DOCSEARCHbyAlgolia
Submit Docs

Agent Sandbox Skill

repository·main·Indexed 18 days ago

https://github.com/disler/agent-sandbox-skill

An extension for AI agents (such as Gemini CLI, Claude Code, and Codex CLI) that provides secure, isolated E2B sandboxes. It enables agents to execute untrusted code, build and host full-stack applications using Vue, FastAPI, and SQLite, and perform browser automation via Playwright. The skill includes a Python-based CLI and a set of backslash commands to orchestrate the full software development lifecycle from planning and building to hosting and testing.

Tokens
41.5K
Snippets
42
Records
175
Agent score
64%

What's inside agent-sandbox-skill

  1. Overview of Chart Sketch Pro

    main
    Chart Sketch Pro is a visual communication tool designed for professionals to create chart collections, dashboards, and presentations. It uses an intuitive slider-based data entry system instead of traditional numeric or spreadsheet input, allowing users to focus on relationships and proportions (e.g., Very Low to Very High) rather than precise numbers. It is ideal for strategy meetings, pitch decks, and stakeholder updates.

  2. Overview of Agent Sandbox Skill

    main

    The Agent Sandbox Skill is an extension for AI agents (such as Gemini CLI, Claude Code, and Codex CLI) that provides managed, isolated execution environments using E2B Sandboxes.

    It allows agents to safely:

    • Execute code: Run untrusted code, tests, and binaries in isolation.
    • Develop full-stack apps: Scaffold, build, and host applications using Vue, FastAPI, and SQLite.
    • Automate browsers: Use Playwright for visual validation and testing.
    • Scale engineering tasks: Run multiple independent agent forks, each with its own sandbox.

    This skill is designed to provide agents with full agency over their environment (installing packages, modifying files, running commands) while protecting your local filesystem and production environments.

  3. Build a Knowledge Base & Documentation Builder

    main
    This prompt provides a comprehensive blueprint for building a modern documentation platform similar to Notion or GitBook. It is designed for developers to implement a system that allows teams to create, organize, and maintain internal knowledge bases with features like rich markdown editing, hierarchical organization, full-text search, real-time collaboration, and version control.

  4. Understand the Agent Sandbox Skill architecture

    main

    The project is structured into three main components:

    • CLI: A Python-based command-line interface using click, e2b, and rich. It is located in .claude/skills/agent-sandboxes/sandbox_cli/.
    • Prompts: Markdown-based prompt templates stored in .claude/skills/agent-sandboxes/prompts/.
    • Skill Definition: The SKILL.md file defines the capabilities and interface exposed to the agent.

    For more details on the underlying sandbox infrastructure, refer to the E2B Documentation.

  6. Implement the Micro-Betting Odds Lab architecture

    main

    To build the Micro-Betting Odds Lab, follow these architectural requirements for the backend and frontend:

    Backend Requirements

    • Data Model: You must implement entities for Users, Wallets, Transactions, Events, Markets, Outcomes, OddsTicks, Bets, Settlements, Limits, and Audits.
    • API & Ingestion:
      • An odds feed ingestion system (simulated) with WebSocket (WS) broadcasting.
      • A bet placement endpoint that performs atomic validation of both the wallet balance and the current odds.
      • An idempotent settlement worker.
      • Admin endpoints for market management.
      • Risk check middleware and audit logging.
    • Real-time Communication: A WebSocket channel for broadcasting live odds and bet status updates.

    Frontend Requirements

    • User Interface:
      • A live event list with markets and auto-refreshing odds via WebSockets.
      • A bet slip UI, wallet balance widget, and transaction history.
      • Error handling for suspended markets or limit breaches.
      • Optimistic UI updates that are gated by server confirmation.
    • Admin Interface:
      • A panel for odds adjustments and market suspension.
      • A liability dashboard for monitoring risk.

  8. Implementation Requirements for Analytics Ops Monitoring Hub

    main

    When building the Analytics Ops Monitoring Hub, ensure the following technical requirements are met across the stack:

    Backend & API Requirements

    • Ingest Pipeline: Implement a queue + processor architecture. Storage should be optimized for time-series data (e.g., using SQLite-friendly rollups).
    • Alerting Engine: An alert scheduler that evaluates rules and SLO burn rates, capable of dispatching via webhooks or email.
    • Security: Implement Role-Based Access Control (RBAC) for teams, audit logs for rule changes, and support for maintenance window suppression.

    Frontend Requirements

    • Dashboard Builder: Include chart presets, SLO cards, and incident overlays.
    • Management UIs: An alert rule editor, an incident timeline UI, a query editor with saved queries, and configuration pages for notification channels.
    • UX Features: Live auto-refresh for dashboards and status badges for alerts and incidents.

    Testing Strategy

    • Backend: Validate ingest logic, rollup correctness, alert evaluation against fixtures, RBAC, and mock webhook/email dispatch.
    • Frontend: Test dashboard rendering, rule editing flows, incident timeline interactions, and team scoping guards.
    • Integration: Perform an end-to-end flow: send sample events $\rightarrow$ create SLO + alert $\rightarrow$ trigger breach to create incident $\rightarrow$ resolve incident $\rightarrow$ verify dashboard updates.

  9. Supported Chart Types in Chart Sketch Pro

    main

    The tool supports 6 core chart types:

    • Bar Chart: Vertical or horizontal bars with slider-controlled heights.
    • Stacked Bar: Multiple series per category showing composition.
    • Pie Chart: Segments with slider-controlled proportions.
    • Donut Chart: A pie variant with a center label or metric.
    • Line Chart: Points connected by lines to show trends.
    • Area Chart: A filled line chart for emphasis.

  10. Use Slider-Based Data Entry Modes

    main

    Data points are controlled via level sliders rather than numeric inputs. There are three available modes:

    • Standard Mode: 5 levels (Very Low → Low → Medium → High → Very High).
    • Extended Mode: 10 levels for finer granularity.
    • Comparison Mode: A negative-to-positive scale ranging from -5 to +5.

    Features include visual level indicators with color feedback, the ability to add/remove points (2-12 per chart), drag-and-drop reordering, and bulk level adjustment (shifting all points up or down simultaneously).

  11. Configure Multi-Series and Annotations

    main

    Charts support advanced visual features:

    Multi-Series Support

    • Add up to 4 series per chart (e.g., 'Before/After', 'Plan/Actual').
    • Each series has a unique name and color.
    • Includes a series legend with toggle visibility.

    Annotations

    • Callouts: Text callouts pointing to specific data points.
    • Highlight Zones: Shaded regions of the chart.
    • Trend Lines: Auto-fit or manual lines.
    • Reference Lines: Horizontal markers (e.g., 'Target', 'Average').
    • Emoji Reactions: Visual emphasis on specific data points.

  12. Core Features of the Content Workflow System

    main

    The system is composed of several functional modules:

    • Idea Management: Kanban-style organization (Backlog, Research, In Progress, Published), tagging by topic/format/priority, and linking related ideas.
    • Content Calendar: Visual views (day, week, month) with drag-and-drop scheduling across multiple channels (YouTube, blog, newsletter, social media).
    • Script & Draft Management: Rich text editing with markdown, script templates, version control with diff viewing, and SEO suggestions.
    • Production Workflow: Customizable stages (e.g., Draft → Review → Edit → Record → Publish), task assignment, and approval workflows.
    • Analytics & Performance: Tracking views, engagement, and clicks via manual entry or API integration (YouTube, Google Analytics).
    • Collaboration Tools: Role-based permissions (creator, editor, viewer), @mentions, activity feeds, and shared asset libraries.