express-insight
关于
This skill translates a complex, multi-domain insight into a clear and actionable communication for a specific audience. It provides a structured method to choose the right format, convey the integrated understanding without oversimplifying, and invite constructive feedback. Use it after `integrate-gestalt` to effectively share cross-domain insights with specialists, generalists, or decision-makers.
快速安装
Claude Code
推荐npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/express-insight在 Claude Code 中复制并粘贴此命令以安装该技能
技能文档
Express Insight
Communicate multi-domain gestalt so it lands — preserve relationships, accessible to audience, honest where integration simplifies or risks distortion. Expression = final step synoptic cycle. Without → integrated remains private + unactionable. Challenge: language linear, insights not — provides structures for multi-dimensional communication without reducing to single dim.
Use When
- After
integrate-gestaltproduces cross-domain understanding to communicate - Finding spans domains, single-domain framing loses relationships
- Audience differs from perspective that generated insight
- Integrated feels clear internally but resists straightforward expression
- Decision depends on seeing how domains interact, not what each says independently
- Previous attempts met w/ confusion or domain-specific pushback
- Synoptic-mind team findings → stakeholders outside team
In
- Required: Integrated insight (output of
integrate-gestaltor equivalent) - Required: Audience (specialists, generalists, decision-makers, mixed)
- Optional: Form constraints ("must fit PR description", "verbal summary")
- Optional: Integrated domains (for attribution)
- Optional: Prior failed attempts (what didn't land)
Do
Step 1: Assess Audience
Who + what need. Same gestalt → 3 audiences = 3 forms.
- Primary:
- Specialists need domain accurate — reject oversimplification even if synthesis correct
- Generalists need big picture — relationships matter more than details
- Decision-makers need actionable implications + trade-offs — want what to do, cost, what if don't
- Mixed → layered: big picture first, domain depth specialists verify
- Existing mental model:
- What do they know each domain?
- Which connections new?
- What assumptions insight challenges?
- Trust requirement: how much justification before accepting cross-domain claim?
- Specialists trust insight respecting rigor
- Generalists trust insight making complexity navigable w/o oversimplify
- Decision-makers trust insight surfacing trade-offs honestly not hiding
→ Clear picture who, what need, what makes credible. Assessment influences every subsequent step.
If err: audience unknown/broad → default mixed-audience: big picture + domain depth on demand. "Everyone" less effective than specific, but better than wrong guess.
Step 2: Choose Form
Form determines what audience can perceive. Not decoration.
-
4 primary:
Form Structure Best for Narrative Story connecting — "X in A creates Y in B, means Z" Complex/novel, audience follows reasoning Diagram Spatial layout — nodes = contributions, edges = connections Structural, topology matters more than sequence Comparison table Each domain's perspective same issue parallel cols Analytical, verify each contribution indep Recommendation "Do X because A, B, C converge on Y, trade-off Z" Decision-makers, need act not understand -
Match form to insight:
- Causal chain across domains → narrative
- Structural relationships → diagram
- Convergence/divergence → comparison table
- What to do next → recommendation
-
Combine: recommendation backed by comparison, or narrative + diagram. But lead 1 primary — multiple formats → cognitive load obscures not clarifies
-
Medium constraints: verbal summary can't carry table; commit msg can't carry narrative. Adjust form not force content into incompatible container
→ Primary form (+ optional secondary) w/ rationale tied to audience + insight.
If err: no form feels right → insight not yet fully integrated. Return integrate-gestalt — difficulty expressing signals incomplete synthesis, not communication.
Step 3: Express Gestalt
Communicate in chosen form, note what integrates, where simplifies, what enables.
- State clearly — 1-3 sentences capturing core. Gestalt itself, not evidence.
- Name domains — explicit which contributed. Not credit — verification. Each = invitation: "check against your expertise."
- Mark simplifications — every multi-domain insight simplifies:
- Which nuances set aside?
- Relationships treated stronger/weaker than might be?
- What would specialist X want to add/qualify?
- State emergent value — what does this enable single-domain doesn't?
- What decision possible now not before?
- What risk visible hidden within individual domains?
- What opportunity appears at intersection no single domain owns?
- Maintain multi-domain texture — resist flattening into 1 domain's language. Integrates engineering + UX → use both. Connects research + ops → preserve both framings. Texture = insight.
- Sequence for comprehension — non-linear insight, sequential communication. Entry point giving best foothold: start where they know, bridge outward unfamiliar. First sentence determines lean in or tune out.
→ Insight audience understands, verifies vs expertise, acts on. Simplifications visible not hidden. Value clear.
If err: feels like list of contributions not integrated whole → gestalt decomposed during communication. Step back + re-express: start from what combination reveals, not what each says separately. Synthesis = message, not parts.
Step 4: Invite Challenge
State strongest reason insight might be wrong. Integrated can feel more certain than are — synthesize many inputs → convergence creates sense of validity unearned. Not disclaimer appended for politeness; structural component making usable.
- Weakest link — which domain connection least well-supported? Where integration relies on analogy not evidence?
- Assumption at risk — what needs to be true, how confident?
- Counter-insight — equal access same domains, diff conclusion, strongest argument?
- Frame challenge as valuable — challenging strengthens. "Strongest objection I see is..." = confidence + openness simultaneously
- Specify what changes mind — evidence/argument revising/collapsing. Makes falsifiable not just persuasive.
→ Honest uncertainty increases not decreases trust. Insight challengeable → improvable.
If err: no weakness → warning sign. All cross-domain involve translation between frameworks, translation always loses. Loss invisible = not found not avoided. Look harder domain boundaries — hidden assumptions live there. Common hiding: shared metaphors working differently each domain, statistical correlations assumed causal across boundaries, analogies holding structurally not quantitatively.
Check
- Audience identified, needs shaped expression
- Form chosen by insight type + audience not habit/convenience
- Stated as coherent whole not decomposed per-domain
- Contributing domains named for verification
- Simplifications stated — set aside + approximated
- Emergent value articulated — enables vs parts
- Multi-domain vocab preserved not flattened
- Entry point starts where audience is, bridges to insight
- Strongest reason wrong stated
- Falsifiable — evidence/arguments revising named
- Specialist reading own contribution recognizes accurate not caricature
Traps
- Domain-by-domain reporting: Presenting each sequential ≠ insight → raw material. Insight = emerges from combination. Lead synthesis, support domain detail if needed.
- False certainty from convergence: 3 domains same way feels strong evidence. Share underlying assumptions/data → less independent than appears. Check truly independent.
- Flatten to audience's domain: Specialist → translate entire into their language. Accessible but destroys multi-domain. Preserve texture — unfamiliar vocab = signal not noise.
- Skip challenge: Omit "why I might be wrong" feels stronger. Not — less trustworthy + less improvable. Epistemic honesty = feature.
- Insight inflation: Claim synthesis reveals more than does. Cross-domain observation ≠ breakthrough. Precise scope: "applies to X in context Y" > "changes everything."
- Premature expression: Express before fully formed → half-insights sound integrated but collapse. Stalling → upstream
integrate-gestaltproblem, not here. - Hide behind complexity: Multi-domain vocab to sound sophisticated not preserve texture. Simpler framing captures same w/o losing relationships → use simpler. Complexity necessary not performative.
→
integrate-gestalt— produces insight this expresses; express-insight = communication phaseargumentation— builds logical case for claim; express-insight communicates perception. Argumentation "here is why X is true"; express-insight "here is what becomes visible when you look A, B, C together"teach— transfers established knowledge; express-insight conveys emergent just formed. Teach transmits; express reveals.shine— channels authentic presence into communication; express-insight uses radiance carrying multi-domain w/o losing warmth/honestyexpand-awareness— widens perceptual field integration possible; express-insight closes cycle communicating widenedadaptic— meta-skill composing full synoptic cycle; express-insight = 5th + final step clear-open-perceive-integrate-express
GitHub 仓库
相关推荐技能
content-collections
元Content Collections 是一个 TypeScript 优先的构建工具,可将本地 Markdown/MDX 文件转换为类型安全的数据集合。它专为构建博客、文档站和内容密集型 Vite+React 应用而设计,提供基于 Zod 的自动模式验证。该工具涵盖从 Vite 插件配置、MDX 编译到生产环境部署的完整工作流。
polymarket
元这个Claude Skill为开发者提供完整的Polymarket预测市场开发支持,涵盖API调用、交易执行和市场数据分析。关键特性包括实时WebSocket数据流,可监控实时交易、订单和市场动态。开发者可用它构建预测市场应用、实施交易策略并集成实时市场预测功能。
creating-opencode-plugins
元该Skill帮助开发者创建OpenCode插件,用于接入命令、文件、LSP等25+种事件。它提供了插件结构、事件API规范和JavaScript/TypeScript实现模式,适合需要拦截操作、扩展功能或自定义事件处理的场景。开发者可通过它快速构建响应式模块来增强OpenCode AI助手的能力。
sglang
元SGLang是一个专为LLM设计的高性能推理框架,特别适用于需要结构化输出的场景。它通过RadixAttention前缀缓存技术,在处理JSON、正则表达式、工具调用等具有重复前缀的复杂工作流时,能实现极速生成。如果你正在构建智能体或多轮对话系统,并追求远超vLLM的推理性能,SGLang是理想选择。
