fix(harness): package provenance and replay corrections

.github/workflows/trunk-check.yml

@@ -29,10 +29,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683  # v4.2.2
 
       - name: Trunk Check
-        uses: trunk-io/trunk-action@d90b9166660d5e5afae248a58172a3a0e99d56d5  # v1.0.4
+        uses: trunk-io/trunk-action@04ba50e7658c81db7356da96657e6e77f220bfa3  # v1.3.1
 
       - name: Trunk Upgrade (on schedule only)
         if: github.event_name == 'schedule'
-        uses: trunk-io/trunk-action@d90b9166660d5e5afae248a58172a3a0e99d56d5  # v1.0.4
+        uses: trunk-io/trunk-action@04ba50e7658c81db7356da96657e6e77f220bfa3  # v1.3.1
         with:
           trunk-args: --upgrade

crates/harness_runner/src/dual_harness.rs

@@ -119,7 +119,8 @@ async fn run_one_helios_task(task: &FixtureTask) -> Result<TaskOutcome, FixtureE
         config.timeout_secs = Some(secs);
     }
     if let Some(env_key) = &adapter.working_dir_env {
-        let dir = std::env::var(env_key).map_err(|_| FixtureError::WorkdirUnset(task.task_id.clone()))?;
+        let dir =
+            std::env::var(env_key).map_err(|_| FixtureError::WorkdirUnset(task.task_id.clone()))?;
         config.working_dir = Some(dir);
     }
 
@@ -129,11 +130,7 @@ async fn run_one_helios_task(task: &FixtureTask) -> Result<TaskOutcome, FixtureE
 
     let passed = match (&task.acceptance, result) {
         (
-            Acceptance {
-                must_error: Some(true),
-                error_class: Some(class),
-                ..
-            },
+            Acceptance { must_error: Some(true), error_class: Some(class), .. },
             Err(RunError::Timeout(_)),
         ) if class == "timeout" => true,
         (acceptance, Ok(run)) => {
@@ -172,11 +169,7 @@ async fn run_one_helios_task(task: &FixtureTask) -> Result<TaskOutcome, FixtureE
     Ok(TaskOutcome {
         task_id: task.task_id.clone(),
         passed,
-        detail: if passed {
-            "ok".into()
-        } else {
-            "acceptance failed".into()
-        },
+        detail: if passed { "ok".into() } else { "acceptance failed".into() },
     })
 }
 

harness/__init__.py

@@ -1 +1,10 @@
 # Helios Harness Benchmarks
+
+from pathlib import Path
+
+
+# Keep the legacy benchmark package importable from the repository root while
+# exposing the installable ``src/harness`` package to in-repo entrypoint users.
+_SOURCE_PACKAGE = Path(__file__).resolve().parent / "src" / "harness"
+if _SOURCE_PACKAGE.is_dir() and str(_SOURCE_PACKAGE) not in __path__:
+    __path__.append(str(_SOURCE_PACKAGE))

harness/pyproject.toml

@@ -3,7 +3,10 @@ name = "helios-harness"
 version = "0.1.0"
 description = "Reusable CLI/API/SDK quality harness for phase-2 evidence generation"
 requires-python = ">=3.12"
-dependencies = []
+dependencies = [
+    "fastjsonschema>=2.21,<3",
+    "jsonschema>=4.23,<5",
+]
 
 [build-system]
 requires = ["hatchling"]

harness/scripts/run-harness.py

@@ -88,6 +88,33 @@
     }
 
 
+def _subject_ref(discovery) -> str:
+    """Return a stable ref for a Git checkout, including detached HEADs."""
+    if discovery.manifest.branch:
+        return discovery.manifest.branch
+    if discovery.manifest.commit:
+        return f"detached:{discovery.manifest.commit}"
+    return ""
+
+
+def _write_unresolved_provenance(payload: dict, out: str, repo: str, subject_ref: str) -> None:
+    """Emit an explicit warning instead of fabricating an envelope for non-Git input."""
+    payload["result_code"] = "WARN"
+    payload["provenance"] = {
+        "status": "unresolved",
+        "reason": "non_git_repository",
+        "collector": "helios-harness",
+        "source_ref": subject_ref or None,
+        "source_sha": None,
+        "repo": repo,
+    }
+    payload["result"] = {
+        "status": "warning",
+        "failure_class": "provenance_unresolved",
+    }
+    Path(out).write_text(json.dumps(payload, indent=2))
+
+
 def run_discovery(root: str, out: str, max_scan_depth: int) -> None:
     from harness.discoverer import Discoverer
     from harness.interfaces import DiscoverInput
@@ -162,8 +189,20 @@
         "plan": commands,
         "command_count": len(commands),
         "reproducibility": _reproducibility_metadata(profile, args),
+        "fixture": {
+            "kind": "discovered-repository",
+            "repo": repo,
+            "ref": _subject_ref(discovery) or None,
+            "commit": discovery.manifest.commit,
+            "plan_sha256": command_hash,
+        },
     }
 
+    subject_ref = _subject_ref(discovery)
+    if not discovery.manifest.commit or not subject_ref:
+        _write_unresolved_provenance(result, out, repo, subject_ref)
+        return
+
     if replay_payload is not None:
         result["replay"] = {
             **replay_payload,
@@ -173,7 +212,14 @@
     if args.dry_run:
         result["result_code"] = "WARN" if not commands else "PASS"
         from harness.benchmark_envelope import add_envelope
-        result = add_envelope(result, repo=repo, profile=profile, plan_hash=command_hash)
+        result = add_envelope(
+            result,
+            repo=repo,
+            profile=profile,
+            plan_hash=command_hash,
+            subject_commit=discovery.manifest.commit or "",
+            subject_ref=subject_ref,
+        )
         Path(out).write_text(json.dumps(result, indent=2))
         return
 
@@ -195,9 +241,13 @@
     payload["reproducibility"] = _reproducibility_metadata(profile, args)
     payload["created_at"] = datetime.now(tz=UTC).isoformat()
     payload["command_count"] = len(commands)
-    from harness.benchmark_envelope import add_envelope
-    payload = add_envelope(payload, repo=repo, profile=profile, plan_hash=command_hash)
-
+    payload["fixture"] = {
+        "kind": "discovered-repository",
+        "repo": repo,
+        "ref": subject_ref or None,
+        "commit": discovery.manifest.commit,
+        "plan_sha256": command_hash,
+    }
     if args.replay:
         replay_path = Path(args.replay)
         if replay_payload is None and replay_path.exists():
@@ -242,6 +292,20 @@
                 "plan_diff": plan_diff,
             }
 
+    if not discovery.manifest.commit or not subject_ref:
+        _write_unresolved_provenance(payload, out, repo, subject_ref)
+        return
+
+    from harness.benchmark_envelope import add_envelope
+    payload = add_envelope(
+        payload,
+        repo=repo,
+        profile=profile,
+        plan_hash=command_hash,
+        subject_commit=discovery.manifest.commit or "",
+        subject_ref=subject_ref,
+    )
+
     Path(out).write_text(json.dumps(payload, indent=2))
 
 
@@ -287,130 +351,16 @@
     print("VALID")
 
 
-# =============================================================================
-# Teammates CLI
-# =============================================================================
-
-
-def cmd_teammates_list(agents_dir: str) -> None:
-    from harness import TeammateRegistry
-
-    registry = TeammateRegistry(agents_dir=Path(agents_dir))
-    teammates = registry.discover()
-
-    if not teammates:
-        print("No teammates found")
-        return
-
-    print(f"Found {len(teammates)} teammates:\n")
-    for t in teammates.values():
-        print(f"  {t.id}: {t.name} ({t.role})")
-        print(f"    {t.description[:60]}...")
-
-
-def cmd_teammates_delegate(teammate_id: str, task: str, timeout: int, profile: str) -> None:
-    import asyncio
-
-    from harness import CodexExecutor, DelegationProtocol, DelegationRequest, Priority, TeammateRegistry
-
-    async def run():
-        registry = TeammateRegistry()
-        registry.discover()
-
-        teammate = registry.get(teammate_id)
-        if not teammate:
-            print(f"Teammate not found: {teammate_id}")
-            return
-
-        protocol = DelegationProtocol()
-        executor = CodexExecutor(profile=profile)
-
-        request = DelegationRequest(
-            teammate_id=teammate_id, task_description=task, priority=Priority.NORMAL, timeout_seconds=timeout
-        )
-
-        result = await protocol.delegate(request, executor)
-
-        print(f"Delegation: {result.delegation_id}")
-        print(f"Status: {result.status}")
-        print(f"Duration: {result.duration_ms}ms")
-        if result.result:
-            print(f"Result: {result.result[:200]}...")
-        if result.error:
-            print(f"Error: {result.error}")
-
-    asyncio.run(run())
-
-
-def cmd_teammates_status(delegation_id: str) -> None:
-    from harness import DelegationProtocol
-
-    protocol = DelegationProtocol()
-    result = protocol.get_status(delegation_id)
-
-    if result:
-        print(f"Delegation: {result.delegation_id}")
-        print(f"Status: {result.status}")
-        print(f"Duration: {result.duration_ms}ms")
-    else:
-        print(f"Delegation not found: {delegation_id}")
-
-
-# =============================================================================
-# Scaling CLI
-# =============================================================================
-
-
-def cmd_scaling_status() -> None:
-    from harness import DynamicLimitController, ResourceSampler
-
-    sampler = ResourceSampler()
-    controller = DynamicLimitController()
-
-    snapshot = sampler.sample()
-
-    print("Resource Status:")
-    print(f"  CPU: {snapshot.cpu_percent:.1f}%")
-    print(f"  Memory: {snapshot.memory_percent:.1f}% ({snapshot.memory_available_mb:.0f}MB available)")
-    print(f"  FDs: {snapshot.fd_count}/{snapshot.fd_limit}")
-    print(f"  Load: {snapshot.load_avg:.2f}")
-    print(f"\nDynamic Limit: {controller.current_limit}")
-    print(f"State: {controller._state}")
-
-
-# =============================================================================
-# Cache CLI
-# =============================================================================
-
-
-def cmd_cache_stats() -> None:
-    from harness import L1Cache
-
-    cache = L1Cache()
-    stats = cache.stats
-
-    print("L1 Cache Stats:")
-    print(f"  Hits: {stats.hits}")
-    print(f"  Misses: {stats.misses}")
-    print(f"  Hit Rate: {stats.hit_rate:.1%}")
-
-
-def cmd_cache_clear() -> None:
-    from harness import L1Cache, L2Cache
-
-    l1 = L1Cache()
-    l2 = L2Cache()
-
-    # Clear L1 (recreate)
-    l1._cache.clear()
-    print("L1 cache cleared")
-
-    # Clear L2
-    l2.clear()
-    print("L2 cache cleared")
-
-
 def main() -> None:
+    from harness.commands import (
+        cmd_cache_clear,
+        cmd_cache_stats,
+        cmd_scaling_status,
+        cmd_teammates_delegate,
+        cmd_teammates_list,
+        cmd_teammates_status,
+    )
+
     p = argparse.ArgumentParser()
     sp = p.add_subparsers(dest="cmd", required=True)