rajudandigam/agent-inspect

Local evidence debugger and trajectory-test toolkit for TypeScript AI agents: inspect causal runs, catch wrong tool paths in CI, and share safe offline evidence.

What it solves

AgentInspect addresses the difficulty of debugging complex TypeScript AI agents that perform multi-step tasks like planning, tool calling, and model interaction. It provides a local, privacy-focused way to visualize execution paths, enforce deterministic checks in CI/CD pipelines, and share verifiable evidence of agent behavior without relying on external SaaS platforms.

How it works

It captures agent execution traces as local JSONL files. These traces act as a single source of truth that can be used for three primary purposes: debugging via a CLI-based execution tree viewer, preventing regressions by running deterministic trajectory checks against your agent's logic, and generating redacted, hash-verifiable evidence bundles for sharing. It integrates with popular frameworks like LangChain, LangGraph, and the Vercel AI SDK.

Who it’s for

Developers building TypeScript-based AI agents who need local observability, automated testing of agent trajectories, and a secure way to share execution logs without uploading sensitive data to third-party services.

Highlights

  • Local-first architecture: No account, collector, or default data upload required.
  • Deterministic CI checks: Fail builds based on specific agent trajectories or tool usage expectations.
  • Privacy-focused: Built-in redaction tools and offline verification of evidence bundles.
  • Framework agnostic: Supports custom implementations and provides official adapters for LangChain, LangGraph, and AI SDK.
  • Comprehensive CLI: Tools for viewing, reporting, explaining, checking, and bundling agent traces.

Related

  • Project
    darkrishabh/agent-skills-eval

    agent‑skills‑eval is a TypeScript/Node.js CLI + SDK that runs an Agent Skill both with and without its `SKILL.md` context, grades the two outputs with a separate LLM judge, and produces JSON artifacts plus a static HTML report. It supports OpenAI‑compatible APIs, custom providers, tool‑call assertions, and full agentskills.io spec validation, making it suitable for skill development, CI testing, and comparative analysis.

  • Project
    statelyai/agent

    Stately Agent is a TypeScript library that builds LLM‑powered agents as XState state machines. The machine owns the control flow and validates every model‑proposed event, guaranteeing that agents can’t take illegal actions. It works with any AI provider via plug‑in executors, supports offline testing, durable snapshots, and self‑improving JSON policies. Ideal for safe, inspectable, and portable AI agents in Node or serverless environments.

  • Project
    iamfakeguru/agent-md

    agent‑md is a lightweight framework that installs rule files, git‑hooks, and helper scripts to turn AI coding assistants (Claude Code, Codex, Cursor, Windsurf) into contract‑driven agents. It provides a single source‑of‑truth markdown file (AGENT.md) plus enforceable checks (type‑check, lint, tests, visual screenshots, state updates) that block the model until the required evidence is present.

  • Project
    traceroot-ai/traceroot

    TraceRoot is an open‑source platform that records AI agent calls, flags bad behaviour with custom detectors, clusters issues into signals, and runs automated evals so a coding‑agent can suggest and verify fixes, closing a self‑improvement loop. It offers a CLI, Python/TS SDKs, and integrations with major agent frameworks and model providers, and can be used via hosted cloud or self‑hosted Docker Compose.

  • Project
    luoguoxiong/aipack

    aipack is a TypeScript‑first, lightweight AI‑agent framework. It includes its own runtime, a Tapable‑based plugin system, and built‑in adapters for 13+ LLM providers. Features include streaming vs. batch execution, persistent session storage, permission‑based tool approval, hybrid BM25 + vector long‑term memory, five‑stage context compression, multi‑agent orchestration (graph, pipeline, router, debate, map‑reduce), two‑way MCP integration, full observability (metrics, dashboard), and an offline evaluation suite. The project ships a CLI assistant, a library for embedding in apps, and example apps demonstrating real‑world use cases.