FixLab coordinates repository-aware diagnosis, scoped code changes, focused tests, and Playwright validation through Agency Copilot or GitHub Copilot. It supports React/.NET applications and backend-only API validation. Application-specific paths, commands, environments, and safety rules stay in the application's repository-owned profile.
Get started
Install and sign in to a supported Copilot runtime, open the application checkout, and paste:
Install or reuse FixLab, register its MCP server, and onboard this repository
with --runtime copilot. Follow FIXLAB-SETUP.md if supplied; otherwise use:
https://github.com/microsoft/FixLab/blob/main/docs/installation.md
Review the real profile and setup plan with me before each change.
Ask me for missing setup details; do not guess paths or settings.
Use Microsoft Edge by default for required browser validation. Review feed
access, dedicated browser SSO and private-state protection; run Doctor.
Ask before starting or reusing the matching dashboard. I will complete MFA.
Preserve my code and running services. Do not start a job or publish changes.
For reusable instructions, copy or attach FIXLAB-SETUP.md without overwriting an existing guide. If Copilot or the file is unavailable, use installation.md and the detailed manual steps. Download the Word quickstart or follow the VS Code guide. Setup still requires reviewed approvals; installing FixLab alone does not establish application readiness or grant permissions.
What you can do
| Workflow | Outcome |
|---|---|
| Fix a bug or small enhancement | Scoped changes, validation evidence, and an authorized pull request |
| Validate an existing change or PR | Findings and evidence without source or PR changes |
| Run focused Playwright validation | Browser or configured API assertions without an implementation job |
Architecture and benefits
FixLab coordinates the developer-facing dashboard and CLI, a repository-owned profile, Copilot execution, and configured validation. It runs locally on the developer or runner machine; the configured AI service supplies model access.
System architecture
FixLab architecture: local orchestration, repository contract, Copilot runtime, evidence, and external services
The dashboard and terminal chat share job state, blockers, and results.
Interrupted work can resume in the same job; a bounded queue keeps later
requests separate.
During the existing PR step, verified reusable learning is captured in a
reviewed repository skill (.github/skills/<topic>/SKILL.md), updating a
matching skill rather than duplicating it. No extra roadmap stage is added.
If there is no new reusable learning, the PR says so instead of adding a file.
End-to-end job flow
FixLab workflow: intake, diagnosis, reproduction, change, review, local stack, live test, and outcome
Efficient delivery defaults to focused checks and an authorized draft PR. Optional broad tests/builds are controlled by saved preferences; required repository checks and CI/merge policies still apply. Failed, deferred, stale, or skipped checks are never reported as passed.
See architecture and benefits for components, benefits, deployment, recovery, and trust boundaries, or the architecture reference for runtime and MCP contracts.
Work in the dashboard
After setup, ask Copilot:
Start or reuse this repository's FixLab dashboard with
--runtime copilotand show me its local URL. Ask me if any setup details are missing. Do not start a job.
Or start it manually and open the printed local URL:
fixlab dashboard C:\source\application --port 4317 --runtime copilot
Follow the dashboard user guide to submit a request, choose run options, handle approvals, review evidence, and resume saved work.
Stop or replace an unused dashboard
At fixlab>, /stop stops the job and leaves it resumable; /exit disconnects
chat without stopping the dashboard. Press Ctrl+C in the dashboard's
terminal to stop the server and release its port.
If Copilot started it in the background, ask:
Shut down FixLab on port 4317 and only its owned processes. Leave other applications running.
Start a replacement using the available port:
fixlab dashboard C:\source\application --port 4317 --runtime copilot
Optionally, in another terminal, connect chat to that same port:
fixlab chat C:\source\application --port 4317 --runtime copilot
Closing a browser tab or exiting Copilot does not stop a detached dashboard. No reinstall or file deletion is needed. Restarting preserves saved jobs; submit a new request for a fresh job.
Documentation
| Start here | Read next |
|---|---|
| First installation | Prerequisites, machine changes, and reviewed setup |
| Detailed installation | Copilot prompt, VS Code, updates, and optional integrations |
| Dashboard user guide | Startup, requests, approvals, evidence, recovery, and shutdown |
| Architecture and benefits | Component diagrams, job flow, deployment, and trust boundaries |
| Repository onboarding | Profiles, worktrees, API-only validation, and linked applications |
| Troubleshooting | Installation, authentication, runtime, and validation blockers |
| All documentation | Technical references, security, rollout, and contributor guides |
Execution and safety
FixLab is a single-user local tool, not a hosted multi-user broker or
managed runner service. The dashboard binds to 127.0.0.1; worktrees isolate
changes, not operating-system privileges.
Authentication, sensitive actions, deployment, and publication retain their approval requirements. Only FixLab-owned processes and artifacts may be cleaned up. Never put credentials, browser state, private endpoints, or customer data in prompts, source, logs, or durable caches. See the security model.
Install from GitHub until npm/marketplace publication is available; consult GitHub releases for published versions. A default-branch install can contain changes newer than a release.
Contributing
See Contributing for reproducible setup and the required
npm run validate completion gate, including the Windows validation runner.
Behavior contracts live in specs/v1; start a repository profile
from templates/repository-profile.json.