Skip to content

causal-memory-layer 0.4.0 — package-validation-ready CML prototype

Choose a tag to compare

View all tags
@safal207 safal207 released this 07 Jun 18:26
· 54 commits to main since this release
v0.4.0
7699cfc

causal-memory-layer 0.4.0 — package-validation-ready CML prototype

This is the first production PyPI release candidate for causal-memory-layer.

CML is an open-source causal audit layer for structured action traces. It focuses on checking whether an action has a valid causal permission / responsibility lineage, not only whether the action happened.

Logs show what happened. CML checks why it was allowed.

Install

pip install causal-memory-layer==0.4.0

Then:

cml --help

For experimental MCP support:

pip install "causal-memory-layer[mcp]==0.4.0"

What is included

  • Core CML audit engine for structured causal records.
  • CLI entry point: cml.
  • Causal chain reconstruction helpers.
  • Audit findings for invalid lineage patterns.
  • Deterministic safety benchmark fixtures and runner.
  • Experimental Cause Band helpers.
  • Experimental MCP agent-audit integration path.
  • Friendly optional dependency guard for cml-mcp when [mcp] is not installed.

Validation evidence

Package validation has passed:

https://github.com/safal207/Causal-Memory-Layer/actions/runs/27063042787/job/79879380002

TestPyPI publication and clean install smoke test have passed:

https://github.com/safal207/Causal-Memory-Layer/actions/runs/27069357386

Confirmed before production release:

  • tests passed,
  • deterministic safety benchmark passed,
  • package build passed,
  • metadata validation passed,
  • wheel install validation passed,
  • TestPyPI publication passed,
  • TestPyPI install smoke test passed,
  • cml --help worked,
  • import cml worked,
  • MCP core import worked,
  • core.health() worked.

Current benchmark snapshot

6/6 matched

See:

docs/evidence/BENCHMARK_EVIDENCE_SNAPSHOT.md

Experimental MCP integration

The package includes an experimental MCP path for agent-audit workflows.

Local development:

pip install -e ".[mcp]"
cml-mcp

Packaged install:

pip install "causal-memory-layer[mcp]==0.4.0"
cml-mcp

MCP tool logic includes:

health
audit_trace
evaluate_cause_band

Known limitations

  • CML is a focused causal-validity audit prototype, not a complete runtime safety platform.
  • Cause Band semantics are experimental and not stable CML/vCML semantics.
  • MCP integration is experimental and local-first.
  • External validation notes are still being collected.
  • The current deterministic benchmark is intentionally small; broader benchmark expansion is future work.
  • This release does not provide a hosted service by itself.

Non-claims

This release does not claim:

  • production AI safety,
  • regulatory compliance,
  • runtime enforcement guarantees,
  • complete jailbreak detection,
  • complete security coverage,
  • stable Cause Band semantics,
  • certification readiness,
  • future prediction,
  • quantum security guarantees.

The correct narrow claim is:

CML 0.4.0 is a package-validated, installable causal-validity audit prototype for structured action traces.

Release docs

Full release notes draft:

docs/release/RELEASE_NOTES_0.4.0.md

Production checklist:

docs/release/PRODUCTION_PYPI_RELEASE_CHECKLIST_0.4.0.md
Assets 2
Loading