auto-mdp-solver — the harness

The Python toolchain behind the auto-mdp-solver Claude Code plugin: a pipeline that turns a verbal dynamic-decision problem into a trained, deployable RL policy. This package is the executable half — the IR schema and interpreter, the gates, and the tuning driver. The plugin (skills, spec, examples) ships separately and installs this package into the workspace venv on first run.

package	what it is	entry point
`mdp_ir`	MDP-IR schema (pydantic), reference interpreter, catalog ⊕ selection layering, execution laws, bit-exact differential runner	`python -m mdp_ir <schema>`, `python -m mdp_ir.laws <dir>`, `python -m mdp_ir.differential <schema>`
`mdp_conformance`	spec-conformance checks on a generated domain	`python -m mdp_conformance <dir>`
`mdp_gates`	eval-gate comparison of a candidate against baselines under the shared seed protocol	`python -m mdp_gates`
`mdp_tuning`	Optuna tuning driver over a domain's training script	`python -m mdp_tuning <dir> ...`
`mdp_stage`	the pipeline's entry gates — where a domain folder stands, and whether an op may start	`python -m mdp_stage <dir> [--for <op>]`

Install

pip install "auto-mdp-solver[domain]"   # everything a generated domain needs (SB3 + torch + pandas + pytest)
pip install auto-mdp-solver             # harness only, torch-free: IR, conformance, gates, tuning
pip install "auto-mdp-solver[dev]"      # harness + pytest
# the same, straight from the repo at a release tag (no PyPI needed):
pip install "auto-mdp-solver[domain] @ git+https://github.com/tong-wang/auto-mdp-solver@v0.10.17#subdirectory=harness"

Requires Python ≥ 3.12. The version number tracks the plugin release it was cut from (v0.10.17 ↔ plugin 0.10.17), so the skills and the harness a workspace holds are comparable by one number.

Documentation, the per-domain spec, the frozen example domains and the test cases live in the repository. MIT license.

Release files for auto-mdp-solver 0.10.17

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for auto-mdp-solver 0.10.17
File	Size	Uploaded
auto_mdp_solver-0.10.17.tar.gz	163.4 kB	Sep 17, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for auto-mdp-solver 0.10.17
File	Interpreter	ABI	Platform	Reset
auto_mdp_solver-0.10.17-py3-none-any.whl 177.2 kB Sep 17, 2026	Python 3	none	any	Details

Total release size: 340.6 kB

Release files / auto_mdp_solver-0.10.17.tar.gz

Download URL	auto_mdp_solver-0.10.17.tar.gz
Size	163.4 kB
Tags	Source
SHA-256 checksum How to use checksums	`414feaf801b69540e5e4c76b9911018c91ce06b1035e1f27b1cb522c3e89636d`
BLAKE2b-256 checksum How to use checksums	`aed1abce8dc0fa4e78c9e48084a718879a5cb7c06f46d191dd1eb9553c8c3bde`
Upload date	Sep 17, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.12.13`

Release files / auto_mdp_solver-0.10.17-py3-none-any.whl

Download URL	auto_mdp_solver-0.10.17-py3-none-any.whl
Size	177.2 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`b34de5bc82f3017c5b37efa9f3552cdfd87fa04dbb4d8c8ce2c293c5b87e42e7`
BLAKE2b-256 checksum How to use checksums	`7cd284cb8ef3308a6a687c79acc2dc04d62f94c092afb8142f6f4d51a9833e3c`
Upload date	Sep 17, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.12.13`