Agents that run your SDLC.

ADLC keeps policy in .adlc/. Dev, Review, and QA run from Azure DevOps, GitHub, and Jira.

Sign inHow it works
ADLC runs list showing agents triggered from work items and pull requests, with cost and status
Azure DevOpsGitHubJira

Board to PR to review, without a new chat every time.

A work-item state change or a pull request starts the matching agent. Each role stays sharp: Dev implements, Code Review judges, QA tests.

  1. Ready

    Dev

    Opens a planning PR and child tasks. The plan is reviewable before anyone implements.

  2. In Progress

    Dev

    Implements the approved plan on the feature branch and opens the work PR.

  3. Pull request

    Code Review

    Posts a verdict on bugs, regressions, and missing tests. It never implements.

  4. Ready for Testing

    QA

    Designs linked cases, then runs them with agent-browser or Playwright in the sandbox.

  5. Every weekday

    Security

    Scans with Trivy and Semgrep. Files High+ bugs on the board.

  6. Every day

    Self Improvement

    Reads run feedback, opens a policy PR, then proposes that PR. Live config is never patched.

Agents are files in the repo.

Instructions, triggers, and secret names live in .adlc/. The dashboard holds keys, webhooks, and on/off. Review a policy change the same way you review code: as a pull request.

Connect ADLC MCP in Cursor or Claude, then install the policy skill. Local agents can list what already exists before they scaffold anything.

# .adlc/agents/dev.md
slug: dev
name: Dev
model_family: grok
triggers:
  - type: webhook
    source: azure-devops
    events: [workitem.updated]
    filters:
      work_item_type: Product Backlog Item
      state: Ready

Mentions, MCP, sandboxes, and cost you can audit.

Mention the agent. Pin the role.

Comment on a PR or work item. /dev forces Dev. Without a slug, the router reads descriptions and instructions and picks once.

you @handle /dev tests failed on the planning PR. Fix the filter signature.

Developers page for installing ADLC MCP with OAuth and the policy skill
Model Context Protocol

ADLC MCP in the IDE

OAuth for Cursor and Claude. Org keys stay under Advanced for sandbox runs.

Isolated Vercel Sandbox

Each run restores the environment snapshot, injects connector secrets, and starts the CLI you pinned.

  • Claude Code
  • Cursor CLI
  • GitHub Copilot CLI
  • Grok Build
Project overview with run cost, tokens, and agent type spend

Cost and ROI Tracking

Agent spend plus sandbox CPU. Cursor confirmed via Admin API. Reports sum confirmed cost only.

Run detail with cost, tokens, file changes, and output summary

Every run is an audit trail

Trigger item in, pull requests out. Feedback on the run feeds Self Improvement the next morning.

The starter pack, not a prompt library.

Repo Initializer writes these into .adlc/. New agents stay disabled until you enable them. Planning is Dev or SpecKit, never both.

  • webhookDevPlans Ready work items, implements In Progress, and fixes bugs. Pick Dev or SpecKit, never both.
  • webhookCode ReviewReviews every pull request for bugs, regressions, and missing tests. Verdict only. It never implements.
  • webhookQA Test GenerationDesigns and links test cases when a story reaches Ready for Testing. Does not execute them.
  • webhookQA Agent BrowserRuns those linked cases in a live browser. Skips if cases are not on the item yet.
  • webhookQA PlaywrightAuthors Playwright coverage for the linked cases and executes them in the same run.
  • cronSecurity ScannerWeekday Trivy and Semgrep. Files High+ bugs on the board.
  • cronSelf ImprovementDaily. Opens a policy PR from run feedback, then proposes that PR. Never patches live config.

Connect the repo. Enable an agent.

Sign in, set GitHub or Azure DevOps as the default repo, run Repo Initializer, then enable Dev or Code Review.

Sign in