execweave 0.7.2__tar.gz → 0.7.3__tar.gz

{execweave-0.7.2 → execweave-0.7.3}/.github/workflows/ci.yml

@@ -150,6 +150,12 @@ jobs:
         run: python scripts/check_correlation_graph.py claude-record-smoke/graph.correlated.json
       - name: Check correlated standalone viewer JavaScript syntax
         run: python scripts/check_viewer_js.py claude-record-smoke/viewer.correlated.html --output claude-record-smoke/viewer.correlated.js
+      - name: Verify Claude materialized conversations
+        run: >
+          python scripts/check_conversation_records.py claude-record-smoke
+          --expect-agent /root
+          --expect-root-text "CI CLAUDE PROMPT"
+          --expect-root-text "CI CLAUDE FINAL ANSWER"
       - name: Gemini run-bound semantic and correlation smoke
         run: execweave-gemini-record --backend auto --no-files --no-network --output-dir gemini-record-smoke -- python scripts/emit_gemini_hook_smoke.py
       - name: Validate Gemini runtime stream
@@ -162,6 +168,12 @@ jobs:
         run: python scripts/check_correlation_graph.py gemini-record-smoke/graph.correlated.json
       - name: Check Gemini correlated viewer JavaScript syntax
         run: python scripts/check_viewer_js.py gemini-record-smoke/viewer.correlated.html --output gemini-record-smoke/viewer.correlated.js
+      - name: Verify Gemini materialized conversations
+        run: >
+          python scripts/check_conversation_records.py gemini-record-smoke
+          --expect-agent /root
+          --expect-root-text "CI GEMINI PROMPT"
+          --expect-root-text "CI GEMINI FINAL ANSWER"
       - name: Cursor run-bound semantic and correlation smoke
         run: execweave-cursor-record --backend auto --no-files --no-network --output-dir cursor-record-smoke -- python scripts/emit_cursor_hook_smoke.py
       - name: Validate Cursor runtime stream
@@ -174,6 +186,12 @@ jobs:
         run: python scripts/check_correlation_graph.py cursor-record-smoke/graph.correlated.json
       - name: Check Cursor correlated viewer JavaScript syntax
         run: python scripts/check_viewer_js.py cursor-record-smoke/viewer.correlated.html --output cursor-record-smoke/viewer.correlated.js
+      - name: Verify Cursor materialized conversations
+        run: >
+          python scripts/check_conversation_records.py cursor-record-smoke
+          --expect-agent /root
+          --expect-root-text "CI CURSOR PROMPT"
+          --expect-root-text "CI CURSOR FINAL ANSWER"
       - name: OpenCode plugin installer smoke
         run: execweave-opencode-plugin --install --root opencode-plugin-smoke
       - name: OpenCode run-bound semantic and correlation smoke
@@ -188,6 +206,40 @@ jobs:
         run: python scripts/check_correlation_graph.py opencode-record-smoke/graph.correlated.json
       - name: Check OpenCode correlated viewer JavaScript syntax
         run: python scripts/check_viewer_js.py opencode-record-smoke/viewer.correlated.html --output opencode-record-smoke/viewer.correlated.js
+      - name: Verify OpenCode materialized conversations
+        run: >
+          python scripts/check_conversation_records.py opencode-record-smoke
+          --expect-agent /root
+          --expect-root-text "CI OPENCODE FINAL ANSWER"
+          --forbid-agent /root/ci-opencode-session
+      - name: Codex hook command smoke
+        run: python scripts/check_codex_hook.py
+      - name: Codex run-bound semantic and correlation smoke
+        run: execweave-codex-record --backend auto --no-files --no-network --output-dir codex-record-smoke -- python scripts/emit_codex_hook_smoke.py
+      - name: Validate Codex runtime stream
+        run: execweave validate codex-record-smoke/events.jsonl
+      - name: Validate Codex merged stream
+        run: execweave validate codex-record-smoke/events.semantic.jsonl
+      - name: Validate Codex correlated stream
+        run: execweave validate codex-record-smoke/events.correlated.jsonl
+      - name: Verify Codex conservative correlation edge
+        run: python scripts/check_correlation_graph.py codex-record-smoke/graph.correlated.json
+      - name: Check Codex correlated viewer JavaScript syntax
+        run: python scripts/check_viewer_js.py codex-record-smoke/viewer.correlated.html --output codex-record-smoke/viewer.correlated.js
+      # Codex reports a subagent stop with the child rollout on agent_transcript_path
+      # and the parent rollout on transcript_path. Archiving the parent is the
+      # v0.7.2 regression, so assert the child conversation survives end to end.
+      - name: Verify Codex agent-local subagent conversations
+        run: >
+          python scripts/check_conversation_records.py codex-record-smoke
+          --min-entries 2
+          --expect-agent /root
+          --expect-agent /root/explorer
+          --expect-root-text "CI ROOT PROMPT"
+          --expect-root-text "CI ROOT FINAL ANSWER"
+          --expect-owned "/root=CI ROOT COMMENTARY"
+          --expect-owned "/root/explorer=CI CHILD PRIVATE REASONING"
+          --forbid-agent /root/01a04cea-67b2-7683-9f6b-cd644497b862
       - name: Model runtime CLI smoke
         run: python scripts/check_model_runtime_cli.py
       - name: Inference gateway CLI smoke

{execweave-0.7.2 → execweave-0.7.3}/.github/workflows/publish.yml

@@ -3,21 +3,30 @@ name: Publish to PyPI
 on:
   release:
     types: [published]
+  workflow_dispatch:
+    inputs:
+      release_tag:
+        description: Release tag to publish
+        required: true
+        type: string
 
 permissions:
   contents: read
 
+env:
+  RELEASE_TAG: ${{ github.event.release.tag_name || inputs.release_tag }}
+
 jobs:
   build:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
+        with:
+          ref: ${{ env.RELEASE_TAG }}
       - uses: actions/setup-python@v5
         with:
           python-version: "3.12"
       - name: Verify release tag matches package version
-        env:
-          RELEASE_TAG: ${{ github.event.release.tag_name }}
         run: |
           python - <<'PY'
           import os
@@ -75,5 +84,4 @@ jobs:
       - name: Attach distributions to GitHub Release
         env:
           GH_TOKEN: ${{ github.token }}
-          RELEASE_TAG: ${{ github.event.release.tag_name }}
         run: gh release upload "$RELEASE_TAG" dist/* --clobber --repo "$GITHUB_REPOSITORY"

execweave-0.7.3/.github/workflows/repair-v073.yml

@@ -0,0 +1,173 @@
+name: Repair v0.7.3 Release
+
+on:
+  workflow_run:
+    workflows: ["CI"]
+    types: [completed]
+    branches: [main]
+
+permissions:
+  actions: write
+  contents: write
+
+jobs:
+  repair:
+    if: ${{ github.event.workflow_run.conclusion == 'success' }}
+    runs-on: ubuntu-latest
+    env:
+      GH_TOKEN: ${{ github.token }}
+      RELEASE_TAG: v0.7.3
+      EXPECTED_VERSION: 0.7.3
+      STALE_TAG_SHA: 25321ec325ecf871ec2a56b55b593fcac854ffcf
+      HEAD_SHA: ${{ github.event.workflow_run.head_sha }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ env.HEAD_SHA }}
+
+      - name: Verify repaired release candidate
+        run: |
+          set -euo pipefail
+          current_main="$(gh api "repos/$GITHUB_REPOSITORY/git/ref/heads/main" --jq '.object.sha')"
+          if [[ "$current_main" != "$HEAD_SHA" ]]; then
+            echo "main moved from $HEAD_SHA to $current_main; refusing release repair" >&2
+            exit 1
+          fi
+          python - <<'PY'
+          import os
+          import tomllib
+          from pathlib import Path
+
+          version = tomllib.loads(Path("pyproject.toml").read_text(encoding="utf-8"))["project"]["version"]
+          expected = os.environ["EXPECTED_VERSION"]
+          if version != expected:
+              raise SystemExit(f"package version {version!r} does not match expected {expected!r}")
+          print(f"Release candidate version: {version}")
+          PY
+
+      - name: Wait for all release gates
+        run: |
+          set -euo pipefail
+          required=("CI" "Package Smoke" "Wheel Install Smoke" "Windows Launcher Compatibility")
+          for attempt in $(seq 1 60); do
+            runs="$(gh api "repos/$GITHUB_REPOSITORY/actions/runs?head_sha=$HEAD_SHA&per_page=100")"
+            pending=0
+            for gate in "${required[@]}"; do
+              state="$(jq -r --arg name "$gate" '
+                .workflow_runs
+                | map(select(.name == $name and .event == "push"))
+                | sort_by(.created_at)
+                | reverse
+                | .[0]
+                | if . == null then "missing"
+                  else ((.status // "") + ":" + (.conclusion // "")) end
+              ' <<<"$runs")"
+              echo "$gate => $state"
+              case "$state" in
+                completed:success) ;;
+                completed:*) echo "$gate failed: $state" >&2; exit 1 ;;
+                *) pending=1 ;;
+              esac
+            done
+            if [[ "$pending" -eq 0 ]]; then
+              exit 0
+            fi
+            sleep 5
+          done
+          echo "release gates did not all complete successfully" >&2
+          exit 1
+
+      - name: Replace stale v0.7.3 tag and release
+        run: |
+          set -euo pipefail
+          tag_sha="$(gh api "repos/$GITHUB_REPOSITORY/git/ref/tags/$RELEASE_TAG" --jq '.object.sha' 2>/dev/null || true)"
+          release_exists=0
+          asset_count=0
+          if gh release view "$RELEASE_TAG" --repo "$GITHUB_REPOSITORY" >/dev/null 2>&1; then
+            release_exists=1
+            asset_count="$(gh release view "$RELEASE_TAG" --repo "$GITHUB_REPOSITORY" --json assets --jq '.assets | length')"
+          fi
+
+          if [[ -n "$tag_sha" && "$tag_sha" != "$STALE_TAG_SHA" && "$tag_sha" != "$HEAD_SHA" ]]; then
+            echo "v0.7.3 tag unexpectedly points to $tag_sha; refusing to overwrite" >&2
+            exit 1
+          fi
+          if [[ "$asset_count" -gt 0 && "$tag_sha" != "$HEAD_SHA" ]]; then
+            echo "existing v0.7.3 release already has assets; refusing destructive replacement" >&2
+            exit 1
+          fi
+
+          if [[ "$tag_sha" != "$HEAD_SHA" ]]; then
+            if [[ "$release_exists" -eq 1 ]]; then
+              gh release delete "$RELEASE_TAG" --cleanup-tag --yes --repo "$GITHUB_REPOSITORY"
+            elif [[ -n "$tag_sha" ]]; then
+              gh api --method DELETE "repos/$GITHUB_REPOSITORY/git/refs/tags/$RELEASE_TAG"
+            fi
+            gh release create "$RELEASE_TAG" \
+              --target "$HEAD_SHA" \
+              --title "ExecWeave v0.7.3" \
+              --generate-notes \
+              --repo "$GITHUB_REPOSITORY"
+          else
+            echo "v0.7.3 tag already points to repaired release candidate"
+          fi
+
+      - name: Dispatch canonical publish workflow
+        run: |
+          set -euo pipefail
+          gh workflow run publish.yml \
+            --ref main \
+            -f release_tag="$RELEASE_TAG" \
+            --repo "$GITHUB_REPOSITORY"
+
+          run_id=""
+          for attempt in $(seq 1 30); do
+            run_id="$(gh run list \
+              --workflow publish.yml \
+              --event workflow_dispatch \
+              --branch main \
+              --limit 20 \
+              --json databaseId,headSha,createdAt \
+              --jq ".[] | select(.headSha == \"$HEAD_SHA\") | .databaseId" \
+              --repo "$GITHUB_REPOSITORY" | head -n 1)"
+            if [[ -n "$run_id" ]]; then
+              break
+            fi
+            sleep 2
+          done
+          if [[ -z "$run_id" ]]; then
+            echo "could not locate dispatched publish workflow" >&2
+            exit 1
+          fi
+          echo "publish_run_id=$run_id" >> "$GITHUB_ENV"
+
+      - name: Wait for publish workflow
+        run: gh run watch "$publish_run_id" --exit-status --repo "$GITHUB_REPOSITORY"
+
+      - name: Verify release assets and PyPI
+        run: |
+          set -euo pipefail
+          assets="$(gh release view "$RELEASE_TAG" --repo "$GITHUB_REPOSITORY" --json assets --jq '.assets[].name')"
+          grep -Fx "execweave-0.7.3-py3-none-any.whl" <<<"$assets"
+          grep -Fx "execweave-0.7.3.tar.gz" <<<"$assets"
+
+          python - <<'PY'
+          import json
+          import time
+          import urllib.error
+          import urllib.request
+
+          url = "https://pypi.org/pypi/execweave/0.7.3/json"
+          for _ in range(60):
+              try:
+                  with urllib.request.urlopen(url, timeout=10) as response:
+                      payload = json.load(response)
+                  if payload.get("info", {}).get("version") == "0.7.3":
+                      print("PyPI ExecWeave 0.7.3 verified")
+                      break
+              except (urllib.error.URLError, json.JSONDecodeError):
+                  pass
+              time.sleep(5)
+          else:
+              raise SystemExit("PyPI did not expose ExecWeave 0.7.3")
+          PY

{execweave-0.7.2 → execweave-0.7.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: execweave
-Version: 0.7.2
+Version: 0.7.3
 Summary: Turn AI-agent runtime behavior into a local interactive execution graph.
 Project-URL: Homepage, https://github.com/Irish-kw/ExecWeave
 Project-URL: Repository, https://github.com/Irish-kw/ExecWeave

{execweave-0.7.2 → execweave-0.7.3}/docs/codex-hooks.de.md

@@ -30,7 +30,6 @@ ExecWeave zeichnet Evidenz aus Codex-Lifecycle-Hooks neben unabhängig erfasster
 - `SubagentStart`
 - `SubagentStop`
 - `Stop`
-- `Interrupt`
 
 Unbekannte oder upstream nicht verfügbare Ereignisse werden nicht erfunden. Hook-Schemata und Dispatch-Abdeckung können sich zwischen Codex-Versionen ändern.
 

{execweave-0.7.2 → execweave-0.7.3}/docs/codex-hooks.fr.md

@@ -30,7 +30,6 @@ ExecWeave enregistre les preuves issues des hooks de cycle de vie Codex à côt
 - `SubagentStart`
 - `SubagentStop`
 - `Stop`
-- `Interrupt`
 
 ExecWeave n'invente pas les événements absents ou indisponibles côté fournisseur. Les schémas et la couverture peuvent changer selon les versions de Codex.
 

{execweave-0.7.2 → execweave-0.7.3}/docs/codex-hooks.ja.md

@@ -30,7 +30,6 @@ ExecWeave は Codex lifecycle-hook evidence を独立した OS runtime telemetry
 - `SubagentStart`
 - `SubagentStop`
 - `Stop`
-- `Interrupt`
 
 未知または利用できない upstream event は作りません。Hook schema と dispatch coverage は Codex version により変わる可能性があります。
 

{execweave-0.7.2 → execweave-0.7.3}/docs/codex-hooks.ko.md

@@ -30,7 +30,6 @@ ExecWeave는 Codex 라이프사이클 훅 증거를 독립적인 OS 런타임
 - `SubagentStart`
 - `SubagentStop`
 - `Stop`
-- `Interrupt`
 
 업스트림에 없거나 사용할 수 없는 이벤트를 ExecWeave가 만들어내지는 않습니다. Codex 버전에 따라 훅 스키마와 디스패치 범위가 달라질 수 있습니다.
 

{execweave-0.7.2 → execweave-0.7.3}/docs/codex-hooks.md

@@ -30,7 +30,6 @@ ExecWeave records Codex lifecycle-hook evidence beside independent OS runtime te
 - `SubagentStart`
 - `SubagentStop`
 - `Stop`
-- `Interrupt`
 
 Unknown or unavailable upstream events are not invented. Hook schemas and dispatch coverage can change between Codex versions.
 

{execweave-0.7.2 → execweave-0.7.3}/docs/codex-hooks.ru.md

@@ -30,7 +30,6 @@ ExecWeave записывает evidence из lifecycle hooks Codex рядом с
 - `SubagentStart`
 - `SubagentStop`
 - `Stop`
-- `Interrupt`
 
 Неизвестные или недоступные upstream events не выдумываются. Hook schemas и dispatch coverage могут меняться между версиями Codex.
 

{execweave-0.7.2 → execweave-0.7.3}/docs/codex-hooks.zh-CN.md

@@ -30,7 +30,6 @@ ExecWeave 将 Codex lifecycle-hook evidence 与独立 OS runtime telemetry 并
 - `SubagentStart`
 - `SubagentStop`
 - `Stop`
-- `Interrupt`
 
 ExecWeave 不会伪造 upstream 未知或不可用的 event。Hook schema 与 dispatch coverage 可能随 Codex 版本改变。
 

{execweave-0.7.2 → execweave-0.7.3}/docs/codex-hooks.zh-TW.md

@@ -30,7 +30,6 @@ ExecWeave 會把 Codex lifecycle-hook evidence 與獨立 OS runtime telemetry
 - `SubagentStart`
 - `SubagentStop`
 - `Stop`
-- `Interrupt`
 
 ExecWeave 不會捏造 upstream 未知或不可用的 event。Hook schema 與 dispatch coverage 可能隨 Codex 版本改變。
 

{execweave-0.7.2 → execweave-0.7.3}/pyproject.toml

@@ -4,8 +4,8 @@ build-backend = "hatchling.build"
 
 [project]
 name = "execweave"
-# ExecWeave v0.7.2 release metadata
-version = "0.7.2"
+# ExecWeave v0.7.3 release metadata
+version = "0.7.3"
 description = "Turn AI-agent runtime behavior into a local interactive execution graph."
 readme = "README.md"
 requires-python = ">=3.10"

{execweave-0.7.2 → execweave-0.7.3}/scripts/check_codex_hook.py

@@ -65,22 +65,16 @@ def main() -> int:
         check=True,
     )
     parsed = json.loads(config.stdout)
-    required = {
-        "PreToolUse",
-        "PermissionRequest",
-        "PostToolUse",
-        "PreCompact",
-        "PostCompact",
-        "SessionStart",
-        "SessionEnd",
-        "UserPromptSubmit",
-        "SubagentStart",
-        "SubagentStop",
-        "Stop",
-        "Interrupt",
-    }
+    # Bound to the shipped event set rather than a second hardcoded copy, which had
+    # drifted to expect an "Interrupt" hook ExecWeave does not register.
+    from execweave.codex_hook_lifecycle import OFFICIAL_CODEX_HOOK_EVENTS
+
+    required = set(OFFICIAL_CODEX_HOOK_EVENTS)
     if set(parsed.get("hooks", {})) != required:
-        raise SystemExit("generated Codex hook config is incomplete")
+        raise SystemExit(
+            "generated Codex hook config does not match the official event set: "
+            f"{sorted(set(parsed.get('hooks', {})) ^ required)}"
+        )
     for event_name in {"PreToolUse", "PermissionRequest", "PostToolUse"}:
         if parsed["hooks"][event_name][0].get("matcher") != "*":
             raise SystemExit(f"Codex {event_name} hook is not configured for all tools")

execweave-0.7.3/scripts/check_conversation_records.py

@@ -0,0 +1,215 @@
+"""Assert that a recorded run actually materialized the conversations it observed.
+
+CI previously ran every provider's record pipeline end to end and never looked at
+``conversations.json``. All four provider smokes were producing ``entry_count: 0``
+while reporting success, which is how a total conversation collapse reached a tagged
+release with green CI.
+
+This checker closes that gap. It reads the final artifact — not an intermediate
+preview object — and fails when a run that observed conversational evidence published
+none, when an expected agent is missing, when one agent execution is published as
+several conversations, or when a message landed in the wrong agent's thread.
+
+Every check walks the real entry list. An earlier version keyed previews by agent path
+before running the ownership checks, so a duplicate silently overwrote its twin and the
+isolation assertions only ever saw one of them.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from collections import defaultdict
+from pathlib import Path
+from typing import Any
+
+
+def _fail(message: str) -> None:
+    print(f"conversation record check FAILED: {message}", file=sys.stderr)
+    raise SystemExit(1)
+
+
+def _load(run_dir: Path) -> dict[str, Any]:
+    path = run_dir / "conversations.json"
+    if not path.is_file():
+        _fail(f"{path} does not exist")
+    try:
+        payload = json.loads(path.read_text(encoding="utf-8"))
+    except json.JSONDecodeError as exc:
+        _fail(f"{path} is not valid JSON: {exc.msg}")
+    if not isinstance(payload, dict):
+        _fail(f"{path} root must be a JSON object")
+    return payload
+
+
+def _entries_by_agent(document: dict[str, Any]) -> dict[str, list[dict[str, Any]]]:
+    """Group every materialized preview under its agent path, keeping duplicates."""
+    grouped: dict[str, list[dict[str, Any]]] = defaultdict(list)
+    for entry in document.get("entries") or []:
+        if not isinstance(entry, dict):
+            continue
+        preview = entry.get("conversation_preview")
+        if isinstance(preview, dict) and preview.get("agent_path"):
+            grouped[str(preview["agent_path"])].append(preview)
+    return dict(grouped)
+
+
+def _texts(preview: dict[str, Any]) -> str:
+    return "\n".join(
+        str(message.get("text") or "") for message in preview.get("messages") or []
+    )
+
+
+def _describe(agent_path: str, previews: list[dict[str, Any]]) -> str:
+    threads = [str(preview.get("thread_id")) for preview in previews]
+    return f"{agent_path} -> {len(previews)} entries with thread ids {threads}"
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(
+        description="Verify a run's materialized conversations.json"
+    )
+    parser.add_argument("run_dir", type=Path, help="ExecWeave run artifact directory")
+    parser.add_argument(
+        "--expect-agent",
+        action="append",
+        default=[],
+        metavar="AGENT_PATH",
+        help="agent path that must have a conversation entry (repeatable)",
+    )
+    parser.add_argument(
+        "--expect-root-text",
+        action="append",
+        default=[],
+        metavar="TEXT",
+        help="text that must appear in the /root thread (repeatable)",
+    )
+    parser.add_argument(
+        "--expect-owned",
+        action="append",
+        default=[],
+        metavar="AGENT_PATH=TEXT",
+        help="text that must appear in one agent's thread and nowhere else (repeatable)",
+    )
+    parser.add_argument(
+        "--forbid-agent",
+        action="append",
+        default=[],
+        metavar="AGENT_PATH",
+        help="agent path that must NOT exist, e.g. a fabricated child (repeatable)",
+    )
+    parser.add_argument(
+        "--allow-duplicate-agent",
+        action="append",
+        default=[],
+        metavar="AGENT_PATH",
+        help=(
+            "permit this agent path to hold several conversation entries. Only for a "
+            "fixture where one agent genuinely runs several independent conversations; "
+            "one execution published twice is a defect, not a configuration choice."
+        ),
+    )
+    parser.add_argument(
+        "--min-entries",
+        type=int,
+        default=1,
+        help="minimum number of materialized conversation entries (default: 1)",
+    )
+    args = parser.parse_args()
+
+    document = _load(args.run_dir)
+    grouped = _entries_by_agent(document)
+    entry_count = document.get("entry_count")
+
+    if not isinstance(entry_count, int) or entry_count < args.min_entries:
+        _fail(
+            f"entry_count is {entry_count!r}; expected at least {args.min_entries}. "
+            "The run observed conversational evidence but published none."
+        )
+    if len(grouped) < args.min_entries:
+        _fail(f"only {len(grouped)} agents carry a conversation; expected {args.min_entries}")
+
+    # One agent execution must be published exactly once. Evidence that names its
+    # thread differently is provenance, not a second conversation.
+    allowed_duplicates = set(args.allow_duplicate_agent)
+    for agent_path, previews in sorted(grouped.items()):
+        if len(previews) > 1 and agent_path not in allowed_duplicates:
+            _fail(
+                "one agent execution is published as several conversations: "
+                + _describe(agent_path, previews)
+                + ". Evidence describing the same execution must merge into one entry."
+            )
+
+    for agent_path in args.expect_agent:
+        if agent_path not in grouped:
+            _fail(f"no conversation entry for {agent_path}; found {sorted(grouped)}")
+        if not any(preview.get("messages") for preview in grouped[agent_path]):
+            _fail(f"{agent_path} materialized an entry with no messages")
+
+    for agent_path in args.forbid_agent:
+        if agent_path in grouped:
+            _fail(f"{agent_path} was materialized but no evidence establishes it")
+
+    if args.expect_root_text and "/root" not in grouped:
+        _fail(f"no /root conversation; found {sorted(grouped)}")
+    for text in args.expect_root_text:
+        if not any(text in _texts(preview) for preview in grouped.get("/root", [])):
+            _fail(f"/root thread is missing expected text {text!r}")
+
+    for spec in args.expect_owned:
+        agent_path, separator, text = spec.partition("=")
+        if not separator:
+            _fail(f"--expect-owned needs AGENT_PATH=TEXT, got {spec!r}")
+        if agent_path not in grouped:
+            _fail(f"no conversation entry for {agent_path}; found {sorted(grouped)}")
+        if not any(text in _texts(preview) for preview in grouped[agent_path]):
+            _fail(f"{agent_path} is missing its own content {text!r}")
+        # Isolation is checked against every published entry, so a duplicate cannot
+        # hide a leak by being overwritten before this runs.
+        for other_path, others in grouped.items():
+            if other_path == agent_path:
+                continue
+            for preview in others:
+                if text in _texts(preview):
+                    _fail(
+                        f"content {text!r} owned by {agent_path} leaked into "
+                        f"{other_path} (thread {preview.get('thread_id')!r})"
+                    )
+
+    # Topology provenance: every parent link must name the evidence behind it, and no
+    # entry may claim a parent that has no conversation of its own.
+    for agent_path, previews in sorted(grouped.items()):
+        for preview in previews:
+            parent = preview.get("parent_agent_path")
+            if preview.get("is_root"):
+                if parent is not None:
+                    _fail(f"{agent_path} is root but names parent {parent!r}")
+                continue
+            if parent is None:
+                _fail(f"{agent_path} is not root but names no parent")
+            if not preview.get("parent_relation_source"):
+                _fail(f"{agent_path} claims parent {parent!r} with no evidence recorded")
+            if parent not in grouped:
+                _fail(f"{agent_path} names parent {parent!r}, which has no conversation")
+
+    print(
+        f"conversation record check passed for {args.run_dir}: "
+        f"{entry_count} entries, agents {sorted(grouped)}"
+    )
+    for agent_path, previews in sorted(grouped.items()):
+        for preview in previews:
+            evidence = preview.get("evidence_thread_ids") or [preview.get("thread_id")]
+            print(
+                f"  {agent_path}: {len(preview.get('messages') or [])} messages, "
+                f"completeness={preview.get('conversation_completeness')}, "
+                f"path_source={preview.get('agent_path_source')}, "
+                f"topology={preview.get('topology_state')}, "
+                f"thread={preview.get('thread_id')}, "
+                f"evidence_threads={len(evidence)}"
+            )
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

{execweave-0.7.2 → execweave-0.7.3}/scripts/emit_claude_hook_smoke.py

@@ -30,12 +30,22 @@ def main() -> int:
     }
     if not os.environ.get("EXECWEAVE_SEMANTIC_SIDECAR"):
         raise SystemExit("run-bound semantic sidecar environment was not inherited")
-    subprocess.run(
-        ["execweave-claude-hook"],
-        input=json.dumps(payload),
-        text=True,
-        check=True,
-    )
+
+    def send(body: dict) -> None:
+        subprocess.run(
+            ["execweave-claude-hook"], input=json.dumps(body), text=True, check=True
+        )
+
+    base = {key: payload[key] for key in ("session_id", "cwd", "permission_mode")}
+    # Conversation evidence, so the run materializes a conversations.json CI can check.
+    send({**base, "hook_event_name": "UserPromptSubmit", "prompt": "CI CLAUDE PROMPT"})
+    send(payload)
+    send({
+        **base,
+        "hook_event_name": "Stop",
+        "stop_hook_active": False,
+        "last_assistant_message": "CI CLAUDE FINAL ANSWER",
+    })
     child = subprocess.Popen(child_argv)
     # Keep the emitter alive while the child is running so the portable sampler
     # gets many observation opportunities on slower hosted runners.