Agent Dashboard CLI
A local web dashboard for agent reliability scans, skill generation and network probes with no separate tool installs
- 62 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The dashboard downloads and runs tools, scans your repositories, writes skills and, during a probe, sends tool code to the configured LLM.
Install
- Works in
Manual install
git clone https://github.com/trustabl/agent-dashboard.gitThen `make run` in the directory, which installs the needed tools and serves the dashboard on `127.0.0.1:4317`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The dashboard downloads and runs tools, scans your repositories, writes skills and, during a probe, sends tool code to the configured LLM.
This is third-party code. Review the repository files before installing.
What it does
`Agent Dashboard` runs three Trustabl tools from one local single-user web UI: Analyzer (`trustabl scan`), Forge (`trustabl forge`, `trustabl forge check`) and Probe (`autofix fix`). They scan agent reliability, generate skills and network policies, and the overview page gathers results, the run log and findings by severity. The tools need no separate install. The dashboard listens only on `127.0.0.1:4317`, with no login and no user or org model. The pinned `trustabl` 0.1.16 and `autofix` 0.5.0 are downloaded by `make run` with SHA-256 checks. Every run has a Copy CLI button with the exact terminal command. State lives in `~/.agent-dashboard` and `make wipe` clears it.
Who it is for. Agent developers who want to check an agent's reliability and network access in their repositories and get generated skills from one dashboard.
Good fit when
- An agent in a local repository is checked for reliability problems before release
- A `SKILL.md` must be generated for a project and checked for freshness
- A network access probe runs in Docker with a local Ollama model
- A run result is to be reproduced in a terminal with the exact command
Not a fit when
- You need a multi-user dashboard with login and roles, since this one is single-user
- You have no Docker or OpenShell and no compatible LLM to run Probe
- Source code must not go to a third-party LLM
Example request
Scan the `~/src/my-agent` repository with Analyzer, show findings by severity and copy the exact run command.Limitations
It needs macOS, Linux or WSL2 with `git`, `curl`, `tar` and `make`, plus Node.js 22.13 or newer (`make run` installs a private copy). Probe also needs Docker or OpenShell and an OpenAI-compatible LLM such as Ollama. Probe sends tool source only to the LLM you configure; keys are read from the environment and the dashboard stores none. Policy and OpenTelemetry telemetry are listed as coming soon. Downloaded tools are pinned by version in `tools.lock.json`.
How to disable. Stop the dashboard process and run `make wipe`, which deletes `~/.agent-dashboard` with the database and runs and, on macOS, the tools' caches. Your repositories and any `SKILL.md` you chose to write stay.
FAQ
How do I run it?
Clone the repository and run `make run` (or `./scripts/run.sh`), then open `http://127.0.0.1:4317` and click Scan repo, Generate skill or Run Probe.
What do the three pages do?
Forge creates and checks skills, Analyzer runs `trustabl scan` and Probe runs `autofix fix` for network policies. Overview gathers statuses and findings.
What does Probe need?
Docker or OpenShell and an LLM with an OpenAI-compatible address, such as Ollama: `--llm-url http://localhost:11434/v1`.
Where does my data go?
Source code stays on the machine. Only Probe sends tool code to the LLM you specify, and the dashboard stores no keys.
How do I repeat a run in a terminal?
Every run has a Copy CLI button that copies the exact command the dashboard ran.
Discussion
No questions or reviews yet.
Related
CLI toolsAgent Reliability Analyzer statically scans AI agent code, finds reliability gaps with 340 plus rules and applies fixes
A local debugger for TypeScript agents: run step trees, CI trajectory checks and portable evidence bundles, with no account.
Vercel CLI for testing coding agents on your framework: runs Claude Code, Codex and others and measures the success rate
agentevals scores AI agent behavior from OpenTelemetry traces and ships MCP tools plus the /eval and /inspect skills.
CLI checks a web page in Chromium: finds every control, uses it and reports what broke. No tests to write and no LLM.
An MCP server and CLI that give an agent its own hidden Linux desktop on X11 or Wayland without touching your screen or browser
README in short
The README describes the dashboard's purpose, a table of pages and commands, first run, Probe setup with a local model, where state lives and `make wipe`, development commands and the Apache-2.0 license.