Last updated July 20, 2026
Insights
Insights are AI-powered analyses that run against your boards to surface things you'd otherwise have to hunt for manually — security gaps, performance bottlenecks, coupling risks, compliance issues. Think of them as automated architecture reviews.
How insights work
The system is built around Insight Skills — reusable analysis templates that define what to look for and how to report it. When you run a skill against a board, it produces structured findings: element-level insights, traced paths through your architecture, and suggested graph changes.
The Insights Bar
Each board has an Insights Bar at the bottom — a configurable strip that shows which insight skills are active for that board. Click a chip to see its latest results, or run a fresh analysis.
Running your first insight
- Open a board with at least a few nodes and edges
- Click the Insights Bar at the bottom of the board
- Select an insight skill (start with "Architecture Health" for a general overview)
- Click Run and wait for analysis to complete
- Review findings — check signals, priorities, and affected elements
- Explore any suggested graph changes before accepting them
Insight categories
Skills are organized into categories:
| Category | What it covers |
|---|---|
| Performance | Latency chains, bottlenecks, caching opportunities |
| Security | Auth gaps, data exposure, attack surfaces |
| Reliability | Single points of failure, blast radius, fallback coverage |
| Scalability | Scaling limits, horizontal vs vertical concerns |
| Maintainability | Coupling, cohesion, tech debt hotspots |
| Cost | Resource waste, right-sizing opportunities |
| Compliance | Regulatory alignment, data residency, audit trails |
| Usability | Developer experience, API ergonomics |
| Feature Planning | Impact analysis for planned changes |
| Discovery | System exploration, dependency mapping |
| Architecture Health | Overall structural quality assessment |
| Other | Custom or uncategorized analyses |
Insight signals
Every finding carries a signal that classifies its nature:
| Signal | Visual | Meaning |
|---|---|---|
| Risk | Red, alert triangle | Something that could go wrong — a vulnerability, a single point of failure, a missing fallback |
| Strength | Green, check circle | Something that's working well — good separation of concerns, proper auth, solid caching |
These are the most common signals. Risks demand attention; strengths confirm good decisions.
Each finding also has a priority: critical, high, medium, or low. A finding can additionally carry a measurement — a quantitative 0–1 score (a coupling score, a dependency depth, a blast-radius size) that drives its visual weight on the board. A measurement rides on a finding alongside its signal; it isn't a signal of its own.
What an analysis produces
An insight run can generate up to four types of results:
The four result types in detail
Element insights
Individual findings attached to specific nodes (or board-wide). Each one has:
- A short title
- A detailed explanation
- A signal (risk, strength, opportunity, or observation)
- A priority (critical, high, medium, or low)
- An optional measurement (a 0–1 score that drives visual emphasis)
- Optional tags for categorization
Affected elements
A map of which nodes are impacted by each finding, along with hints about how to emphasize them visually:
- Relationship types: primary, dependent, dependency, related, context
- Emphasis hints: highlight, pulse, glow, outline, focus
Insight paths
Traced routes through your architecture — request flows, blast radius propagation, dependency chains. Each path has ordered steps linking node-to-node with labels and annotations.
Graph suggestions
Proposed structural changes: add a cache layer, remove a redundant edge, modify a connection. Each suggestion includes:
- An action — add, remove, or modify
- A target — node, edge, or container
- A rationale — why this change is recommended
Not every insight run produces all four result types. A simple security scan might only return element insights and affected elements. A dependency analysis might focus heavily on insight paths. Chat picks what's relevant.
Results modes
Skills declare how their results combine with previous runs:
| Mode | Behavior |
|---|---|
| Replace | New results completely replace previous ones. Only one active result set per skill per board. |
| Append | New results are added alongside previous ones. Builds a history of analyses over time. |
This maps to display behavior in the Insights Bar: replace-mode skills show the latest result directly; append-mode skills open a list panel for browsing historical runs.
Insights aren't just pretty reports — they produce actionable graph suggestions. Found a missing cache layer? The suggestion tells you exactly which node to add and where. You review, accept, and your board evolves.
From finding to intent
On an app board, findings don't stop at being read. Review a finding or a graph suggestion and create an intent from it — a work item that flows to a developer through the plugin's /intents command, carrying the original finding along with it. From there the chain stays linked end to end: the signal that triggered it, the analysis, the released intent, and the commit that closed it. The full flow lives in Intents.