Mushroom PCE is a VS Code extension that helps you understand code faster with AI explanations and an interactive Circuit Mode.

Mushroom PCE Circuit Mode

Key Features

File analysis from the active editor in two modes:
- Developer Mode: AI explanation with structured sections.
- List Mode: deterministic static inventory (no model call).
Symbol-aware output with clickable links (Go To Function).
Blueprint Planner (plain-English feature planning workspace):
- Planner chat that is grounded in scanned src/ structure + known functions/exports.
- Reuse-first planning suggestions before proposing new files/functions.
- Generates both a structured JSON spec and a plain-text implementation prompt.
- Saves artifacts to docs/feature-plans/*.md.
- Resizable sections (Blueprint Planner Chat, JSON Spec, Prompt Output).
- One-click copy buttons for chat transcript, JSON spec, and prompt output.
Circuit Mode for architecture/runtime exploration:
- Current File graph
- Full Architecture graph
- Runtime CodeFlow graph
- Skeleton subgraph focus mode
Node Details panel with node-scoped chat.
Copyable AI output in Node Details (copy icons for answer/code blocks).

AI-Enhanced Circuit Mode

AI Insights for selected graph scope.
AI edge suggestions overlay:
- Dashed suggested edges
- Confidence display
- Apply/Reject all
- Per-edge Apply/Reject controls
Relation Explain workflow:
- Dedicated HUD controls: Set From, Set To, Explain, Reset
- Quick-pick relation state and AI explanation for the selected pair
HUD polish:
- Compact HUD sizing
- Scrollable HUD cards
- Better visibility of graph behind controls

Circuit HUD Guide

This section explains every control visible in the Circuit Mode HUD, how they differ, and when to use each one.

Feature-by-feature reference

HUD feature	What it does	Difference vs similar controls	Practical use case
`Lock view`	Freezes camera interaction so pan/zoom do not change the viewport.	This does not lock graph data or selection; it only locks camera movement.	Keep the scene stable while comparing node details or writing notes.
`Minimize HUD`	Collapses the HUD to reduce visual footprint.	Different from `Maximize HUD`; minimize hides density, maximize increases reading area.	When you want more canvas space while inspecting graph layout.
`Maximize HUD`	Expands HUD cards for easier reading of details/AI text.	Different from `Minimize HUD`; this is for readability, not canvas focus.	Reviewing longer AI insights or relation explanations.
`View Mode: Architecture`	Layer-oriented structural view.	Focuses on module/layer topology, not runtime call/data flow ports.	System design review, boundary checks, ownership mapping.
`View Mode: Runtime`	Function/data-flow oriented view with ports and flow edges.	Focuses on execution and movement flow, not layered architecture grouping.	Debugging flow path, tracing function sequence, finding runtime bottlenecks.
`Layout: Collapse All`	Collapses architecture layers/groups.	Opposite of `Expand All`. Mostly meaningful in Architecture view.	Reduce clutter to get a macro view quickly.
`Layout: Expand All`	Expands all collapsed layers/groups.	Opposite of `Collapse All`. Mostly meaningful in Architecture view.	Dive back into layer internals after high-level scan.
`Scope: Full Architecture`	Loads project-wide architectural graph.	Broader than `Current File`; less step-focused than `CodeFlow`.	Cross-file dependency and layering analysis.
`Scope: Current File`	Restricts graph to active file context.	Narrowest scope; fastest and least noisy.	Local refactor, one-file deep dive, quick comprehension.
`Scope: CodeFlow`	Runtime-oriented step graph with codeflow edges.	More sequence/flow oriented than the other scopes.	"What happens next" analysis and path walkthroughs.
`Dependencies: Imports/Exports`	Shows dependency graph using import/export relations.	Structural dependency only; no call hierarchy enrichment.	Clean dependency map for architecture hygiene.
`Dependencies: + Call Hierarchy`	Adds call hierarchy signals on top of imports/exports.	Richer but noisier than Imports/Exports-only mode.	Impact analysis before changing shared functions.
`Edges: All`	Shows all eligible edges under current filters.	Broadest edge visibility.	Exploration and discovery when you need full context.
`Edges: API-high`	Filters to high-confidence API edges (plus key context edges).	Higher precision, lower recall than `All`.	Reduce noise when validating trustworthy links.
`AI Insights`	Requests AI summarization/enrichment for current graph context.	This generates analysis; it does not change edges by itself.	Fast orientation in unfamiliar areas of code.
`Apply AI`	Applies pending AI-suggested edges to the graph.	Commits suggestions; opposite intent of `Reject AI`.	Keep useful suggested links after validation.
`Reject AI`	Discards pending AI-suggested edges.	Removes suggestion queue; opposite intent of `Apply AI`.	Reset noisy suggestions and keep graph strict/manual.
`Relation Explain: Set From`	Pins current selection as source node for relation query.	Source endpoint only; pair with `Set To`.	Build repeatable A -> B relation checks.
`Relation Explain: Set To`	Pins current selection as target node for relation query.	Target endpoint only; pair with `Set From`.	Confirm destination impact or dependency rationale.
`Relation Explain: Explain`	Requests AI explanation for the current `From -> To` pair.	Consumes pinned relation state; does not auto-select nodes.	"Why are these connected?" and path explanation.
`Relation Explain: Reset`	Clears the pinned `From/To` relation state.	Clears relation context only, not graph or selection.	Start a fresh relation query without stale endpoints.
`Selection: Include external neighbors`	Expands focused selection with connected hidden neighbors.	Adds context around selected node, not global scope switch.	Grow local neighborhood without jumping to full architecture.
`Selection: Disconnect connection`	Disconnects selected node from Context Bot if connected.	Context Bot edge operation only.	Clean up temporary context wiring in Runtime CodeFlow.
`+` FAB -> `Context Bot`	Adds Context Bot node in Runtime CodeFlow.	Runtime utility node for context aggregation, not a source-code symbol.	Build ad-hoc context bundles from multiple nodes.
Runtime interaction: click output port -> click Context Bot	Connects selected node output into Context Bot context input; repeat to detach.	Port-based wiring flow specific to Runtime CodeFlow.	Assemble multi-node context before asking AI in Node Details.
Runtime interaction: `Ctrl+click node`, then `Ctrl+click Context Bot`	Fast toggle connect/disconnect between a node and Context Bot.	Shortcut path for the same context-bot connection intent.	Rapid connect/disconnect during iterative tracing.
Canvas interaction: double-click background	Toggles Hand mode (pan behavior).	Navigation mode switch, not data/filter mode.	Comfortable camera movement in dense graphs.
Canvas interaction: drag/scroll	Drag nodes to rearrange, scroll to zoom.	Visual organization/navigation only; graph semantics unchanged.	Reduce overlap and improve readability while investigating.

What users often mix up

Pair	Difference
`View Mode` vs `Scope`	`View Mode` changes how nodes/edges are interpreted visually (architecture vs runtime). `Scope` changes which graph dataset is loaded (current file vs project vs codeflow).
`Dependencies` vs `Edges`	`Dependencies` chooses dependency signal type (imports only vs imports+calls). `Edges` chooses visibility strictness (all vs API-high).
`Relation Explain` vs Context Bot wiring	`Relation Explain` is an AI reasoning workflow over a selected pair. Context Bot wiring is runtime context aggregation for node chat/context workflows.
`Minimize/Maximize HUD` vs `Lock view`	Min/max are HUD layout controls. Lock view controls camera motion.

Practical mix-and-match recipes

Architecture review for a large repo: View Mode: Architecture + Scope: Full Architecture + Dependencies: Imports/Exports + Edges: API-high.
Impact analysis before editing a function: View Mode: Architecture + Scope: Full Architecture + Dependencies: + Call Hierarchy + Edges: All.
Focused refactor in one file: Scope: Current File + View Mode: Runtime + Edges: All + Include external neighbors when needed.
Runtime path debugging: Scope: CodeFlow + View Mode: Runtime + Context Bot wiring (output port -> Context Bot) + Node Details chat.
Relationship explanation workflow: click node A -> Set From, click node B -> Set To, then Explain, and Reset before testing a different pair.
Low-noise AI-assisted exploration: Scope: Full Architecture + Edges: API-high + AI Insights, then selectively Apply AI only when suggestions look credible.

Commands

Mushroom PCE: Start Mushroom PCE (mushroom-pce.start)
Mushroom PCE: Analyze Active File (mushroom-pce.analyzeActive)
Mushroom PCE: Select Model (mushroom-pce.selectModel)
Mushroom PCE: Set List Mode (mushroom-pce.setListMode)
Mushroom PCE: Set Developer Mode (mushroom-pce.setDeveloperMode)
Mushroom PCE: Set Definition Mode (mushroom-pce.setDefinitionMode)
Mushroom PCE: Toggle Graphify Context (mushroom-pce.toggleGraphifyContext)
Mushroom PCE: Open Circuit Mode (mushroom-pce.openCircuit)
Mushroom PCE: Open Blueprint (mushroom-pce.openBlueprint)
Mushroom PCE: Go To Function (mushroom-pce.goToFunction)

Project Structure

src/
  app/
    activate.ts
    state/
      analysisCache.ts
      modelState.ts

  commands/
    registerCommands.ts
    startPceCommand.ts
    openCircuitCommand.ts

  controllers/
    mushroom/
      MushroomPanelController.ts
    blueprint/
      BlueprintPanelController.ts
    circuit/
      CircuitPanelController.ts
      CircuitDetailsPanelController.ts

  services/
    ai/
      requestModelText.ts
    analysis/
      buildListModeOutput.ts
      frequency.ts
    language/
      detectLanguageWarning.ts
    prompts/
      buildPrompt.ts
    symbols/
      parseSymbolLocations.ts
    blueprint/
      buildBlueprintPlan.ts
      generateBlueprintCode.ts
      scanWorkspaceBlueprint.ts
    circuit/
      buildGraph.ts
      buildGraphHybrid.ts
      buildProjectArchitectureGraph.ts
      buildProjectGraph.ts
      buildCodeFlowGraph.ts
      buildSkeletonGraph.ts
      ai/
        enrichCircuitGraph.ts
        explainNodeRelation.ts

  views/
    mushroom/
      mushroomHtml.ts
      mushroomStyles.ts
    circuit/
      circuitHtml.ts
      circuitStyles.ts
    details/
      detailsHtml.ts
      detailsStyles.ts

  webview/
    circuit/
      circuitApp.ts

  shared/
    types/
      appTypes.ts
      circuitTypes.ts
    webview/
      messages.ts

  utils/
    escapeHtml.ts
    getNonce.ts
    markdownToHtml.ts
    markdownToChatHtml.ts
    regex.ts

  extension.ts

Requirements

VS Code ^1.110.0
Access to at least one VS Code chat/language model (Developer Mode, Node Chat, AI Circuit features)
Node.js 18+ (development only)

Development

npm install
npm run compile

Useful scripts:

npm run watch
npm run check-types
npm run lint
npm run lint:fix
npm run test

Pre-commit quality gate:

Husky + lint-staged is configured for this repo.
On git commit, staged *.ts/*.tsx files run through eslint --fix --max-warnings=0.
Commits are blocked if lint errors remain.

Reusable bootstrap for other projects:

Run this from this repo and point it to any JS/TS project folder:

powershell -ExecutionPolicy Bypass -File .\scripts\bootstrap-quality.ps1 -ProjectPath "D:\path\to\your-project"

What it does:
- Installs husky + lint-staged as dev dependencies
- Adds/updates prepare and lint:fix scripts in package.json
- Adds a lint-staged rule for *.{js,jsx,ts,tsx}
- Creates .husky/pre-commit with npx lint-staged
- Sets git config core.hooksPath .husky when the target is a git repo

Notes

Graph quality depends on language server symbol/call-hierarchy support.
Large workspaces may take longer for Full Architecture/CodeFlow views.
Some edges can be heuristic/fallback-labeled when high-confidence API data is unavailable.
Blueprint planning context is currently sourced from src/ and sampled file insights.