mdx-rust

A Rust-native safe-change system for codebases.

mdx-rust points at Rust code, finds scoped hardening opportunities, validates changes in isolation, checks project policy and behavior evals when supplied, and only lands edits that pass Rust gates. It still supports agent optimization, but v0.5 is aimed at ordinary Rust crates and service backends too.

The CLI is the supported product surface. The library crates are published for installation and inspection, but their APIs remain unstable before 1.0.

Current Scope

mdx-rust is an early public beta. It is useful for experimentation and dogfooding on Rust agent crates, but it is intentionally conservative. In plain terms: v0.5 is good at review-first, safety-gated hardening of scoped Rust modules, simple policy mapping, deterministic behavior evals, plan-first refactor impact analysis, approved plan execution for low-risk candidates, and the existing agent prompt/fallback optimizer. It is not a general autonomous refactoring engine yet.

Today it supports:

• Rust-aware source analysis with syn and tree-sitter-rust.
• Process-based agent invocation with lifecycle traces.
• Prompt and AST-guarded fallback-behavior improvement strategies.
• Review-first scoped Rust hardening for normal modules through improve.
• Structured markdown policy parsing and policy-to-finding matches.
• Workspace behavior evals through .mdx-rust/evals.json.
• Optional behavior eval gates for improve --eval-spec.
• Repo doctor risk summaries with prioritized next actions.
• Plan-first refactor impact analysis with public API, module edge, long function, large file, and patchable hardening candidates.
• apply-plan execution for approved low-risk refactor candidates, with stale source snapshot rejection and all real edits routed through hardening transactions.
• apply-plan --all execution queues for reviewing or applying every executable low-risk candidate in a saved plan, with per-step validation.
• Bounded hardening transactions with all touched files snapshotted and rolled back on final validation failure.
• Isolated validation with cargo check and cargo clippy -- -D warnings.
• Net-positive scoring, final real-tree validation, and rollback on failure.
• Versioned audit packets for accepted optimizer changes and hardening reports for scoped module improvements.
• JSON Schema derivations for agent-facing records such as candidates, hooks, traces, eval datasets, audit packets, and validation command records.
• Human CLI output plus machine-parseable --json output.
• Deterministic static audit checks for risky agent surfaces.

What v0.5 Adds

v0.5 is the first release where mdx-rust becomes a guardrailed refactoring workflow instead of only a scoped hardening tool.

• Run mdx-rust plan <target> to produce a non-mutating refactor plan.
• Review impact before editing: public items, module edges, file size, long functions, policy references, behavior eval references, source snapshots, and candidate risk.
• Execute approved low-risk candidates with mdx-rust apply-plan --candidate <id>.
• Execute the whole safe queue with mdx-rust apply-plan --all.
• Get JSON artifacts for every plan and apply run so humans and agents can audit what happened.

The aggressive part is that apply-plan --all can now apply multiple approved low-risk changes from one plan. The disciplined part is that every real edit still goes through freshness checks, isolated validation, final validation, and the hardening transaction path.

Not yet supported:

• Arbitrary multi-file accepted edits outside the hardening transaction model.
• General autonomous refactoring.
• Direct application of stale plans or plan-only/high-risk candidates.
• Stable library APIs.
• Coverage, mutation testing, or full semantic behavior proofs.
• External hook runners.
• Multi-language optimization.

Safety Model

The acceptance contract is the center of the project:

1. Build a targeted ProposedEdit for one file.
2. Run pre-edit and pre-command hooks.
3. Apply the edit in an isolated workspace.
4. Run cargo check and cargo clippy -- -D warnings with timeouts.
5. Score the patched isolated workspace.
6. Require a strictly positive score delta.
7. Run pre-accept hooks.
8. Land the already validated edit on the real tree.
9. Run final validation on the real tree.
10. Roll back if final validation fails or times out.
11. Count the change as accepted only after landing and final validation pass.

The full non-bypass contract lives in SAFETY_INVARIANTS.md.

The implementation also uses typed rejection records and internal stage wrappers so accepted changes cannot be represented the same way as proposed or rejected candidates.

The hardening path for ordinary Rust modules is review-first by default: mdx-rust improve validates candidate changes in an isolated workspace and requires --apply before touching the real tree. In v0.5, passing --eval-spec also requires the behavior commands in that spec to pass in the isolated workspace and again after final application.

The refactor path is plan-first by design. mdx-rust plan never edits files. It writes a versioned plan artifact, classifies candidate risk, snapshots source hashes, surfaces public API impact, and identifies which candidates are executable. mdx-rust apply-plan can review or execute approved low-risk candidates, but it rejects stale source snapshots and still routes real edits through the existing hardening transaction gates.

For higher-leverage cleanup, mdx-rust apply-plan --all builds an execution queue from the saved plan, de-duplicates executable candidates by file, checks freshness before each step, and validates each applied step before continuing.

Quick Start

Install the CLI:

cargo install mdx-rust

Try the built-in example from a checkout:

git clone https://github.com/dhotherm/mdx-rust
cd mdx-rust

cargo run -p mdx-rust -- init
cargo run -p mdx-rust -- register example examples/rig-minimal-agent
cargo run -p mdx-rust -- optimize example --iterations 2
cargo run -p mdx-rust -- audit example
cargo run -p mdx-rust -- invoke example --input '{"query":"What is 9 + 10?"}'

For your own Rust agent:

cd your-rust-agent
mdx-rust init
mdx-rust register my-agent .
mdx-rust optimize my-agent --iterations 3 --budget medium --review

Artifacts are written under .mdx-rust/agents/<name>/.

For an ordinary Rust crate or backend module:

cd your-rust-service
mdx-rust init
mdx-rust doctor
mdx-rust audit --policy policies/backend-safety.md
mdx-rust eval --spec .mdx-rust/evals.json
mdx-rust improve src/api/config.rs
mdx-rust plan src/api
mdx-rust apply-plan .mdx-rust/plans/refactor-plan-...json --candidate <id>
mdx-rust apply-plan .mdx-rust/plans/refactor-plan-...json --candidate <id> --apply
mdx-rust apply-plan .mdx-rust/plans/refactor-plan-...json --all
mdx-rust apply-plan .mdx-rust/plans/refactor-plan-...json --all --apply
mdx-rust improve src/api/config.rs --eval-spec .mdx-rust/evals.json --apply

Hardening artifacts are written under .mdx-rust/hardening/. Refactor plan artifacts are written under .mdx-rust/plans/.

Behavior eval specs execute local commands from your repository. Treat them as trusted project code, review changes to them like test scripts, and prefer deterministic commands such as cargo test, golden CLI checks, or service contract smoke tests.

Key Commands

mdx-rust init
mdx-rust register my-agent ./path/to/agent
mdx-rust doctor
mdx-rust spec my-agent
mdx-rust doctor my-agent
mdx-rust audit --policy policies/backend-safety.md
mdx-rust audit my-agent
mdx-rust improve src/lib.rs
mdx-rust plan src/lib.rs
mdx-rust plan src/api --policy policies/backend-safety.md --eval-spec .mdx-rust/evals.json
mdx-rust apply-plan .mdx-rust/plans/refactor-plan-...json --candidate plan-hardening-src-lib-rs-2
mdx-rust apply-plan .mdx-rust/plans/refactor-plan-...json --candidate plan-hardening-src-lib-rs-2 --apply
mdx-rust apply-plan .mdx-rust/plans/refactor-plan-...json --all --max-candidates 10
mdx-rust apply-plan .mdx-rust/plans/refactor-plan-...json --all --apply --max-candidates 10
mdx-rust improve src/lib.rs --eval-spec .mdx-rust/evals.json --apply
mdx-rust eval --spec .mdx-rust/evals.json
mdx-rust eval my-agent --dataset .mdx-rust/agents/my-agent/dataset.json
mdx-rust optimize my-agent --iterations 3 --budget medium --review
mdx-rust schema audit-packet --json
mdx-rust schema hardening-run --json
mdx-rust schema behavior-eval-report --json
mdx-rust schema project-policy --json
mdx-rust schema refactor-plan --json
mdx-rust schema refactor-apply-run --json
mdx-rust schema refactor-batch-apply-run --json

Every command intended for automation supports --json.

Audit Packets And Hardening Reports

Accepted changes produce versioned JSON audit packets in the experiment directory. The optimizer 0.2 schema records:

• Agent name and iteration.
• Single-file edit scope contract.
• Accepted diff and diff hash.
• Dataset version and hash.
• Policy path and hash when available.
• Scorer id and version.
• Diagnosis model metadata and whether a live model was used.
• Hook decisions.
• Isolated and final validation command outcomes.
• Baseline, patched, delta, and holdout scores.
• Rollback status if rollback was attempted.

See docs/provenance.md for the schema contract. v0.4 and later hardening runs produce versioned JSON reports under .mdx-rust/hardening/ with findings, proposed changes, validation outcomes, transaction status, rollback status, policy matches, behavior eval outcomes, and workspace metadata.

v0.5 refactor plans produce versioned JSON reports under .mdx-rust/plans/ with impact summaries, source snapshot hashes, public API pressure, module edges, required gates, policy/eval references, and candidate actions. Plan artifacts are evidence for review and orchestration; they are not proof that a change has been applied. apply-plan reports are also written under .mdx-rust/plans/ and record whether a candidate or execution queue was reviewed, applied, rejected, stale, partially applied, or unsupported.

Print the current JSON Schemas with:

mdx-rust schema audit-packet --json
mdx-rust schema hardening-run --json
mdx-rust schema behavior-eval-report --json
mdx-rust schema refactor-plan --json
mdx-rust schema refactor-apply-run --json
mdx-rust schema refactor-batch-apply-run --json

API Stability

mdx-rust, mdx-rust-core, and mdx-rust-analysis are all published so the CLI can be installed from crates.io.

For 0.5.x:

• The mdx-rust CLI is supported.
• The mdx-rust-core and mdx-rust-analysis APIs are unstable.
• Public library types may change before 1.0.
• The intended facade is documented on docs.rs, but direct module usage is not a stability promise.

See docs/api-stability.md.

Project Docs

• SAFETY_INVARIANTS.md - acceptance loop and non-bypass rules.
• docs/architecture.md - module and lifecycle overview.
• docs/provenance.md - audit packet schema.
• docs/release-readiness.md - release gates and manual checks.
• ROADMAP.md - current scope and next work.
• CONTRIBUTING.md - development and safety expectations.

Contributor Rails

This repo uses a Justfile as the canonical local command surface:

just ci
just audit
just machete
just release-candidate

These commands mirror the public CI expectations and keep coding agents from guessing which checks matter.

Status

v0.5.0 is in development as the first guardrailed refactoring release. It adds impact analysis, stale-plan-resistant plan execution for low-risk candidates, and keeps broader refactors behind explicit review and future transaction work.

License

MIT