Skip to main content

Carrier-agnostic artifact governance and controlled-execution services that consume SCLite contract lifecycle artifacts.

Project description

GovEngine

CI: pytest Package: govengine 0.13.0 Python: 3.11+ Dependency: SCLite >=1.0.2 License: MIT

GovEngine is a carrier-agnostic deterministic governed-runtime kernel for portable artifact governance and policy-gated controlled execution.

It consumes SCLite as its contract lifecycle layer and provides reusable services around artifact state/transition boundaries, policy decisions, execution-contract shaping, execution-ticket checks, command-shape normalization, dry-run result assembly, and neutral runtime/control projections. Security-domain action, tool, scope, and signal behavior is host-owned; the published 0.12 alpha line removes the former Ravenclaw-derived compatibility helpers.

Why it exists

AI-assisted security workflows need a hard boundary between:

  1. what an agent or caller wants;
  2. what policy allows;
  3. what execution shape was prepared;
  4. what was approved;
  5. what was dry-run or executed;
  6. what evidence can be reviewed.

SCLite defines the auditable contract artifacts for that lifecycle. GovEngine is the reusable Python service layer that consumes those contracts and helps a host runtime enforce them without relying on prompt text alone.

Dependency direction

Ravenclaw -> GovEngine -> SCLite
  • SCLite owns schema-backed lifecycle artifacts and validation.
  • GovEngine owns reusable governed-execution helpers that consume SCLite artifacts.
  • Ravenclaw remains the reference security runtime/control plane and concrete integration host.
  • Tecrax is reserved as a future infrastructure-operations runtime/profile on the same foundation.

GovEngine is not Ravenclaw, Tecrax, Logdash, an LLM agent loop, a scanner, or a protocol adapter.

What GovEngine includes now

  • a public surface registry covering neutral artifact-governance, planning, admission/policy, evidence-review, domain-profile, runtime-proof, and controlled-execution surfaces;
  • serializable kernel/profile/runtime/SCLite boundary contracts and a machine-readable boundary report;
  • execution-contract shaping/redaction helpers;
  • artifact descriptor/state/transition boundary helpers;
  • SCLite lifecycle status bridge and lightweight lifecycle transition gate/controller;
  • guarded-root replay checks for already-verified SCLite Kernel Guard sidecars;
  • high-level guarded-strict verification plus replay-fresh runtime decisions;
  • artifact deconfliction/change-order helpers and lightweight state-index summaries;
  • signature/trust policy bridge helpers with host-provided signer/verifier ports and deterministic demo ports for fixtures;
  • approved-spec and execution-ticket validation helpers;
  • controlled execution gate helpers with dry-run as the default runner path;
  • command-shape helpers;
  • dry-run result assembly helpers;
  • deterministic orchestration handoff, governance event envelope, run-state, and between-step control-decision contracts;
  • neutral runtime-shell contracts for host control actions, queue snapshots, runtime snapshots, and scheduler-tick metadata;
  • neutral planning contracts for task, plan-intent, and planner-port handoffs;
  • neutral admission, policy, approval, and audit contracts for host runtime gates;
  • explicit SCLite integration seams;
  • focused standalone pytest coverage and GitHub Actions CI.

The 0.13.0 line also adds:

  • one admission decision you can actually read — a single RuntimeAdmissionResult record that summarizes whether a prepared request may proceed, what blocked it, and what to fix next; helpers compose and validate that record from separate policy, ticket, trust, guard, replay, runner, and receipt signals without running live work themselves;
  • replay freshness — remember which verified SCLite guarded roots were already used, so the same protected bundle cannot silently count as “fresh” twice;
  • receipt and evidence chain checks — confirm that a runner receipt still points at the right admission and ticket, and that later evidence or review references stay within the bounds of that receipt;
  • GovEngine-owned record signing for fixtures — deterministic digests and signed-record helpers for tests and reviewer demos, not production PKI;
  • a development-only audit trail adapter — append and verify a local hash-chained audit log during development, without claiming a production database;
  • runner safety posture — supervision helpers that keep dry-run as the default and treat an optional local subprocess runner as not ready until explicit host safety gates exist;
  • operator inspect without executingscripts/inspect_runtime_admission.py lets you read and summarize an admission record read-only, with no runner request, replay claim, audit write, or live execution.

What it intentionally does not include yet

  • live subprocess execution backend;
  • raw artifact storage/writes;
  • Logdash UI/API routes;
  • OpenClaw, MCP, A2A, or other protocol adapters;
  • LLM provider integrations;
  • Ravenclaw-specific personas, workspace state, or campaign UX;
  • production-readiness claims;
  • PKI, CA, KMS, key storage, or production identity proof;
  • a shipped LocalSubprocessRunner implementation (LocalSubprocessRunnerReadiness is a gating contract only);
  • production replay or audit persistence (ReplayClaimStore and JsonlAuditLedgerAdapter are host-owned or development-only adapters).

Current status

GovEngine is an alpha package 0.13.0 (0.13.0). It keeps the neutral artifact-governance, planning, admission/policy, controlled-execution, runner-supervision, runtime-shell, evidence-review, profile, and proof surfaces while removing the former optional security-profile facade and Ravenclaw-derived helper modules. The published dependency line is sclite-core>=1.0.2,<1.1.

Current roadmap direction

The governed-runtime MVP on main includes a canonical RuntimeAdmissionResult record as the bounded admission decision surface and compose_runtime_admission_result() as the neutral gate-summary composition helper. The helper composes prepared execution contract status, policy decision, execution ticket status, trust decision, guarded-strict SCLite verification when runtime-consumable, GovEngine replay freshness, runner profile, receipt obligation, blockers, next actions, and bounded artifact references into that record. normalize_admission_artifact_refs() is an alpha helper for bounded review references and existing digest strings; it does not compute content digests or claim SCLite canonicalization.

compose_runtime_admission_result() composes host-supplied gate summaries; it does not validate SCLite tickets, verify signatures, record replay state, or execute live work.

The operator-facing MVP flow is documented in docs/GOVERNED_RUNTIME_MVP_RUNBOOK.md. It ties admission, trust ports, guarded SCLite verification, replay freshness, runner profile, receipt obligation, and evidence/review binding into one inspectable dry-run/default-safe chain.

This roadmap does not make intent execution authority. It keeps profile/domain policy meaning, production identity, key management, operator authorization, raw evidence storage, and live backend behavior host-owned until explicit ports, negative tests, and safety gates justify any additional kernel surface.

Installation

Install the latest published package from PyPI:

python -m pip install govengine

GovEngine depends on the PyPI distribution sclite-core while preserving the Python import package sclite.

For local development:

python -m venv .venv
. .venv/bin/activate
python -m pip install -e '.[dev]'
python -m pytest -q
python scripts/validate_public_truth.py

Minimal smoke example

from govengine import public_surface_index
from govengine.execution.runner import approved_spec_dry_run_result

assert [surface.name for surface in public_surface_index()] == [
    "artifact_governance_core",
    "planning_contracts_core",
    "admission_policy_core",
    "evidence_review_core",
    "domain_profile_sdk",
    "runtime_contract_proofs",
    "controlled_execution_core",
]

receipt = approved_spec_dry_run_result(
    approved_execution_spec={
        "action_type": "bounded_request",
        "capability": "fixture_review",
        "resolved_tool": "fixture",
        "execution_mode": "dry_run",
    },
    planned_commands=[["fixture", "review"]],
)
assert receipt["status"] == "dry-run"

Documentation

License and provenance

GovEngine is MIT-licensed. It was extracted from Ravenclaw in contract-first stages, so LICENSE preserves the copyright notice for the originating Ravenclaw contribution lineage. The author metadata in pyproject.toml identifies the GovEngine package maintainer; it does not replace or reassign the originating copyright notice.

Safety boundary

GovEngine should preserve deterministic governance over prompt-only behavior. GovEngine must never execute directly from raw intent: execution requires a prepared execution contract, valid policy decision, approved execution ticket, valid signature/trust decision, and allowed runner profile. When a SCLite bundle is runtime-consumable, the execution gate also requires a guarded-strict SCLite verification result and replay-fresh GovEngine decision; review-only bundles can remain on weaker review/integrity postures without becoming execution authority.

DryRunRunner/dry-run behavior remains the default. Live execution backends are disabled by default; any future LocalSubprocessRunner must be optional, policy-enabled, negative-tested, and never the default. Controlled execution depends on lifecycle gates and signing/trust gates, with Ravenclaw retaining the concrete runtime adapter until reviewed. Demo signing helpers are fixture ports only: they bind a deterministic signature to an artifact digest for tests/reviewer demos and must not be presented as cryptographic identity, PKI, CA, KMS, or trust-store support.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

govengine-0.13.0.tar.gz (132.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

govengine-0.13.0-py3-none-any.whl (99.5 kB view details)

Uploaded Python 3

File details

Details for the file govengine-0.13.0.tar.gz.

File metadata

  • Download URL: govengine-0.13.0.tar.gz
  • Upload date:
  • Size: 132.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.4

File hashes

Hashes for govengine-0.13.0.tar.gz
Algorithm Hash digest
SHA256 6eab6ae1859697d5fde3e7b1811c3833ba8b69c817dec497cfa447548d21399c
MD5 dabaad3dfa3327feefa30748ee371781
BLAKE2b-256 be88cc5c9602f035cd7893690f0fc36cb680a49c4889d4d3618a9d9325dcf4e3

See more details on using hashes here.

File details

Details for the file govengine-0.13.0-py3-none-any.whl.

File metadata

  • Download URL: govengine-0.13.0-py3-none-any.whl
  • Upload date:
  • Size: 99.5 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.4

File hashes

Hashes for govengine-0.13.0-py3-none-any.whl
Algorithm Hash digest
SHA256 1ea6fad29dc0db05e318be2e8a5684e768da15c4c8b7ffcb9d21c008af645382
MD5 7635bbd02222ef5b466634a5f35988cc
BLAKE2b-256 5fd08accabb2a4b47bcab6c006d2a0dcc1f1cc15749235ff4ec76e84a9bdfb01

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page