Skip to content

miethe/CCDash

Repository files navigation

CCDash — Agentic Project Dashboard & Analytics Platform

CCDash is a local-first dashboard designed to orchestrate, monitor, and analyze the work of AI Agents within a software project. It bridges the gap between traditional project management (Kanban/Docs) and AI-driven development (Session logs, Tool usage, Token metrics).

🚀 Core Philosophy

  1. Agent-First: Every task, commit, and document changes is traceable back to specific Agent Sessions.
  2. Forensics & Debugging: Detailed introspection into Agent "thought processes," tool usage, and costs.
  3. Local Context: Tightly coupled with the local filesystem, Git history, and Markdown frontmatter.

🛠️ Technology Stack

  • Frontend: React 19, TypeScript, Vite
  • Styling: Tailwind CSS (Slate dark mode theme)
  • Icons: Lucide React
  • Visualization: Recharts (Area, Bar, Pie, Line, Composed charts)
  • Routing: React Router DOM (v7)
  • AI Integration: Google Gemini SDK (@google/genai)

📦 Feature Specification

1. Global Navigation & Layout

  • Collapsible Sidebar: Fluid transition sidebar with icon-only mode.
  • Notifications: Badge system for system alerts (e.g., cost overruns, quality drops).
  • Theme: Deep "Slate" dark mode optimized for long engineering sessions.

2. Dashboard (Overview)

  • KPI Cards: High-level metrics for Total Spend, Average Session Quality, Hallucination Rate, and Shipping Velocity.
  • AI Insights: Integrated Google Gemini analysis that reads current metrics/tasks and generates executive summaries on project health.
  • Visualizations:
    • Cost vs. Quality Area Chart: Tracks spending against code quality over time.
    • Model Usage Bar Chart: Breakdown of underlying LLM usage (Claude vs. Gemini).

3. Feature Board (Aggregate Delivery View)

  • Feature-Centric View: Redesigned board that groups work into Features discovered from project documentation (PRDs and Implementation Plans).
  • Document-First Discovery: Automatically cross-references PRDs, Implementation Plans, and Progress files to build a cohesive view of each feature.
  • Views: Toggle between Kanban Board (grouped by feature stage: Backlog, In Progress, Review, Done) and List View.
    • Features with deferred steps still land in Done when all tasks are terminal (done or deferred), and show a deferred caveat indicator.
  • Drill-Down Modal:
    • Overview: Visualize linked documents (PRDs, Plans, Reports), category badges, and related feature variants (v1, v2).
    • Phases tab: Accordion view of implementation phases. Each phase expands to show a checklist of individual tasks with their real-time status.
    • Documents tab: Quick access to all documentation files associated with the feature.
  • Filtering: Search features by name/slug/tag, filter by category/status (including deferred caveat), and sort by update date or total task count.
    • In the feature modal, the Phases tab supports phase-status and task-status filtering, including deferred.

4. Project Management

  • Dynamic Project Switching: Easily switch between multiple local projects from the sidebar.
  • Project Context: Each project maintains its own configuration for session logs, plan documentation, and progress tracking.
  • Project Creation: Add new projects by specifying local paths and metadata, which are persisted for future sessions.

5. Plan Catalog (Documentation)

  • Views:
    • Card Grid: Visual overview of PRDs, RFCs, and Architecture docs.
    • List: Sortable table view.
    • Folder/Explorer: 3-pane IDE-style file explorer for navigating documentation hierarchies.
  • Document Modal:
    • Markdown Preview: Renders document content.
    • Metadata: Parses frontmatter (Authors, Version, Status).
    • Version Control: Lists linked Commits and PRs.
    • Bi-directional Linking: Tabs showing "Linked Files" and "Linked Entities" (Tasks/Sessions).

6. Session Inspector (Agent Forensics)

The core debugging loop for AI interactions.

  • Session Index: Grid view of Active (Live) vs. Historical sessions with cost and health indicators.
  • Deep Dive View (Tabbed Interface):
    1. Transcript:
      • 3-pane fluid layout (Log list, Detail view, Metadata sidebar).
      • Message/Tool/Skill Support: distinct visual styling for different log types.
      • Inline Expansion: Inspect tool arguments and large outputs without losing context.
    2. Files:
      • Table of all files touched in the session.
      • Diff stats (Additions/Deletions).
      • "Open in IDE" simulation or "View Doc" if the file exists in the Plan Catalog.
    3. Artifacts:
      • Visual cards for generated Memories, Request Logs, and Knowledge Base entries.
    4. App Impact:
      • Codebase Impact Chart: Line chart tracking LOC added/removed and file touch-counts over the session duration.
      • Test Stability Chart: Area chart visualizing Test Pass vs. Fail counts over time.
    5. Analytics (Advanced):
      • Interactive Charts: Click on any chart (Active Agents, Tool Usage, Model Allocation) to view detailed stats (Cost, Tokens, Count) and deep-link to filtered transcript views.
      • Token Timeline: Switch between summary bar charts and a detailed cumulative area chart overlaying discrete events (Tool Executions, File Edits).
      • Master Timeline: Full-width correlation view of session lifecycle events against token consumption.
    6. Agents:
      • Card view of all participating agents (e.g., Architect, Coder, Planner).
      • Click-to-filter transcript by specific agent.

7. Settings

  • Alert Rules Engine: Configure thresholds for active monitoring (e.g., "Alert if Session Cost > $5.00").
  • Toggle System: Activate/Deactivate specific rules.

📊 Data Models

Feature

The primary unit of delivery. Aggregates:

  • linkedDocs: References to PRDs, Implementation Plans, and Reports.
  • phases: Implementation phases containing granular ProjectTask items.
  • relatedFeatures: Bi-directional links to other version variants of the same feature.
  • deferredTasks: Count of terminal-complete tasks deferred for later follow-up.

AgentSession

The atomic unit of work. Contains:

  • logs: The conversation and tool execution stream.
  • impactHistory: Time-series data of code/test changes.
  • updatedFiles: List of file modifications.
  • linkedArtifacts: References to external systems (SkillMeat, MeatyCapture).

ProjectTask

Represents a specific unit of implementation.

  • status: Mapped from frontmatter (pending/backlog, in-progress, review, completed/done, deferred).
  • cost: Derived from estimated effort.

PlanDocument

Represents Markdown documentation. Contains:

  • frontmatter: Metadata for linking back to Tasks and Sessions.

🚀 Running the Project

  1. Install frontend dependencies:

    npm install
  2. Install backend dependencies and create backend/.venv:

    npm run setup
  3. Start full local development stack (backend + frontend):

    npm run dev
  4. Useful scripts:

    • npm run dev:backend - backend only (reload mode)
    • npm run dev:frontend - frontend only
    • npm run build - build frontend assets
    • npm run start:backend - production-style backend startup
    • npm run start:frontend - serve built frontend (vite preview)
  5. Environment Variables:

    • GEMINI_API_KEY: Enables AI insight features.
    • CCDASH_BACKEND_HOST / CCDASH_BACKEND_PORT: Backend bind host/port for startup scripts.
    • CCDASH_API_PROXY_TARGET: Vite proxy target for /api requests.

For detailed setup, troubleshooting, and deployment startup guidance, see docs/setup-user-guide.md.

About

Dashboard for Agentic AI sessions

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published