baseplate

Status: Folding. See the component map.

baseplate is the one repo in the family with no architectural position of its own, and it is being dissolved into the components that actually use its contents — tracked in #4. It calls itself a substrate, but a 115-byte library surface with two consumers whose largest file is a cxpak MCP client is a grab-bag, not a floor. A client belongs with its server; verification types belong with the verifier.

Nothing is archived and no published version is yanked. Every release on crates.io stays available and anything depending on one keeps working. Destinations are still being decided per-module; this README is updated with the map once the moves land, rather than announcing them in advance.

A small, dependency-light crate of shared types and helpers, with two consumers. Publicly, attestr — the Verifier — pins baseplate = "0.2"; privately, dotclaude-core does the same. Nothing else in the family links it.

baseplate is deliberately not a tool with a behavioral contract of its own, and it is not the floor under the toolkit either — that is the claim #4 retracts. What is actually here: Promise-Theory verification value types, the trace/finding types that go with them, root/path resolution, a cxpak MCP client, YAML registry loading, and Java-test detection. Depend on it and use only the pieces you need.

Part of the Barnett Studios agentic-harness toolkit → cxpak · commitward · abproof · cascadr · cordon · slicr · corpus · attestr · baseplate

What's inside

Module	Responsibility
model	Promise-Theory verification value types — Observation (Kept/Broken/Partial/Skipped), Confidence, PromiseType, PromiseSpec, ReviewDecision. Not a model registry: there are no model identifiers, tiers or aliases here. The module name is a misnomer, and it is left in place rather than renamed because the module is moving out under #4.
trace	The shared trace/finding value types tools exchange (a turn's observed edits).
paths	Root resolution — $BASEPLATE_HOME-anchored, git-tree aware, with no hard-coded home.
registry	Loading and merging YAML promise/checkpoint registries with a stable schema. Unrelated to model.
cxpak	A thin rmcp client for the cxpak MCP server (child-process transport).
java_test	Detection of Java test files (unit *Test.java, integration *IT.java, system *SIT.java).
patterns	Shared regex primitives compiled once, reused across tools.

Use

[dependencies]
baseplate = "0.2"

While it is Folding, baseplate = "0.2" keeps working exactly as it does today. No release is yanked and no API is removed without a version bump.

use baseplate::{model, registry, trace};

// Load a YAML registry (promises/checkpoints) from an explicit path.
let reg = registry::load(std::path::Path::new("registry.yaml"), None)?;

Root resolution honours $BASEPLATE_HOME (the distribution-safe anchor for a binary installed outside a git tree), else the git tree the binary lives in, else the current directory — no hard-coded $HOME path. It pulls a small, boring dependency set (serde, regex, tokio, rmcp) and nothing language-model-specific.

The query CLI — the invocable ops as a one-shot container

baseplate's dominant value is the shared types other crates compile against — a role that has no runtime surface and stays a compile-time crate dependency. But a subset of its modules have genuinely-invocable ops, and those ship as a self-contained CLI (a cli feature) packaged as a container image, so any harness can use them without linking the crate. Each op reads a JSON request on stdin and writes an ADR-0052 envelope ({schema_version, status, body}) on stdout, network-free:

$ echo '{"paths":["src/test/java/x/FooTest.java","src/main/java/x/Foo.java"]}' \
    | docker run --rm -i --network none ghcr.io/barnett-studios/baseplate java-test analyze
{"schema_version":"1","status":"ok","body":{"results":[{"path":"src/test/java/x/FooTest.java","is_java_test":true},...]}}

Op	Request	Answers
java-test analyze	{paths: [String]}	which paths are Java test files
patterns match	{pattern, content, case_insensitive?}	does a regex compile, and where it matches
registry load	{registry_yaml, overrides_yaml?}	parse + summarize a promise registry (valid? version, promises)

An uncompilable pattern or invalid registry is a result (compiles: false / valid: false), not a failure; only a malformed request yields status: "error" + a non-zero exit, so a consumer falls back to its in-process path rather than trusting an empty result.

Honest scope. model/trace (shared types) and the cxpak client (needs a live MCP server) have no mount-free surface and are intentionally not exposed here — this image is baseplate's consumable face, not its primary role, which stays a linked crate. The image is standalone: no other component derives FROM it. Build the binary from source with cargo build --release --features cli; it's also on the Homebrew tap (brew install barnett-studios/tap/baseplate).

Stability

Pre-1.0: the surface may change between minor versions. The cxpak client tracks the cxpak MCP tool contract (op-parameterized intent tools); a breaking change there is called out in the release notes. Downstream tools pin a compatible minor.

License

Licensed under either of MIT or Apache-2.0 at your option. Unless you explicitly state otherwise, any contribution you intentionally submit for inclusion in the work shall be dual-licensed as above, without any additional terms.

---

Built by Barnett Studios — part of the agentic-harness toolkit. The one public crate that depends on baseplate is attestr: baseplate = "0.2" on every release from 0.2.0 to 0.4.1, and 0.1.0 on this crate's former name, dotclaude-support. The other consumer is private. Every other published version of cxpak, commitward, cascadr and abproof declares no baseplate dependency (checked against the crates.io index, all versions, not just the latest), and cordon, slicr and corpus are not Rust crates. They are siblings in the toolkit; none of them links this one. A footer claiming otherwise is what #4 opened over.