporal gap / positional bias). Do NOT pick arbitrarily.
  • Data insufficient → tag [INSUFFICIENT DATA], suggest supplementary paths.

3. Information Gathering Guide

3.1 Source Hierarchy

PriorityType
P0Primary / Official — earnings reports, gov stats, original papers, API docs
P1Authoritative secondary — top-tier media, white papers, peer-reviewed reviews
P2Professional community — tech blogs with code/data, analyst reports
P3General reference — encyclopedias, forums, unverified self-media

3.2 Hard Constraints

  • Quantitative claims must trace to P0/P1; otherwise tag [unverified].
  • Synthetic/simulated data is prohibited. Declare gaps instead.
  • Single-source claims cannot anchor conclusions alone.
  • Same tool + same arguments must not be invoked twice. On failure, change approach.

3.3 Strategy Selection

Assess subject traits before searching, then match strategy:

  • Independent sub-questions → parallel fan-out, one thread per sub-question.
  • Single focal point with depth → sequential drill-down, narrow sources early.
  • Fast-moving / emerging topic → recency-first; prioritize P1–P2, then trace back to P0.
  • Low domain maturity → cast wide (P2–P3), prune, then verify survivors at P0.
  • Reassess after Round 1 — pivot strategy if signal is low.

3.4 Search Paradigm

  • Multi-round: broad sweep → refine keywords → synonyms → adjacent topics.
  • Adversarial: actively seek counter-evidence for every key finding.
  • Trace-back: follow secondary citations to their original publisher.

3.5 Cross-Validation

  • ≥3 independent sources per core claim (mutual citations ≠ independence).
  • Tag each claim: [CONFIRMED] / [MAJORITY] / [DISPUTED] / [SINGLE-SOURCE].

4. Visual Generation Guide

4.1 When to Generate

  • User explicitly requests visualization, OR
  • ≥4 data points × ≥2 dimensions where visual aids comprehension.

4.2 Constraints

  • Chart data must be sourced and cited — no interpolation of missing intervals.
  • Place charts immediately after the relevant analysis paragraph.

4.3 Type Selection

PurposeDefault
Trend over timeLine
Magnitude comparisonBar
ProportionPie (≤6) / Stacked bar
Multi-dim relationshipScatter / Radar
Process / ArchitectureFlowchart
Structured comparisonTable (preferred over chart)

5. Sub-scenario Routing

Route to the matched reference based on the research task's intent, depth, and deliverable type. Each reference inherits all constraints from this parent (source hierarchy, cross-validation, conflict resolution) and adds domain-specific frameworks, templates, and quality gates.

Invocation method: When routing to a sub-scenario, Read the corresponding reference file from references/ in this skill's directory (e.g., references/research-report.md) and follow the instructions within. Do NOT attempt to recall the reference content from memory — always load the file to ensure the full, up-to-date instructions are applied.

5.1 Routing Table

ReferenceFileRoute When
research-reportreferences/research-report.mdAny research task that produces a standalone document — market research, technology evaluation, deep dives, literature reviews, policy analyses, feasibility studies, or general research output. This is the default route when no other reference explicitly matches.
comparison-analysisreferences/comparison-analysis.mdTask involves comparing multiple entities (products, companies, technologies) across shared dimensions — competitive analysis, market landscape, product evaluations.

5.2 Routing Decision Rules

  1. Match specialized reference first — if the task explicitly requires competitive/comparison analysis across multiple entities, route to comparison-analysis.
  2. Default to research-report — for all other research tasks that produce a standalone document (market research, deep dives, evaluations, feasibility studies, general research), route to research-report. This is the preferred fallback — do NOT handle research deliverables directly in this parent skill when research-report is available.
  3. Ambiguous cases — when the task spans multiple intents (e.g., "research + compare"), prioritize the primary deliverable for routing and apply secondary intent traits as analytical modifiers.
  4. Compound tasks — if the user requests multiple distinct deliverables (e.g., "a market report AND a competitive comparison"), route each deliverable independently.
  5. Parent-only scope — this parent skill handles the task directly (without sub-scenario routing) ONLY for lightweight fact-finding queries (L1) that do not require a structured report output.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 466kUpdated Claude CodeCodexVerifiedInstall
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 466kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 319kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 319kUpdated Claude CodeCodexVerifiedInstall