chore(release): cut v0.9.0 — event-sourced kernel + runtime verbs

.claude-plugin/plugin.json

@@ -1,11 +1,25 @@
 {
   "name": "loop-engineer",
-  "version": "0.8.0",
+  "version": "0.9.0",
   "description": "Design, launch, verify, repair, and improve agent loops. A Claude-Code-native architect+operator for long-running, verifiable, self-improving agentic-coding systems.",
-  "author": { "name": "Sollan Systems", "url": "https://github.com/SollanSystems" },
+  "author": {
+    "name": "Sollan Systems",
+    "url": "https://github.com/SollanSystems"
+  },
   "homepage": "https://github.com/SollanSystems/loop-engineer",
   "repository": "https://github.com/SollanSystems/loop-engineer",
-  "keywords": ["agent", "loop", "agentic", "verification", "harness", "orchestration", "self-improvement", "claude-code", "eval", "repair"],
+  "keywords": [
+    "agent",
+    "loop",
+    "agentic",
+    "verification",
+    "harness",
+    "orchestration",
+    "self-improvement",
+    "claude-code",
+    "eval",
+    "repair"
+  ],
   "license": "MIT",
   "hooks": {
     "Stop": [

CHANGELOG.md

@@ -14,6 +14,68 @@ All notable changes to `loop-engineer` are documented here.
   `WORKFLOW.md` and `README.md` are reworded to describe the mechanism; the 0.3.4
   history is left intact.
 
+## 0.9.0 — 2026-07-17
+
+**The event-sourced kernel.** The contract gains a durable runtime substrate:
+`loop-engineer/event@1` events in an append-only SQLite store
+(`.loop/events.db` — WAL, `synchronous=FULL`, no-update/no-delete triggers,
+`expected_sequence` compare-and-swap) folded by a deterministic reducer that
+enforces FSM legality, the all-required completion gate, and terminal
+immutability through the same `loop.fsm`/`loop.completion` modules the
+writers use (#62). Around it: a canonical intermediate-state FSM with
+unknown-state validation and writer timestamps (#59); `loop-engineer/plan@1`
+— a Loop Plan IR with a `plan-lint` verb and a typed capability vocabulary
+(#61); `loop-engineer/evidence@1` — hashed evidence objects with provenance,
+re-verifiable byte-for-byte (#63); administrative `terminal_superseded`
+events — the only event a terminated run admits, so a wrong terminal is
+corrected on the record instead of edited in place (#64); a run-control event
+vocabulary for approval, pause, and resume (#73); and explicit validation
+modes `--mode basic|strict|release` across the CLI (#60).
+
+**Runtime verbs.** `loop run` — event-sourced single-step dispatch with
+crash-safe resume (#71); read-only `loop status` and `loop replay` (#70);
+`loop simulate` — strictly read-only dry-run dispatch prediction (#75);
+`approve` / `pause` / `resume` / `cancel` (#74); a subprocess-isolated
+verifier runner — shlex argv, no shell, wall-clock cap, bounded output tail,
+typed failure classes — plus typed fail-loud stubs for the not-yet-shipped
+run modes (#72); `loop architect` as a typed fail-loud deferral (#76). And
+`loop doctor` now composes an event-store consistency gate over
+`status`/`replay` whenever `.loop/events.db` exists — one hard gate across
+the file layer and the event log (#77).
+
+**Completion-semantics hardening (Phase 0).** `Succeeded` now requires
+**all** required criteria satisfied plus evidence — the prior any-true
+reading was a correctness bug, fixed in one shared `loop/completion.py`
+wired into the writer API, the integrations projection, and the contract
+validator's G1 check. `terminal_state.json` is create-once (atomic;
+`force` always raises); iteration ids are canonical ints (legacy decimal
+strings stay read-compatible); `terminal@1` gains an additive
+`completion_policy` (#48).
+
+**Adversarial hardening.** A property-based kernel suite over FSM legality,
+G1 completion, supersession, and replay determinism (#65), and a
+process/security suite — real crash injection at code-controlled barriers,
+verifier tampering, workspace-escape TOCTOU (#66). The `verify_evidence()`
+symlink-swap TOCTOU that suite discovered (#67) is closed by fd-pinned
+open-then-verify (#69), retiring the strict xfail marker.
+
+**Scoreboard, funnel, housekeeping.** The ST5 harness scoreboard — 9 public
+harnesses read through a foreign-layout registry with 8 vendored fixtures
+(#41); contributor-funnel fixes — trigger-phrase disambiguation (#42),
+documentation-completeness labels on the self-eval (#43),
+`approval_requested`/`replanned` recognized as honest-red metrics tokens
+(#44), metrics-clean RUNLOG seeding (#45); Dependabot version updates and
+least-privilege workflow permissions (#46, #47).
+
+**Known limitation.** Read verbs over the event store leave SQLite
+`-wal`/`-shm` sidecars next to `.loop/events.db` (read-only connections
+recreate them and cannot checkpoint on close). Store content is never
+mutated, but tree-byte-identity checks over a store-backed workspace will
+notice them.
+
+Test baseline: 933 passed / 16 skipped with the `yaml`+`schemas` extras;
+864 / 85 in structural-fallback mode (PyYAML only).
+
 ## 0.8.0 — 2026-07-09
 
 **ST3 — integration adapters.** `loop/integrations.py`: an engine-neutral,

README.md

@@ -5,7 +5,7 @@
 [![CI](https://github.com/SollanSystems/loop-engineer/actions/workflows/ci.yml/badge.svg?branch=main)](https://github.com/SollanSystems/loop-engineer/actions/workflows/ci.yml)
 [![Python 3.10–3.12](https://img.shields.io/badge/python-3.10%E2%80%933.12-blue)](https://www.python.org/downloads/)
 [![License: MIT](https://img.shields.io/badge/license-MIT-green)](LICENSE)
-[![Release](https://img.shields.io/badge/release-0.8.0-blue)](https://github.com/SollanSystems/loop-engineer/tags)
+[![Release](https://img.shields.io/badge/release-0.9.0-blue)](https://github.com/SollanSystems/loop-engineer/tags)
 
 Long-running agents commit **false completion**. After context compaction they
 forget what "done" meant, optimize to the visible test, patch in circles, and
@@ -20,7 +20,11 @@ ships, on disk and runnable today:
 - a **held-out gate** and an **anti-cheat scan** built to catch a loop gaming its
   own verifier;
 - a runnable example whose `false_completion: false` is backed by a committed,
-  real gate verdict — `.loop/artifacts/holdout-verdict.json`, not a hand-set flag.
+  real gate verdict — `.loop/artifacts/holdout-verdict.json`, not a hand-set flag;
+- an **event-sourced runtime** — `run`, `status`, `replay`, `simulate`, and
+  approve/pause/resume/cancel over an append-only SQLite event log
+  (`.loop/events.db`), folded by a deterministic reducer that enforces the same
+  completion gate as the writers, with crash-safe single-step resume.
 
 ![The inspector scores a self-asserted DIY loop 0/weak, then the gate-backed example 90/strong — both runs live](docs/demo.gif)
 
@@ -280,6 +284,12 @@ The portable core lives in `loop/` and validates schema-bearing artifacts in
 - `loop-engineer/state@1`
 - `loop-engineer/tasks@1`
 - `loop-engineer/terminal@1`
+- `loop-engineer/receipt@1`
+- `loop-engineer/repair@1`
+- `loop-engineer/rollout@1`
+- `loop-engineer/plan@1`
+- `loop-engineer/event@1`
+- `loop-engineer/evidence@1`
 
 ### Claude Code plugin
 
@@ -318,7 +328,7 @@ present. Recipe:
 **CI** — one workflow step validates the contract and publishes a scorecard:
 
 ```yaml
-- uses: SollanSystems/loop-engineer@v0.8.0
+- uses: SollanSystems/loop-engineer@v0.9.0
   with:
     path: "."
 ```
@@ -411,8 +421,8 @@ doctor` and the contract's own `verify-*` scripts).
 
 ## Status
 
-- Version: `0.8.0`
-- Release tag: `v0.8.0` (PyPI publish trigger; plugin tags through 0.6.0 used `loop-engineer--v<version>`)
+- Version: `0.9.0`
+- Release tag: `v0.9.0` (PyPI publish trigger; plugin tags through 0.6.0 used `loop-engineer--v<version>`)
 - License: MIT
 - Primary interface: Claude Code plugin
 - Portable core: Python CLI + JSON schemas

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "loop-engineer"
-version = "0.8.0"
+version = "0.9.0"
 description = "Portable Loop Contract Core: validate and inspect repo-native operating contracts for agent loops."
 readme = "README.md"
 requires-python = ">=3.10"

scripts/test_docs_version.py

@@ -15,9 +15,9 @@ def test_readme_has_no_stale_seven_skills():
     assert "all 9 skills" in readme
 
 
-def test_plugin_version_is_0_8_0():
+def test_plugin_version_is_0_9_0():
     plugin = json.loads(_read(".claude-plugin/plugin.json"))
-    assert plugin["version"] == "0.8.0"
+    assert plugin["version"] == "0.9.0"
 
 
 def test_pyproject_version_matches_plugin():
@@ -30,6 +30,7 @@ def test_pyproject_version_matches_plugin():
 
 def test_changelog_has_current_and_historical_entries():
     changelog = _read("CHANGELOG.md")
+    assert "## 0.9.0" in changelog
     assert "## 0.8.0" in changelog
     assert "## 0.7.0" in changelog
     assert "## 0.6.1" in changelog