loop-engineer 0.7.0__tar.gz → 0.9.0__tar.gz

{loop_engineer-0.7.0 → loop_engineer-0.9.0}/.claude-plugin/plugin.json

@@ -1,11 +1,25 @@
 {
   "name": "loop-engineer",
-  "version": "0.7.0",
+  "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": [

loop_engineer-0.9.0/.github/dependabot.yml

@@ -0,0 +1,16 @@
+version: 2
+updates:
+  # Covers .github/workflows/*.yml and the root composite action.yml.
+  - package-ecosystem: "github-actions"
+    directory: "/"
+    schedule:
+      interval: "weekly"
+    groups:
+      actions:
+        patterns: ["*"]
+
+  # Runtime is stdlib-only; this watches the optional extras and any future pins.
+  - package-ecosystem: "pip"
+    directory: "/"
+    schedule:
+      interval: "weekly"

{loop_engineer-0.7.0 → loop_engineer-0.9.0}/.github/workflows/ci.yml

@@ -5,6 +5,9 @@ on:
     branches: [main]
   pull_request:
 
+permissions:
+  contents: read
+
 jobs:
   gates:
     name: gates (py${{ matrix.python-version }})
@@ -14,14 +17,14 @@ jobs:
       matrix:
         python-version: ["3.10", "3.11", "3.12"]
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
 
-      - uses: actions/setup-python@v5
+      - uses: actions/setup-python@v6
         with:
           python-version: ${{ matrix.python-version }}
 
       - name: Install gate dependencies
-        run: python -m pip install --upgrade pip pyyaml pytest jsonschema
+        run: python -m pip install --upgrade pip pyyaml pytest jsonschema hypothesis
 
       - name: Frontmatter gate
         run: python -B scripts/validate_frontmatter.py
@@ -45,27 +48,55 @@ jobs:
 
       - name: Quickstart smoke test
         run: |
-          python -B -m loop doctor examples/coverage-repair
+          python -B -m loop doctor --mode release examples/coverage-repair
           python -B -m loop inspect examples/coverage-repair
 
+      - name: Plan lint smoke test
+        run: |
+          python -B -m loop plan-lint --mode release examples/plans/coverage-repair.plan.json
+          if python -B -m loop plan-lint --mode release examples/plans/invalid/cyclic-dependency.plan.json; then
+            echo "::error::cyclic-dependency.plan.json unexpectedly passed plan-lint"
+            exit 1
+          fi
+          if python -B -m loop plan-lint --mode release examples/plans/invalid/missing-goal.plan.json; then
+            echo "::error::missing-goal.plan.json unexpectedly passed plan-lint"
+            exit 1
+          fi
+
   recipe-langgraph:
     name: recipe (langgraph)
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
-      - uses: actions/setup-python@v5
+      - uses: actions/checkout@v7
+      - uses: actions/setup-python@v6
         with:
           python-version: "3.12"
       - name: Install recipe dependencies
         run: python -m pip install --upgrade pip pyyaml pytest jsonschema langgraph
       - name: LangGraph recipe end-to-end
-        run: python -B -m pytest -q -p no:cacheprovider scripts/test_langgraph_recipe.py
+        run: python -B -m pytest -q -p no:cacheprovider scripts/test_langgraph_recipe.py scripts/test_langgraph_recipe_st3.py
+
+  recipe-temporal:
+    name: recipe (temporal)
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: actions/setup-python@v6
+        with:
+          python-version: "3.12"
+      - name: Install recipe dependencies
+        run: python -m pip install --upgrade pip pyyaml pytest jsonschema temporalio
+      - name: Temporal recipe end-to-end
+        # start_local() downloads the Temporal dev-server binary (~20MB) to
+        # $TMPDIR/temporal-sdk-python-<version> — ephemeral and quick, so (like
+        # recipe-langgraph) this job carries no cache step.
+        run: python -B -m pytest -q -p no:cacheprovider scripts/test_temporal_recipe.py
 
   action-dogfood:
     name: action (dogfood on flagship example)
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
       # The repo root's live .loop/ is gitignored, so it does not exist in a fresh
       # CI checkout — the tracked flagship example is the contract the action gates
       # (doctor-clean, inspect 90/strong in the action's dependency-free install).

{loop_engineer-0.7.0 → loop_engineer-0.9.0}/.github/workflows/publish.yml

@@ -4,13 +4,16 @@ on:
   push:
     tags: ["v*"]
 
+permissions:
+  contents: read
+
 jobs:
   build:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
 
-      - uses: actions/setup-python@v5
+      - uses: actions/setup-python@v6
         with:
           python-version: "3.12"
 
@@ -38,7 +41,7 @@ jobs:
           /tmp/smoke/bin/loop doctor /tmp/smoke-loop
           /tmp/smoke/bin/loop inspect /tmp/smoke-loop || true
 
-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@v7
         with:
           name: dist
           path: dist/
@@ -50,7 +53,7 @@ jobs:
     permissions:
       id-token: write
     steps:
-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@v8
         with:
           name: dist
           path: dist/

{loop_engineer-0.7.0 → loop_engineer-0.9.0}/.gitignore

@@ -1,19 +1,20 @@
-# Ephemeral workbench
-.tmp/
+# Ephemeral workbench (root-anchored: examples/*-run fixtures may legitimately
+# contain same-named dirs — .claude/, .taskmaster/, memory/ — that MUST track)
+/.tmp/
 
 # Loop run telemetry (operating-contract state for the self-improvement run; not plugin content)
-.loop/
+/.loop/
 
 # Machine-local Claude Code config + GSD audit telemetry (not plugin content)
-.claude/
-.gsd/
+/.claude/
+/.gsd/
 
 # Local session memory / private build narrative (not plugin content)
-memory/
+/memory/
 
 # Workbench dirs (review scratch + roadmap drafts; not plugin content)
-review/
-roadmap/
+/review/
+/roadmap/
 
 # Python
 __pycache__/
@@ -33,3 +34,6 @@ Thumbs.db
 # Secrets — never commit
 .env
 *.local
+
+# Local cold-start context — deliberately untracked (see its own do-not-commit note)
+/CLAUDE.md

{loop_engineer-0.7.0 → loop_engineer-0.9.0}/CHANGELOG.md

@@ -14,6 +14,99 @@ 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,
+pure-stdlib projection (`EngineOutcome` + `to_terminal_state`) from any
+engine's "the run ended" signal onto the 7 typed terminal states, with the
+fixed precedence safety → human → blocked → budget → spec-gap → gate verdict.
+`Succeeded` is reachable only through a green `holdout_gate.decide` verdict,
+a clean anticheat sweep, a met criterion, and evidence; `false_completion` is
+copied from the gate, never synthesized; missing gate/anticheat input fails
+closed to `FailedUnverifiable`. The LangGraph recipe is upgraded in place to
+this bar (its run now scores clean under `loop metrics` — closing the
+recorded FCR-1.0 follow-up) and a Temporal recipe lands
+(`examples/temporal-certify/`, certify-activity pattern, cancellation →
+`AbortedByHuman`, retry exhaustion → `FailedBlocked`, timeout →
+`FailedBudget`). Both recipes pin the false-completion invariant
+(visible-green/holdout-red → `FailedUnverifiable` with
+`false_completion: true`, never `Succeeded`) and pass the doctor round-trip.
+
+**ST4 — contributor funnel.** `loop inspect` now recognizes a foreign
+Superpowers-style run dir read-only (`loop/foreign.py` — a layout mapper onto
+the existing `LoopPaths` seam; the M2/M3-hardened scorer is untouched and a
+foreign harness with no gate and no terminal record scores honestly low).
+The reading is checked in as `docs/gap-reports/superpowers.md` — the §14
+conformance checklist evaluated against a vendored, sanitized fixture
+(`examples/superpowers-run/`). A second runnable example lands:
+`examples/flaky-test-triage/` — doctor-clean, gate-backed, and the showcase
+for repair records (`loop metrics` derives a non-null repair-productivity of
+1.0 from its anchored red→green repair). Seven gate-backed starter issues are
+drafted under `docs/contributing/issues/` and filed at release;
+CONTRIBUTING gains the start-here funnel.
+
 ## 0.7.0 — 2026-07-08
 
 **ST2 — the portable standard.** The on-disk contract is now a documented,

{loop_engineer-0.7.0 → loop_engineer-0.9.0}/CONTRIBUTING.md

@@ -11,7 +11,7 @@ you open a PR, all of these must be green (CI runs them on every push):
 ```bash
 # from the repo root
 python3 scripts/validate_frontmatter.py     # SKILL.md frontmatter
-python3 scripts/self_eval.py                # 13 structural invariants
+python3 scripts/self_eval.py                # 13 structural/doc-completeness invariants
 python3 -m pytest -q scripts                # the test suite
 python3 -m py_compile loop/*.py scripts/*.py
 python3 -m loop doctor   examples/coverage-repair   # quickstart still works
@@ -20,6 +20,35 @@ python3 -m loop inspect  examples/coverage-repair
 
 If you don't have the deps, prefix with `uv run --with pyyaml --with pytest`.
 
+`scripts/test_adversarial_kernel.py` also uses the dev-only `hypothesis` dependency: install it with `pip install -e ".[dev]"`, or use `uv run --with hypothesis --with pytest`. A plain install never pulls it in; without it, those property tests skip rather than fail.
+
+`scripts/test_adversarial_process.py` spawns short-lived child processes and sends real `SIGKILL` signals to simulate mid-transaction crashes; it is POSIX-only (CI uses `ubuntu-latest`), requires no new dependency, and adds a few seconds of subprocess wall time.
+
+Most `self_eval.py` checks verify documentation-completeness — that the suite's
+canonical vocabulary (terminal states, repair-record fields, eval layers +
+metrics) is present in the skill prose — not that a running loop enforces it. The
+runtime/behavioral gate is `python3 -m loop doctor` and the contract's own
+`verify-*` scripts, which is why both appear in the list above.
+
+## Start here — the contributor funnel
+
+Every open starter issue names **the gate that proves the fix** — a
+deterministic command that is red before your change and green after. That is
+the whole review bar (see the ground rule above).
+
+- [`good first issue`](https://github.com/SollanSystems/loop-engineer/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22)
+  — small, bounded, gate-verifiable fixes.
+- [`help wanted`](https://github.com/SollanSystems/loop-engineer/issues?q=is%3Aissue+is%3Aopen+label%3A%22help+wanted%22)
+  — integration recipes (OpenHands, ruflo — designs already written in
+  `docs/superpowers/specs/2026-06-30-st3-integration-adapters.md`) and
+  foreign-harness gap reports (`docs/gap-reports/`).
+
+The contribution target for anything that emits or consumes contract
+artifacts is the standard: `reference/repo-os-contract.md` — a harness that
+satisfies the §14 conformance checklist (A1–E1) may claim it emits a
+Loop-Engineer-conformant contract v1. Drafts for the seeded issues live in
+`docs/contributing/issues/` and are filed on GitHub at release time.
+
 ## Repository layout
 
 | Path | What lives here |
@@ -34,7 +63,10 @@ If you don't have the deps, prefix with `uv run --with pyyaml --with pytest`.
 
 ## Authoring or editing a skill
 
-The `self_eval.py` checks encode hard rules — match them or the gate fails:
+The `self_eval.py` checks encode hard rules — match them or the gate fails. Most
+are documentation-completeness checks (the canonical vocabulary must appear in the
+skill prose), not behavioral enforcement of a running loop — that gate is `loop
+doctor` and the contract's own `verify-*` scripts:
 
 - **Frontmatter** must have `name:` and `description:`. The directory name **must equal**
   the frontmatter `name`. Keep the `description` a *quoted* YAML scalar (the suite quotes

{loop_engineer-0.7.0 → loop_engineer-0.9.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: loop-engineer
-Version: 0.7.0
+Version: 0.9.0
 Summary: Portable Loop Contract Core: validate and inspect repo-native operating contracts for agent loops.
 Project-URL: Homepage, https://github.com/SollanSystems/loop-engineer
 Project-URL: Repository, https://github.com/SollanSystems/loop-engineer
@@ -29,6 +29,8 @@ License: MIT License
 License-File: LICENSE
 Keywords: agent,agentic,claude-code,eval,harness,loop,orchestration,repair,self-improvement,verification
 Requires-Python: >=3.10
+Provides-Extra: dev
+Requires-Dist: hypothesis>=6; extra == 'dev'
 Provides-Extra: schemas
 Requires-Dist: jsonschema>=4; extra == 'schemas'
 Provides-Extra: yaml
@@ -42,7 +44,7 @@ Description-Content-Type: text/markdown
 [![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-0.7.0-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
@@ -57,7 +59,11 @@ 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)
 
@@ -161,6 +167,7 @@ directory, and need no agent runtime to run.
 {
   "ok": true,
   "validation_mode": "structural-fallback",
+  "requested_mode": "auto",
   "schemas_checked": [
     "loop-engineer/manifest@1",
     "loop-engineer/state@1",
@@ -176,6 +183,15 @@ default, or real JSON-Schema validation against `schemas/*.json` when the
 optional `jsonschema` dependency is present (`pip install -e ".[schemas]"`), in
 which case it reads `"jsonschema"`.
 
+For `doctor`, `validate`, and `verify`, `--mode basic|strict|release` makes the
+choice explicit. With no flag, the current auto-detect behavior is unchanged
+for backward compatibility with existing foreign and scoreboard callers.
+`--mode basic` always uses the structural fallback, regardless of whether
+`jsonschema` happens to be installed. `--mode strict` and `--mode release`
+require real JSON-Schema validation and fail non-zero rather than silently
+downgrade if the `jsonschema` extra is missing. `release` currently has the
+same behavior as `strict`; it is reserved for a stricter future policy.
+
 `inspect` is a static contract linter: it scores the loop contract's structure —
 what proof machinery is present and what is missing — without running the loop:
 
@@ -253,7 +269,7 @@ terminal outcome:
   .loop/
     manifest.yaml        # contract metadata
     state.json           # live FSM cursor
-    terminal_state.json  # final exit record; written once — emit refuses overwrite without force=True
+    terminal_state.json  # final exit record; immutable once written
     artifacts/           # evidence bundles and intermediate outputs
     approvals/           # approval requests and resolutions
     checkpoints/         # recoverable snapshots
@@ -307,6 +323,12 @@ 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
 
@@ -338,13 +360,14 @@ otherwise, the stop is blocked with the exact doctor issues. No-op without
 **Any Python runtime** — `loop.emit` is a pure-stdlib writer for foreign
 orchestrators (LangGraph, or anything that can call four functions):
 `open_contract`, `append_iteration`, `append_receipt`, `terminate`. The writer
-refuses an evidence-free `Succeeded` at write time. Recipe:
+refuses `Succeeded` unless every declared criterion is true and evidence is
+present. Recipe:
 [docs/integrations/langgraph.md](docs/integrations/langgraph.md).
 
 **CI** — one workflow step validates the contract and publishes a scorecard:
 
 ```yaml
-- uses: SollanSystems/loop-engineer@v0.7.0
+- uses: SollanSystems/loop-engineer@v0.9.0
   with:
     path: "."
 ```
@@ -426,20 +449,23 @@ claude plugin validate --strict .claude-plugin/plugin.json
 ```
 
 The structural self-eval checks skill presence, frontmatter, cross-links,
-terminal-state coverage, repair-record fields, eval metrics, templates, secret
-patterns, dispatch examples, the bring-your-own-verifier default, the MIT
-license, and README differentiation.
+templates, secret patterns, dispatch examples, the bring-your-own-verifier
+default, the MIT license, and README differentiation. Three further checks are
+documentation-completeness scans: they confirm the canonical vocabulary for
+terminal-state coverage, repair-record fields, and eval metrics is present in the
+skill prose — not that a running loop enforces it (that runtime gate is `loop
+doctor` and the contract's own `verify-*` scripts).
 
 ---
 
 ## Status
 
-- Version: `0.7.0`
-- Release tag: `v0.7.0` (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
-- Current reference example: `examples/coverage-repair`
+- Current reference examples: `examples/coverage-repair`, `examples/flaky-test-triage`
 
 ---
 

{loop_engineer-0.7.0 → loop_engineer-0.9.0}/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-0.7.0-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)
 
@@ -124,6 +128,7 @@ directory, and need no agent runtime to run.
 {
   "ok": true,
   "validation_mode": "structural-fallback",
+  "requested_mode": "auto",
   "schemas_checked": [
     "loop-engineer/manifest@1",
     "loop-engineer/state@1",
@@ -139,6 +144,15 @@ default, or real JSON-Schema validation against `schemas/*.json` when the
 optional `jsonschema` dependency is present (`pip install -e ".[schemas]"`), in
 which case it reads `"jsonschema"`.
 
+For `doctor`, `validate`, and `verify`, `--mode basic|strict|release` makes the
+choice explicit. With no flag, the current auto-detect behavior is unchanged
+for backward compatibility with existing foreign and scoreboard callers.
+`--mode basic` always uses the structural fallback, regardless of whether
+`jsonschema` happens to be installed. `--mode strict` and `--mode release`
+require real JSON-Schema validation and fail non-zero rather than silently
+downgrade if the `jsonschema` extra is missing. `release` currently has the
+same behavior as `strict`; it is reserved for a stricter future policy.
+
 `inspect` is a static contract linter: it scores the loop contract's structure —
 what proof machinery is present and what is missing — without running the loop:
 
@@ -216,7 +230,7 @@ terminal outcome:
   .loop/
     manifest.yaml        # contract metadata
     state.json           # live FSM cursor
-    terminal_state.json  # final exit record; written once — emit refuses overwrite without force=True
+    terminal_state.json  # final exit record; immutable once written
     artifacts/           # evidence bundles and intermediate outputs
     approvals/           # approval requests and resolutions
     checkpoints/         # recoverable snapshots
@@ -270,6 +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
 
@@ -301,13 +321,14 @@ otherwise, the stop is blocked with the exact doctor issues. No-op without
 **Any Python runtime** — `loop.emit` is a pure-stdlib writer for foreign
 orchestrators (LangGraph, or anything that can call four functions):
 `open_contract`, `append_iteration`, `append_receipt`, `terminate`. The writer
-refuses an evidence-free `Succeeded` at write time. Recipe:
+refuses `Succeeded` unless every declared criterion is true and evidence is
+present. Recipe:
 [docs/integrations/langgraph.md](docs/integrations/langgraph.md).
 
 **CI** — one workflow step validates the contract and publishes a scorecard:
 
 ```yaml
-- uses: SollanSystems/loop-engineer@v0.7.0
+- uses: SollanSystems/loop-engineer@v0.9.0
   with:
     path: "."
 ```
@@ -389,20 +410,23 @@ claude plugin validate --strict .claude-plugin/plugin.json
 ```
 
 The structural self-eval checks skill presence, frontmatter, cross-links,
-terminal-state coverage, repair-record fields, eval metrics, templates, secret
-patterns, dispatch examples, the bring-your-own-verifier default, the MIT
-license, and README differentiation.
+templates, secret patterns, dispatch examples, the bring-your-own-verifier
+default, the MIT license, and README differentiation. Three further checks are
+documentation-completeness scans: they confirm the canonical vocabulary for
+terminal-state coverage, repair-record fields, and eval metrics is present in the
+skill prose — not that a running loop enforces it (that runtime gate is `loop
+doctor` and the contract's own `verify-*` scripts).
 
 ---
 
 ## Status
 
-- Version: `0.7.0`
-- Release tag: `v0.7.0` (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
-- Current reference example: `examples/coverage-repair`
+- Current reference examples: `examples/coverage-repair`, `examples/flaky-test-triage`
 
 ---
 

{loop_engineer-0.7.0 → loop_engineer-0.9.0}/action.yml

@@ -29,7 +29,7 @@ inputs:
 runs:
   using: "composite"
   steps:
-    - uses: actions/setup-python@v5
+    - uses: actions/setup-python@v6
       with:
         python-version: ${{ inputs.python-version }}