dotnet-modernize 0.1.0

Agentic .NET Framework -> modern .NET upgrade skill, packaged with deterministic CLI helpers (baseline build capture, graphify knowledge-graph wiring, task scaffolding/tracking) and one-command installers for Claude Code, Cursor, Codex, and other agent tools.

dotnet-modernize

An agentic .NET Framework → modern .NET upgrade workflow (assessment → plan → build-validated execution, in the style of Visual Studio Copilot's "Modernize" agent), packaged for pip with:

• One-command install of the skill (SKILL.md + references/) into Claude Code's skill directory (or any other agent tool via --dest).
• Deterministic CLI helpers for the parts of the workflow that don't need an LLM: scenario scaffolding, baseline build capture, wiring up the graphify knowledge graph, and tracking tasks.md completion.

The reasoning-heavy stages — reading the codebase, deciding the upgrade strategy, diagnosing build errors — are not reimplemented in Python. That's what the bundled skill is for: point an agent (Claude Code, or any tool that can read a SKILL.md) at your solution and let it run the workflow, using this CLI underneath for the mechanical bits.

Install

pip install dotnet-modernize
# or: pipx install dotnet-modernize / uv tool install dotnet-modernize

Quickstart

cd path/to/your/solution

# 1. Put the skill where your agent will find it
dotnet-modernize install                 # -> .claude/skills/dotnet-modernize/

# 2. Scaffold an upgrade scenario
dotnet-modernize init dotnet-version-upgrade --target-tfm net10.0

# 3. Capture the pre-upgrade baseline (never edits code)
dotnet-modernize baseline dotnet-version-upgrade

# 4. Build the knowledge graph the agent will query during assessment/execution
dotnet-modernize graph --auto-install

# 5. Hand off to your agent:
#    "Upgrade this solution to net10.0 using scenario dotnet-version-upgrade,
#     following the dotnet-modernize skill."

# 6. Check progress any time (reads tasks.md, no agent needed)
dotnet-modernize status dotnet-version-upgrade

Commands

Command	What it does
install [--target claude-project|claude-user|generic] [--dest PATH] [--force]	Copies the bundled skill folder into an agent's skill directory.
uninstall	Removes a previously installed copy.
init SCENARIO_ID [--target-tfm] [--scope] [--solution-name]	Creates .github/upgrades/<id>/ with scenario-instructions.md and tasks.md skeletons.
list	Lists existing scenarios.
archive SCENARIO_ID	Renames a scenario folder with a timestamp so a new run can start clean without losing history.
baseline SCENARIO_ID [--solution PATH] [--no-tests]	Runs dotnet restore/build/test as-is and writes baseline-log.txt. Read-only w.r.t. source.
graph [--path] [--auto-install]	Runs graphify extract at the solution root; prints the report path.
query "<question>" [--path]	Queries an existing graph without an agent in the loop.
update-graph [--path]	Incrementally refreshes the graph after a committed task.
status SCENARIO_ID	Parses tasks.md checkboxes and shows completion %.
doctor	Checks for dotnet, git, graphify, and whether the skill/scenarios exist.

Why a hybrid (skill + CLI) instead of pure Python?

Upgrading a real solution requires judgment a script can't safely automate: classifying a WCF server as a rewrite, deciding bottom-up vs top-down order, root-causing a build error to the correct breaking-change entry instead of pattern-spraying fixes. That's delegated to an LLM agent reading the skill. Everything around that judgment — is git clean, does it build today, is the graph fresh, how many tasks are left — is exactly the kind of thing that should be a boring, deterministic, testable CLI command instead of something re-derived by the agent every session. This package draws that line explicitly instead of blurring it.

What's bundled

src/dotnet_modernize/skill_data/
├── SKILL.md
└── references/
    ├── artifact-templates.md
    ├── breaking-changes-catalog.md
    ├── project-type-playbooks.md
    ├── knowledge-graph.md
    ├── frontend-and-static-assets.md
    └── graphify/            # graphify's own canonical agent workflow, embedded
        ├── PIPELINE.md
        ├── query.md
        ├── update.md
        ├── extraction-spec.md
        ├── hooks.md
        ├── github-and-merge.md
        └── transcribe.md

License

MIT