Cargo.toml

[workspace]
resolver = "2"
members = [
    "crates/plexus-core",
    "crates/plexus-probe",
    "crates/plexus-module",
    "crates/plexus-conformance",
]

# ck-plexus — the CortexKit connectors plane: a catalog-driven substrate for
# credentialed access to external services through MCP, REST, local scripts,
# and native helpers. This slice provides the pure manifest/storage domain core
# and the thin ck-plexus wire module, plus MCP transport/catalog discovery and
# the CKCRED route adapter.
#
# Shared dependencies come from crates.io. Update their pinned versions and
# this workspace's lock together, deliberately; CI checks out this repo alone.
[workspace.dependencies]
# From crates.io, matching the source subc-daemon uses. Two copies would each
# carry their own "installed logger" static, so the in-process test daemon would
# log through a copy no subscriber was ever installed on, and go silent.
cortexkit-log = "0.3.6"
tracing = "0.1"
# Exact pins, moved together on purpose. A caret on subc-daemon alone resolved
# 0.20.25, which requires protocol 0.25 and control 0.18, and the lock then held
# two copies of each wire crate: a type from one does not match the same type
# from the other, and the in-process test daemon would speak a different
# protocol from the module it tests. The pinned wire crates below all require
# protocol 0.33, which cortexkit-role-tool-provider 0.13 (the tool-result/v1
# format) requires. Client 0.32.14 needs protocol 0.33.5 and transport 0.14.1;
# daemon 0.38.0 needs control 0.33.3 and os 0.1.22. When the protocol moves,
# review and resolve this whole block together.
subc-protocol = "=0.33.5"
subc-os = "=0.1.22"
subc-transport = "=0.14.1"
subc-control = "=0.33.4"
subc-daemon = "=0.38.0"
subc-jsonc = "=0.1.2"
subc-client-rs = "=0.32.14"

# Exact pins, moved together. cortexkit-log already depends on the registry
# cortexkit-store-types; a second copy from anywhere else would be a distinct
# type to the compiler.
cortexkit-store = { version = "=0.2.5", features = ["sqlite"] }
cortexkit-store-types = "=0.2.2"

serde = { version = "1", features = ["derive"] }
serde_json = { version = "1", features = ["float_roundtrip"] }
# Reports WHICH field a deserialization error came from. serde_json's own error
# names the bad value and the expected type but never the path, so a one-field
# typo in a reviewed vendor manifest produced "invalid type: string, expected
# u32" with nothing saying where. Adds exactly one crate to the graph: its two
# dependencies (itoa, serde_core) are already present via serde_json.
serde_path_to_error = { version = "0.1" }
ryu-js = "1.0"
sha2 = "0.10"
hex = "0.4"
httpdate = "1"
zeroize = "1"
rusqlite = { version = "0.32", features = ["bundled"] }
tokio = { version = "1", features = ["rt-multi-thread", "macros", "net", "io-util", "sync", "time"] }

# The probe sandbox interprets guests with Wasmtime's Pulley interpreter (see
# crates/plexus-probe). Unoptimized, that interpreter is slow enough that the
# probe tests' heavy guest runs past the sandbox's 5 s deadline and reports a
# timeout instead of the memory ceiling it exists to test. Optimizing just the
# interpreter keeps test timings close to the release binary's; release builds
# are already optimized and unaffected.
[profile.dev.package.pulley-interpreter]
opt-level = 3

# The conformance harness runs subc-daemon in-process, so the daemon's
# provenance probe hashes the test binary itself. On Linux that is a full
# SHA-256 of an unstripped debug executable with embedded debug info, and
# unoptimized sha2 makes that first provenance request take far longer than
# the harness's control deadline. Optimizing sha2 in dev builds keeps it close
# to release speed; release builds are already optimized and unaffected.
[profile.dev.package.sha2]
opt-level = 3