Skip to content

Changelog

All notable changes to the Animus monorepo will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Fixed — Truth Baseline (2026-06-27)

  • scripts/truth-baseline.py — check_test_count now fails when the shell command exits non-zero, even when expected is not set in the TOML. Previously, missing python in the environment produced a silent false positive (0 tests reported as PASS).
  • truth-baseline.toml — added expected = 1 and op = ">=" to core_tests and forge_tests checks, plus python3 fallback when python is unavailable.
  • New version_alignment check — reads all packages/*/pyproject.toml and pwa/package.json versions, reports mismatches. contracts/ documented as exception (no manifest). Surfaces actual misalignment: core 2.3.0, forge 1.9.0, bootstrap 0.8.0, kernel 0.1.0, quorum 1.2.0, types 0.1.0, pwa 0.1.0.

Fixed — Test Infrastructure (2026-06-27)

  • Root pyproject.toml — Added [tool.pytest.ini_options] with pythonpath covering all 8 package source directories, enabling pytest to discover sibling packages when run from repo root.
  • truth-baseline.toml — core_tests and forge_tests commands updated to run from repo root with -c pyproject.toml flag, bypassing per-package config files that lacked sibling package paths.
  • Results: core 2,832 tests (was 0), forge 10,431 tests (was 9,425 with 25 errors), kernel 99 tests (was 0), quorum 961 tests (was 861 with 2 errors), types 67 tests (was 22 with 3 errors). Truth baseline now 11/12 PASS (only version_alignment remains FAIL, documented).

Added — Version Compatibility Matrix (2026-06-27)

  • COMPATIBILITY_MATRIX.md — Documents actual versions across all 8 packages, inter-package dependencies, known incompatibilities, and verification command. Includes dependency graph and SemVer bump guidance.
  • Truth baseline updated: compatibility_matrix check added (PASS when file exists). version_alignment expected to FAIL until unified versioning is adopted; mismatches are now documented rather than hidden.

Added — Schema Compiler (2026-06-27)

  • scripts/compile_schemas.py — New script using datamodel-code-generator to auto-generate Pydantic v2 models from packages/contracts/*.schema.json.
  • Generated 20 schema modules in packages/types/src/animus_types/ (action.py, event.py, assessment.py, etc.) plus common.py with shared enums (ArtifactType, SubjectDomain, SecurityClass, etc.) and base Common model.
  • packages/types/src/animus_types/__init__.py — Updated to re-export all generated models + legacy types.
  • packages/types/tests/test_schemas.py — 22 tests: importable check for all 20 schemas + round-trip validation for action and event.
  • packages/contracts/pyproject.toml — New package manifest, making contracts a real installable package.

Added — Scaffold Cleanup (2026-06-27)

  • Root contracts/ deleted — Empty directory removed; canonical schemas live in packages/contracts/.
  • READMEs added to apps/, modules/, database/, infra/, evidence/releases/ — each explains what belongs there, boundary vs packages/, and owner.
  • ADRs written: ADR-002 (Audience-Based Docs Tree), ADR-003 (Schema Compiler), ADR-004 (Truth Baseline Fix), ADR-005 (Kernel Extraction).
  • truth-baseline.toml updated: removed contracts/ from root_architecture_dirs, added root_architecture_readmes and adr_coverage checks.

Added — Evidence Bundle MVP (2026-06-27)

  • scripts/assemble_evidence_bundle.py — New script that collects test counts, validates JSON schemas, captures git provenance, locks dependencies, and produces a human-readable markdown report with pass/fail badges.
  • Output: timestamped directory in evidence/releases/evidence-YYYY-MM-DD-HHMMSS/ containing manifest.json, test-results.json, schema-validation.json, git-info.txt, dependencies.lock, report.md.
  • Flags: --output-dir for custom path, --allow-dirty to permit uncommitted changes.
  • Integration: documented in docs/operators/deployment.md as a pre-release step.

Added — Documentation Reorganization (2026-06)

  • Audience-based docs tree — docs/ reorganized from flat graveyard (44 files, no index) into structured lanes:
  • getting-started/ — quickstart, installation, concepts
  • architecture/ — overview (rewritten for v2.1, 8 planes), packages, decisions, standards
  • packages/ — per-package documentation with version matrix
  • contributing/ — setup, workflow, debugging
  • operators/ — deployment, configuration, monitoring, troubleshooting
  • reference/ — glossary, FAQ, changelog, security, whitepapers
  • roadmap/ — current priorities and plans
  • 14 new placeholder files written with minimal viable content across all doc lanes
  • docs-validate.py — CI validation script for internal links, anchor references, trailing whitespace, and redirect stubs
  • Test count reconciliation — Core 2,865, Forge 10,304, Bootstrap 2,048, Quorum 961 (total 16,178)
  • Gorgon→Forge naming cleanup in packages/forge/ (tests/README, skills/README, 6 SKILL.md files)
  • Bootstrap config reference expanded in docs/operators/configuration.md — full schema covering identity, ollama, gateway, intelligence, channels, self-improvement, proactive, personas
  • Package READMEs written for Quorum, PWA, Contracts (previously missing)
  • Whitepaper index populated at docs/reference/whitepapers/README.md
  • MkDocs site — Material theme, dark/light mode, search, code copy buttons. Deploys to GitHub Pages via .github/workflows/docs-deploy.yml on every push to main that touches docs

Changed

  • docs/architecture/overview.md — rewritten from aspirational 4-layer hardware diagram to v2.1 reality (8 technical planes, no stale hardware references)
  • .github/workflows/ci.yml — Docs Validation job now calls scripts/docs-validate.py instead of inline Python/shell
  • Root README.md — ASCII tree expanded to show all 8 packages, nav links updated to new docs tree

Fixed

  • 6 broken internal links in docs/reference/faq.md and glossary.md (path depth errors)
  • Trailing whitespace in 29 session-modified files
  • Stale test counts understated by ~11% across docs/packages/README.md and root README

Added — Quorum v2 Week 1: EventLog bitemporal + signal bridge

  • Bitemporal-lite on CoordinationEvent — valid_from (world-time) and recorded_at (observation-time) fields ported from memboot's pattern. Backward-compatible: timestamp preserved as alias.
  • Idempotent SQLite migration — EventLog._migrate() adds the two columns and backfills both from timestamp on legacy databases; safe to re-run.
  • EventLog.query() bitemporal range filters — valid_from_since / valid_from_until for world-time windows; since / until continue to filter by observation-time.
  • Signal bridge — EventLog(signal_bus=bus) mirrors every recorded event onto the bus as Signal(signal_type=f"event.{event_type.value}"). Best-effort; failures swallowed so bus issues never break record().
  • 4 of 5 mutation sites wired to EventLog.record() — closes Constitutional Principle P3 (Transparency) on INTENT_PUBLISHED (resolver), VOTE_CAST (triumvirate), DECISION_MADE (triumvirate), MARKER_LEFT (stigmergy). SCORE_UPDATED and INTENT_RESOLVED defer to Week 3-4 scorer registry.
  • Quorum test count: 920 → 957 (+37 tests across 4 new test files).

See docs/specs/quorum_v2_week1_event_log_extension.md and ADL-20260510-001.

[2.2.0] - 2026-03-05

Added

  • Dual-model routing — Claude as brain (planning, code gen) + Ollama as hands (summarization, formatting)
  • Task classification — classify_task() heuristic routes HEAVY vs LIGHT tasks to appropriate model
  • Autonomous build pipeline — 4-agent Forge workflow (planner → coder → verifier → fixer) with quality gates and $2.00 budget cap
  • Constrained tool selection — Numbered menu + key:value parsing for reliable Ollama tool use
  • Task outcome tracking — Records outcomes in MemoryLayer, recalls similar past tasks, detects failure patterns, tracks success rates
  • New archetypes — planner, coder, verifier added to ForgeAgent
  • New slash commands — /build, /model, /stats in chat agent
  • create_local_think_tool() — Lets Claude offload cheap subtasks to Ollama during agentic loop
  • Core test count: 1879 → 2046 (+167 tests)

Changed

  • think_with_tools() dispatch: non-Anthropic models now route to constrained loop instead of markdown loop
  • chat.py fully wired to Animus Core (CognitiveLayer, ToolRegistry, MemoryLayer, TaskOutcomeTracker)

[2.0.0] - 2026-02-20

Added

  • Monorepo consolidation — Gorgon and Convergent merged into Animus as Forge and Quorum packages
  • Four independently installable packages: Core, Forge, Quorum, Bootstrap
  • Monorepo CI pipeline with per-package test jobs, security scanning, and benchmarks
  • Bootstrap package (v0.5.0): install daemon, onboarding wizard, dashboard, gateway, intelligence layer, persona system

Changed

  • Forge: test_ai module renamed to animus_forge
  • Quorum: keeps convergentAI PyPI name, imports as convergent
  • Core: lightweight forge/swarm sub-engines for embedded orchestration

Fixed

  • OAuth test patch target and race condition
  • Forge async test failures (stale from prior codebase)

[1.0.0] - 2026-02-20

Added

  • Forge/Swarm revise gate loop-back with ReviseRequestedError
  • Register translation (LLM-based formal/casual/technical)
  • Native Anthropic tool_use in cognitive layer
  • /workflow command in chat.py for running Forge YAML workflows

Fixed

  • OAuth test patching correct target
  • datetime.utcnow() deprecation warnings

[0.7.0] - 2026-02-19

Added

  • Swarm parallel agent orchestration with stigmergic coordination
  • DAG analysis via Kahn's topological sort with parallel execution stages
  • Stage-level atomic checkpoints
  • Coverage hardening: 584 to 1475 tests (46% to 91%)

Fixed

  • Gitleaks shallow clone issue (fetch-depth: 0)
  • Python 3.10 compatibility (datetime.UTC to timezone.utc)
  • 30 CodeQL alerts resolved to 0

[0.6.0] - 2026-02-19

Added

  • Forge multi-agent orchestration MVP
  • YAML workflows with agent archetypes, token budgets, quality gates
  • SQLite WAL checkpoints for workflow state
  • Google Calendar and Gorgon integrations
  • Sync protocol and learning system