DevManager-Agent

DevManager-Agent is an Agentic DevOps platform for engineering management and code quality governance. It analyzes repository changes, reviews code through specialized agents, tracks delivery health, and produces evidence-based engineering reports with human-governed automation.

The system is read-only by default. It is designed to support engineering managers and technical leads with traceable findings, deterministic scoring, audit records, and controlled approval boundaries.

Core Capabilities

• Repository ingestion with previous-successful-baseline tracking.
• Diff normalization, file classification, and change-risk analysis.
• Agent-assisted review for correctness, security, reliability, maintainability, and test risk.
• Structured finding validation, deduplication, and deterministic scoring.
• Engineering reports backed by commits, files, CI signals, audit events, and database records.
• Policy-governed automation with no production write actions unless explicitly approved.

Repository Layout

• apps/: API gateway, web UI, worker, orchestrator, report-service, and application entry points.
• services/: domain services such as PM (requirements platform) integration.
• packages/: shared Python, TypeScript, Java, database, Git, LLM, scoring, reporting, and contract modules.
• migrations/postgres/: PostgreSQL SQL migration scripts that are versioned and uploaded with the code.
• scripts/: local development, migration, and operational scripts.
• .github/: CI workflow configuration.

Product design documents, runtime logs, local caches, virtual environments, and frontend build artifacts are intentionally excluded from version control. Pytest test suites (under apps/, packages/, services/, and top-level tests/) are tracked in version control.

Technology Stack

• Python 3.12 with uv workspace management.
• FastAPI for the API gateway.
• ARQ and Redis for background jobs.
• PostgreSQL 16 for persistent storage.
• React, TypeScript, and Vite for the web UI.
• Docker Compose for local PostgreSQL and Redis services.

Database Migrations

SQL migrations are stored in migrations/postgres/ and are part of the uploadable source set. Apply pending migrations with:

uv run python scripts/migrate.py

The current migration set includes:

• 001_initial_schema.sql
• 002_add_clone_url.sql
• 002_add_repository_token.sql
• 003_add_run_repo_fullname.sql
• 004_create_settings.sql
• 005_add_llm_settings.sql
• 006_add_llm_base_url.sql
• 007_audit_events_workflow_id_nullable.sql
• 008_team_ops_foundation.sql
• 009_team_ops_mirror.sql
• 010_team_ops_derived.sql
• 011_team_ops_suggestion.sql
• 012_pm_sync_cursor.sql
• 013_team_ops_knowledge.sql

Testing

Run the full test suite (unit + integration):

uv run pytest

make test and the CI test job both delegate to the same command. Tests are collected from apps/, packages/, services/, and top-level tests/ (see [tool.pytest.ini_options].testpaths in pyproject.toml). Integration tests under tests/integration/ require a live PostgreSQL — the CI workflow applies migrations via scripts/migrate.py before running pytest.

Local Development

Install dependencies:

uv sync
cd apps/web && npm install

Start PostgreSQL and Redis with Docker Compose:

docker compose up -d postgres redis

Note on the PostgreSQL image: migrations/postgres/013_team_ops_knowledge.sql uses the vector extension (pgvector) for embedding columns and HNSW indexes. The stock postgres:16 image does not ship this extension. If you bring up PostgreSQL via docker compose, switch the postgres service image to pgvector/pgvector:pg16 (drop-in replacement that bundles the vector extension) before running migrations. The CI workflow already uses this image.

Run database migrations:

uv run python scripts/migrate.py

Start the local development stack:

./start.sh

Stop local services:

./stop.sh

Configuration

Create a local .env file from .env.example and adjust values for your environment. Local .env files are ignored by Git and must not be committed.

Common settings:

• DATABASE_URL
• REDIS_URL
• API_SECRET_KEY
• LLM provider credentials and base URLs, when enabled.

Safety Defaults

• No automatic production deployment.
• No automatic pull request merge.
• No unrestricted repository shell execution.
• No employee performance, compensation, promotion, or termination decisions.
• External comments, issue creation, remediation pull requests, and notifications require explicit policy approval.

License

This project is licensed under the MIT License. See LICENSE for details.