execweave 0.6.3__tar.gz → 0.6.4__tar.gz

{execweave-0.6.3 → execweave-0.6.4}/.github/workflows/i18n.yml

@@ -1,6 +1,7 @@
 name: Documentation i18n
 
 on:
+  workflow_dispatch:
   push:
     branches: [main]
     paths:
@@ -30,5 +31,5 @@ jobs:
           python-version: "3.12"
       - name: Check eight-language navigation and coverage
         run: python scripts/sync_i18n_nav.py --check
-      - name: Check German, French, and Russian structural parity
+      - name: Check translation structural parity
         run: python scripts/audit_i18n_parity.py --languages fr de ru

{execweave-0.6.3 → execweave-0.6.4}/.github/workflows/publish.yml

@@ -38,6 +38,8 @@ jobs:
           python -m pip install build twine
           python -m build
           python -m twine check dist/*
+      - name: Clean wheel install smoke
+        run: python scripts/check_wheel_install.py
       - uses: actions/upload-artifact@v4
         with:
           name: python-package-distributions

{execweave-0.6.3 → execweave-0.6.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: execweave
-Version: 0.6.3
+Version: 0.6.4
 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
@@ -76,9 +76,30 @@ ExecWeave is an open-source, local-first observability project that turns AI-age
 > **Event is ground truth. The graph is a materialized view.**
 
 <p align="center">
-  <img src="docs/assets/execweave-viewer.png" alt="ExecWeave Live execution graph" width="100%">
+  <img src="docs/assets/execweave-launch-demo-v5-x.gif" alt="ExecWeave animated live demo" width="100%">
 </p>
 
+<!-- execweave-demo:start -->
+## Reproduce this demo
+
+The animation above is a real ExecWeave v0.6.3 live session. This workload deliberately creates enough activity to make the execution graph useful: multiple Python modules, JSON/CSV files, tests, file inspection, and outbound HTTP requests.
+
+Run a local Agent CLI under ExecWeave. For example:
+
+```bash
+execweave live --open -- claude
+```
+
+Then paste this workload prompt into the Agent:
+
+```text
+Create a small Python project in ./execweave-demo with 8 modules,
+generate sample JSON and CSV data, run the program and tests,
+inspect the generated files, and fetch example.com plus the GitHub API.
+```
+The same workload works with `codex`, `gemini`, `cursor`, or `opencode`. Exact node, edge, event, process, and endpoint counts vary by OS, Agent version, and environment. ExecWeave records observed runtime evidence; the animation is one concrete run, not a fixed expected graph.
+<!-- execweave-demo:end -->
+
 ## Install
 
 ExecWeave is published on PyPI as a standard Python wheel/sdist. Install the latest published release with:
@@ -118,39 +139,43 @@ execweave live --open -- python my_agent.py
 execweave live --open -- ollama serve
 ```
 
-`execweave live` streams process, file, and network evidence for the command tree it launches. It does **not** automatically inject Agent semantic hooks, model-runtime API metadata, or inference-gateway routing metadata into the Live Viewer.
+`execweave live` streams process, file, and network evidence for the command tree it launches. In v0.6.4, configured Claude/Codex/Gemini/Cursor hooks and the OpenCode plugin automatically feed the per-run live sidecar. Ollama, llama.cpp, and vLLM server launches also receive automatic local model-catalog probes.
 
 #### Live capability matrix
 
 | Integration | Direct OS-runtime live | Specialized metadata | Automatically in Live Viewer |
 | --- | --- | --- | --- |
-| Claude Code | Yes | `execweave-claude-record` / hooks | No |
-| OpenAI Codex | Yes | `execweave-codex-record` / hooks | No |
-| Gemini CLI | Yes | `execweave-gemini-record` / hooks | No |
-| Cursor | Yes | `execweave-cursor-record` / hooks | No |
-| OpenCode | Yes | `execweave-opencode-record` / plugin | No |
-| Ollama | Yes, when launched under ExecWeave (for example `ollama serve`) | `execweave-model-runtime event/probe --runtime ollama` | No |
-| llama.cpp | Yes, when its local server is launched under ExecWeave | `execweave-model-runtime event/probe --runtime llamacpp` | No |
-| vLLM | Yes, when its local server is launched under ExecWeave | `execweave-model-runtime event/probe --runtime vllm` | No |
-| LM Studio | Yes only for a local process launched under ExecWeave; an already-running server is not attached | `execweave-model-runtime event/probe --runtime lmstudio` | No |
-| LiteLLM Proxy | Yes, when the local proxy is launched under ExecWeave | `execweave-inference-gateway event --gateway litellm` | No |
+| Claude Code | Yes | `execweave-claude-record` / hooks | Yes (configured hook/plugin) |
+| OpenAI Codex | Yes | `execweave-codex-record` / hooks | Yes (configured hook/plugin) |
+| Gemini CLI | Yes | `execweave-gemini-record` / hooks | Yes (configured hook/plugin) |
+| Cursor | Yes | `execweave-cursor-record` / hooks | Yes (configured hook/plugin) |
+| OpenCode | Yes | `execweave-opencode-record` / plugin | Yes (configured hook/plugin) |
+| Ollama | Yes, when launched under ExecWeave (for example `ollama serve`) | `execweave-model-runtime event/probe --runtime ollama` | Yes (automatic local probe) |
+| llama.cpp | Yes, when its local server is launched under ExecWeave | `execweave-model-runtime event/probe --runtime llamacpp` | Yes (automatic local probe) |
+| vLLM | Yes, when its local server is launched under ExecWeave | `execweave-model-runtime event/probe --runtime vllm` | Yes (automatic local probe) |
+| LM Studio | Yes only for a local process launched under ExecWeave; an already-running server is not attached | `execweave-model-runtime event/probe --runtime lmstudio` | Yes (automatic after successful `lms server start --port <port>`) |
+| LiteLLM Proxy | Yes, when the local proxy is launched under ExecWeave | `execweave-inference-gateway event --gateway litellm` / configured callback | Yes (configured callback) |
 | OpenRouter | No direct remote-service process; run the local client/Agent under `live` instead | `execweave-inference-gateway event/generation --gateway openrouter` | No |
 
+Agent rows marked **Yes** require the provider hook/plugin to be configured once; `execweave live` then supplies the per-run `EXECWEAVE_SEMANTIC_SIDECAR` automatically. Ollama, llama.cpp, and vLLM rows marked **Yes** use automatic loopback model-catalog probes only when ExecWeave launches the corresponding local server. LM Studio is also automatic for `lms server start` when an explicit `--port` is supplied, the compatible endpoint was absent before launch, and the launcher exits successfully. LiteLLM Proxy is **Yes** after its ExecWeave callback has been configured once and the proxy is launched inside the current `execweave live` environment. OpenRouter remains **No** because remote routing metadata is not available from OS/network observation alone.
+
 For an already-running Ollama server, use `execweave-model-runtime probe --runtime ollama` to snapshot loaded-model state. For OpenRouter, `live` can observe the local client and its network activity, while gateway routing/usage metadata remains a separate evidence layer.
 
-<!-- v0.6.3-live -->
-### v0.6.3 live observability
+<!-- v0.6.4-live -->
+### v0.6.4 live observability
 
-Use the same live session in either the browser or terminal:
+`top` keeps the Agent fully interactive in the terminal where you launched it and opens the dashboard in a **separate terminal window**:
 
 ```bash
-execweave top -- codex          # Terminal dashboard
-execweave top --open -- codex   # Terminal + Web Viewer
+execweave top -- codex          # Agent here + detached Top dashboard
+execweave top --open -- codex   # Agent here + detached Top dashboard + Web Viewer
 ```
 
+The detached dashboard is an attach-only client for the same localhost live session; it never launches a second Agent process. On Windows it uses a new console, on macOS it opens Terminal, and on desktop Linux it uses an available terminal emulator. In a headless environment, ExecWeave prints an attach command instead of taking over the Agent terminal.
+
 Live updates use incremental snapshots/deltas with bounded history instead of repeatedly rebuilding and transferring the full graph. Live and standalone viewers support a persistent Dark/Light theme switch. On Linux, very large recursive filesystem scopes are preflighted and automatically fall back from inotify to polling when needed, so an exhausted inotify watch pool does not abort the session.
 
-`live` is a generic OS-runtime view rather than an integration whitelist. Specialized Agent semantic, model-runtime, and gateway metadata remain separate evidence layers and are not automatically injected into the Live Viewer in v0.6.3.
+v0.6.4 gives each live run a shared specialized-evidence sidecar. Configured Claude Code, Codex, Gemini CLI, Cursor, and OpenCode integrations inherit that sidecar automatically, so their provider semantic evidence can appear incrementally beside OS-runtime evidence in the same live graph. This live normalization is provisional; after the command exits, the final graph is rebuilt from the canonical runtime + semantic merge. Missing specialized evidence is never invented.
 
 Run the reproducible graph scalability benchmark with `execweave-scalability`; CI covers 10k, 100k, and 1M synthetic events.
 
@@ -403,9 +428,9 @@ Security findings remain explicit about evidence limits. A possible sensitive-fi
 
 ## Current status
 
-ExecWeave `main` is currently **v0.6.3** and under active development.
+ExecWeave `main` is currently **v0.6.4** and under active development.
 
-The baseline includes runtime collection, graph materialization/querying, standalone/live Viewer, Claude/Codex/Gemini/Cursor/OpenCode semantic integrations, conservative Tool → Process correlation, OpenRouter/LiteLLM gateway metadata, Ollama/llama.cpp/vLLM/LM Studio runtime metadata, exact Gateway ↔ Model Runtime request identity, published PyPI wheel/sdist packaging, reproducible overhead benchmarking, cross-platform command-launcher compatibility, large-graph browser safety guards, incremental Live JSONL tail/cache, and cross-platform CI on Python 3.10/3.12.
+The baseline includes runtime collection, graph materialization/querying, standalone/live Viewer, detached terminal Top dashboard, provisional multi-layer live evidence with canonical final semantic merge, Claude/Codex/Gemini/Cursor/OpenCode semantic integrations, conservative Tool → Process correlation, OpenRouter/LiteLLM gateway metadata, Ollama/llama.cpp/vLLM/LM Studio runtime metadata, exact Gateway ↔ Model Runtime request identity, published PyPI wheel/sdist packaging, reproducible overhead benchmarking, cross-platform command-launcher compatibility, large-graph browser safety guards, incremental Live JSONL tail/cache, and cross-platform CI on Python 3.10/3.12.
 
 ## Privacy
 
@@ -435,4 +460,4 @@ Contributions are welcome, especially around native OS collectors, additional Ag
 
 ## License
 
-See [`LICENSE`](LICENSE).
+See [`LICENSE`](LICENSE).

{execweave-0.6.3 → execweave-0.6.4}/README.de.md

@@ -26,9 +26,31 @@ ExecWeave ist ein Open-Source-, local-first Observability-Projekt, das die Aktiv
 > **Das Event ist die ground truth; der Graph ist eine materialized view.**
 
 <p align="center">
-  <img src="docs/assets/execweave-viewer.png" alt="ExecWeave Live execution graph" width="100%">
+  <img src="docs/assets/execweave-launch-demo-v5-x.gif" alt="ExecWeave Live execution graph" width="100%">
 </p>
 
+<!-- execweave-demo:start -->
+## Diese Demo reproduzieren
+
+Der Screenshot oben stammt aus einer echten ExecWeave-v0.6.3-Live-Session. Dieser Workload erzeugt absichtlich genügend Aktivität für einen aussagekräftigen execution graph: mehrere Python modules, JSON/CSV files, tests, file inspection und ausgehende HTTP requests.
+
+Starte eine lokale Agent CLI unter ExecWeave, zum Beispiel:
+
+```bash
+execweave live --open -- claude
+```
+
+Füge danach diesen workload prompt in den Agent ein:
+
+```text
+Create a small Python project in ./execweave-demo with 8 modules,
+generate sample JSON and CSV data, run the program and tests,
+inspect the generated files, and fetch example.com plus the GitHub API.
+```
+
+Derselbe workload funktioniert auch mit `codex`, `gemini`, `cursor` oder `opencode`. Die genaue Anzahl von nodes, edges, events, processes und endpoints hängt von OS, Agent version und environment ab. ExecWeave zeichnet tatsächlich beobachtetes runtime evidence auf; der Screenshot zeigt einen konkreten Lauf und keinen fest erwarteten graph.
+<!-- execweave-demo:end -->
+
 ## Installation
 
 ExecWeave wird auf PyPI als normales Python wheel/sdist veröffentlicht. Installieren Sie die neueste Release-Version mit:
@@ -68,41 +90,41 @@ execweave live --open -- python my_agent.py
 execweave live --open -- ollama serve
 ```
 
-`execweave live` streamt Process-, File- und Network-Evidence für den von ExecWeave gestarteten Command Tree. Agent-Semantic-Hooks, Model-Runtime-API-Metadaten oder Inference-Gateway-Routing-Metadaten werden **nicht** automatisch in den Live Viewer injiziert.
+`execweave live` streamt Process-, File- und Network-Evidence für den gestarteten Command Tree. Seit v0.6.4 speisen konfigurierte Claude/Codex/Gemini/Cursor-Hooks und das OpenCode-Plugin automatisch den run-spezifischen Live-Sidecar; unter ExecWeave gestartete Ollama-, llama.cpp- und vLLM-Server erhalten zusätzlich einen automatischen lokalen Model-Catalog-Probe.
 
 #### Live-Capability-Matrix
 
 | Integration | Direct OS-runtime live | Specialized metadata | Automatisch im Live Viewer |
 | --- | --- | --- | --- |
-| Claude Code | Ja | `execweave-claude-record` / hooks | Nein |
-| OpenAI Codex | Ja | `execweave-codex-record` / hooks | Nein |
-| Gemini CLI | Ja | `execweave-gemini-record` / hooks | Nein |
-| Cursor | Ja | `execweave-cursor-record` / hooks | Nein |
-| OpenCode | Ja | `execweave-opencode-record` / plugin | Nein |
-| Ollama | Ja, wenn es unter ExecWeave gestartet wird, z. B. `ollama serve` | `execweave-model-runtime event/probe --runtime ollama` | Nein |
-| llama.cpp | Ja, wenn der lokale Server unter ExecWeave gestartet wird | `execweave-model-runtime event/probe --runtime llamacpp` | Nein |
-| vLLM | Ja, wenn der lokale Server unter ExecWeave gestartet wird | `execweave-model-runtime event/probe --runtime vllm` | Nein |
-| LM Studio | Nur für einen lokalen Prozess, der unter ExecWeave gestartet wurde; ein bereits laufender Server wird nicht attached | `execweave-model-runtime event/probe --runtime lmstudio` | Nein |
-| LiteLLM Proxy | Ja, wenn der lokale Proxy unter ExecWeave gestartet wird | `execweave-inference-gateway event --gateway litellm` | Nein |
+| Claude Code | Ja | `execweave-claude-record` / hooks | Ja (konfigurierter Hook/Plugin) |
+| OpenAI Codex | Ja | `execweave-codex-record` / hooks | Ja (konfigurierter Hook/Plugin) |
+| Gemini CLI | Ja | `execweave-gemini-record` / hooks | Ja (konfigurierter Hook/Plugin) |
+| Cursor | Ja | `execweave-cursor-record` / hooks | Ja (konfigurierter Hook/Plugin) |
+| OpenCode | Ja | `execweave-opencode-record` / plugin | Ja (konfigurierter Hook/Plugin) |
+| Ollama | Ja, wenn es unter ExecWeave gestartet wird, z. B. `ollama serve` | `execweave-model-runtime event/probe --runtime ollama` | Ja (automatischer lokaler Probe) |
+| llama.cpp | Ja, wenn der lokale Server unter ExecWeave gestartet wird | `execweave-model-runtime event/probe --runtime llamacpp` | Ja (automatischer lokaler Probe) |
+| vLLM | Ja, wenn der lokale Server unter ExecWeave gestartet wird | `execweave-model-runtime event/probe --runtime vllm` | Ja (automatischer lokaler Probe) |
+| LM Studio | Nur für einen lokalen Prozess, der unter ExecWeave gestartet wurde; ein bereits laufender Server wird nicht attached | `execweave-model-runtime event/probe --runtime lmstudio` | Ja (automatischer Probe nach erfolgreichem Start mit `--port`) |
+| LiteLLM Proxy | Ja, wenn der lokale Proxy unter ExecWeave gestartet wird | `execweave-inference-gateway event --gateway litellm` / configured callback | Ja (konfigurierter Callback) |
 | OpenRouter | Kein direkt startbarer Remote-Service-Prozess; stattdessen den lokalen Client/Agent unter `live` ausführen | `execweave-inference-gateway event/generation --gateway openrouter` | Nein |
 
 Für einen bereits laufenden Ollama-Server kann `execweave-model-runtime probe --runtime ollama` den geladenen Modellzustand snapshotten. Bei OpenRouter kann `live` den lokalen Client und dessen Network-Aktivität beobachten; Gateway-Routing-/Usage-Metadaten bleiben eine separate Evidence-Layer.
 
-<!-- v0.6.3-live -->
-### v0.6.3 Live-Observability
+<!-- v0.6.4-live -->
+### v0.6.4 Live-Observability
 
-Dieselbe Live-Session kann im Browser oder Terminal betrachtet werden:
+`top` lässt den Agent im ursprünglichen Terminal interaktiv und öffnet das Dashboard in einem separaten Terminalfenster:
 
 ```bash
-execweave top -- codex          # Terminal dashboard
-execweave top --open -- codex   # Terminal + Web Viewer
+execweave top -- codex
+execweave top --open -- codex
 ```
 
-Live-Updates verwenden inkrementelle Snapshots/Deltas mit begrenzter History, statt den vollständigen Graph wiederholt neu zu bauen und zu übertragen. Live- und Standalone-Viewer unterstützen einen persistenten Dark/Light-Theme-Schalter. Unter Linux werden sehr große rekursive Filesystem-Scopes vorab geprüft und bei Bedarf automatisch von inotify auf Polling zurückgestuft, sodass ein erschöpfter inotify-Watch-Pool die Session nicht abbricht.
+Live-Updates verwenden inkrementelle Snapshots/Deltas mit begrenzter History. Live- und Standalone-Viewer behalten die Dark/Light-Auswahl. Unter Linux werden sehr große rekursive Filesystem-Scopes vorab geprüft und bei Bedarf automatisch von inotify auf Polling zurückgestuft.
 
-`live` ist eine generische OS-Runtime-Ansicht und keine Integrations-Whitelist. Spezialisierte Agent-Semantic-, Model-Runtime- und Gateway-Metadaten bleiben getrennte Evidence-Layer und werden in v0.6.3 nicht automatisch in den Live Viewer injiziert.
+v0.6.4 erstellt für jeden Live-Run einen gemeinsamen Specialized-Evidence-Sidecar. Konfigurierte Claude/Codex/Gemini/Cursor-Hooks und das OpenCode-Plugin erscheinen automatisch im selben Live Graph; unter ExecWeave gestartete Ollama-, llama.cpp- und vLLM-Server werden automatisch über ihre Loopback-API nach dem Model Catalog abgefragt. Diese spezialisierten Live-Events sind provisional; nach Ende des Befehls wird der finale Graph aus dem canonical Runtime+Semantic-Merge neu aufgebaut. Fehlende Evidence wird nicht erfunden.
 
-Der reproduzierbare Graph-Scalability-Benchmark läuft mit `execweave-scalability`; CI deckt 10k, 100k und 1M synthetische Events ab.
+Mit `execweave-scalability` lässt sich der Graph-Scalability-Benchmark reproduzieren; CI deckt 10k, 100k und 1M synthetische Events ab.
 
 #### Scalability-Benchmark
 
@@ -353,7 +375,7 @@ Security Findings bleiben hinsichtlich ihrer Evidence-Grenzen explizit. Ein mög
 
 ## Aktueller Stand
 
-ExecWeave `main` ist derzeit **v0.6.3** und wird aktiv weiterentwickelt.
+ExecWeave `main` ist derzeit **v0.6.4** und wird aktiv weiterentwickelt.
 
 Der Baseline umfasst Runtime Collection, Graph-Materialisierung und -Abfragen, Standalone/Live Viewer, Claude/Codex/Gemini/Cursor/OpenCode Semantic Integrations, konservative Tool → Process Correlation, OpenRouter/LiteLLM Gateway Metadata, Ollama/llama.cpp/vLLM/LM Studio Runtime Metadata, exakte Gateway ↔ Model Runtime Request Identity, veröffentlichte PyPI wheel/sdist-Pakete, reproduzierbares Overhead Benchmarking, Cross-Platform Command-Launcher-Kompatibilität, Large-Graph Browser Safety Guards, inkrementelles Live JSONL Tail/Cache und Cross-Platform CI unter Python 3.10/3.12.
 
@@ -385,4 +407,4 @@ Beiträge sind willkommen, insbesondere zu native OS collectors, zusätzlichen A
 
 ## Lizenz
 
-Siehe [`LICENSE`](LICENSE).
+Siehe [`LICENSE`](LICENSE).

{execweave-0.6.3 → execweave-0.6.4}/README.fr.md

@@ -26,9 +26,31 @@ ExecWeave est un projet d’observabilité open source et local-first qui transf
 > **L’Event est la ground truth ; le Graph est une materialized view.**
 
 <p align="center">
-  <img src="docs/assets/execweave-viewer.png" alt="ExecWeave Live execution graph" width="100%">
+  <img src="docs/assets/execweave-launch-demo-v5-x.gif" alt="ExecWeave Live execution graph" width="100%">
 </p>
 
+<!-- execweave-demo:start -->
+## Reproduire cette démo
+
+La capture ci-dessus provient d'une véritable session live ExecWeave v0.6.3. Ce workload provoque volontairement assez d'activité pour rendre l'execution graph parlant : plusieurs modules Python, des fichiers JSON/CSV, des tests, l'inspection de fichiers et des requêtes HTTP sortantes.
+
+Exécutez un Agent CLI local sous ExecWeave, par exemple :
+
+```bash
+execweave live --open -- claude
+```
+
+Puis collez ce workload prompt dans l'Agent :
+
+```text
+Create a small Python project in ./execweave-demo with 8 modules,
+generate sample JSON and CSV data, run the program and tests,
+inspect the generated files, and fetch example.com plus the GitHub API.
+```
+
+Le même workload fonctionne avec `codex`, `gemini`, `cursor` ou `opencode`. Le nombre exact de nodes, edges, events, processes et endpoints dépend de l'OS, de la version de l'Agent et de l'environnement. ExecWeave enregistre le runtime evidence réellement observé ; la capture montre une exécution concrète, pas un graph attendu fixe.
+<!-- execweave-demo:end -->
+
 ## Installation
 
 ExecWeave est publié sur PyPI sous forme de wheel/sdist Python standard. Installez la dernière release avec :
@@ -68,41 +90,41 @@ execweave live --open -- python my_agent.py
 execweave live --open -- ollama serve
 ```
 
-`execweave live` diffuse les preuves process, file et network de l’arbre de commandes qu’il lance. Il **n’injecte pas automatiquement** dans le Live Viewer les hooks sémantiques Agent, les métadonnées API du Model Runtime ni les métadonnées de routage de l’Inference Gateway.
+`execweave live` diffuse les preuves process, file et network de l’arbre de commandes qu’il lance. Depuis v0.6.4, les hooks Claude/Codex/Gemini/Cursor configurés et le plugin OpenCode alimentent automatiquement le live sidecar du run ; les serveurs Ollama, llama.cpp et vLLM lancés sous ExecWeave reçoivent aussi un probe local automatique du catalogue de modèles.
 
 #### Matrice des capacités Live
 
 | Integration | Direct OS-runtime live | Specialized metadata | Automatiquement dans le Live Viewer |
 | --- | --- | --- | --- |
-| Claude Code | Oui | `execweave-claude-record` / hooks | Non |
-| OpenAI Codex | Oui | `execweave-codex-record` / hooks | Non |
-| Gemini CLI | Oui | `execweave-gemini-record` / hooks | Non |
-| Cursor | Oui | `execweave-cursor-record` / hooks | Non |
-| OpenCode | Oui | `execweave-opencode-record` / plugin | Non |
-| Ollama | Oui, lorsqu’il est lancé sous ExecWeave, par exemple `ollama serve` | `execweave-model-runtime event/probe --runtime ollama` | Non |
-| llama.cpp | Oui, lorsque son serveur local est lancé sous ExecWeave | `execweave-model-runtime event/probe --runtime llamacpp` | Non |
-| vLLM | Oui, lorsque son serveur local est lancé sous ExecWeave | `execweave-model-runtime event/probe --runtime vllm` | Non |
-| LM Studio | Oui uniquement pour un processus local lancé sous ExecWeave ; un serveur déjà actif n’est pas attaché | `execweave-model-runtime event/probe --runtime lmstudio` | Non |
-| LiteLLM Proxy | Oui, lorsque le proxy local est lancé sous ExecWeave | `execweave-inference-gateway event --gateway litellm` | Non |
+| Claude Code | Oui | `execweave-claude-record` / hooks | Oui (hook/plugin configuré) |
+| OpenAI Codex | Oui | `execweave-codex-record` / hooks | Oui (hook/plugin configuré) |
+| Gemini CLI | Oui | `execweave-gemini-record` / hooks | Oui (hook/plugin configuré) |
+| Cursor | Oui | `execweave-cursor-record` / hooks | Oui (hook/plugin configuré) |
+| OpenCode | Oui | `execweave-opencode-record` / plugin | Oui (hook/plugin configuré) |
+| Ollama | Oui, lorsqu’il est lancé sous ExecWeave, par exemple `ollama serve` | `execweave-model-runtime event/probe --runtime ollama` | Oui (probe local automatique) |
+| llama.cpp | Oui, lorsque son serveur local est lancé sous ExecWeave | `execweave-model-runtime event/probe --runtime llamacpp` | Oui (probe local automatique) |
+| vLLM | Oui, lorsque son serveur local est lancé sous ExecWeave | `execweave-model-runtime event/probe --runtime vllm` | Oui (probe local automatique) |
+| LM Studio | Oui uniquement pour un processus local lancé sous ExecWeave ; un serveur déjà actif n’est pas attaché | `execweave-model-runtime event/probe --runtime lmstudio` | Oui (probe auto après démarrage réussi avec `--port`) |
+| LiteLLM Proxy | Oui, lorsque le proxy local est lancé sous ExecWeave | `execweave-inference-gateway event --gateway litellm` / configured callback | Oui (callback configuré) |
 | OpenRouter | Pas de processus de service distant à lancer directement ; exécutez plutôt le client/Agent local sous `live` | `execweave-inference-gateway event/generation --gateway openrouter` | Non |
 
 Pour un serveur Ollama déjà actif, utilisez `execweave-model-runtime probe --runtime ollama` afin de prendre un snapshot de l’état des modèles chargés. Pour OpenRouter, `live` peut observer le client local et son activité réseau, tandis que les métadonnées de routage/usage du gateway restent une couche de preuve distincte.
 
-<!-- v0.6.3-live -->
-### Observabilité Live v0.6.3
+<!-- v0.6.4-live -->
+### Observabilité Live v0.6.4
 
-Utilisez la même session Live dans le navigateur ou le terminal :
+`top` garde l’Agent interactif dans le Terminal d’origine et ouvre le dashboard dans une fenêtre Terminal séparée :
 
 ```bash
-execweave top -- codex          # Terminal dashboard
-execweave top --open -- codex   # Terminal + Web Viewer
+execweave top -- codex
+execweave top --open -- codex
 ```
 
-Les mises à jour Live utilisent des snapshots/deltas incrémentaux avec un historique borné au lieu de reconstruire et retransférer continuellement le graphe complet. Les Viewers Live et standalone proposent un sélecteur Dark/Light persistant. Sous Linux, les très grands scopes filesystem récursifs sont pré-évalués et basculent automatiquement d’inotify vers le polling si nécessaire, afin qu’un pool de watches inotify épuisé n’interrompe pas la session.
+Les mises à jour Live utilisent des snapshots/deltas incrémentaux avec un historique borné. Les Viewers Live et standalone conservent le choix Dark/Light. Sous Linux, les très grands scopes filesystem récursifs sont pré-évalués et basculent automatiquement d’inotify vers le polling si nécessaire.
 
-`live` est une vue OS-runtime générique et non une whitelist d’intégrations. Les métadonnées spécialisées Agent semantic, Model Runtime et Gateway restent des couches de preuve séparées et ne sont pas automatiquement injectées dans le Live Viewer en v0.6.3.
+v0.6.4 crée un specialized-evidence sidecar partagé pour chaque run live. Les hooks Claude/Codex/Gemini/Cursor configurés et le plugin OpenCode arrivent automatiquement dans le même Live Graph ; les serveurs Ollama, llama.cpp et vLLM lancés sous ExecWeave sont interrogés automatiquement via leur API loopback pour le catalogue de modèles. Ces événements spécialisés live sont provisoires ; après la fin de la commande, le graphe final est reconstruit depuis le merge canonical runtime + semantic. Aucune preuve absente n’est inventée.
 
-Lancez le benchmark reproductible de scalabilité du graphe avec `execweave-scalability` ; la CI couvre 10k, 100k et 1M événements synthétiques.
+Lancez `execweave-scalability` pour reproduire le benchmark de scalabilité ; la CI couvre 10k, 100k et 1M événements synthétiques.
 
 #### Benchmark de scalabilité
 
@@ -353,7 +375,7 @@ Les findings de sécurité restent explicites sur les limites de la preuve. Un p
 
 ## État actuel
 
-ExecWeave `main` est actuellement en **v0.6.3** et reste en développement actif.
+ExecWeave `main` est actuellement en **v0.6.4** et reste en développement actif.
 
 Le baseline comprend la collecte runtime, la matérialisation/interrogation du graphe, les Viewers standalone/live, les intégrations semantic Claude/Codex/Gemini/Cursor/OpenCode, la corrélation conservatrice Tool → Process, les métadonnées gateway OpenRouter/LiteLLM, les métadonnées runtime Ollama/llama.cpp/vLLM/LM Studio, l’identité exacte Gateway ↔ Model Runtime request, les packages wheel/sdist publiés sur PyPI, un benchmark de surcharge reproductible, la compatibilité cross-platform du launcher, les garde-fous de navigateur pour grands graphes, le tail/cache Live JSONL incrémental et la CI cross-platform sous Python 3.10/3.12.
 
@@ -385,4 +407,4 @@ Les contributions sont les bienvenues, notamment autour des native OS collectors
 
 ## Licence
 
-Voir [`LICENSE`](LICENSE).
+Voir [`LICENSE`](LICENSE).

{execweave-0.6.3 → execweave-0.6.4}/README.ja.md

@@ -20,9 +20,31 @@ ExecWeave は、AI エージェントの活動をインタラクティブな exe
 > **Event is ground truth. The graph is a materialized view.**
 
 <p align="center">
-  <img src="docs/assets/execweave-viewer.png" alt="ExecWeave Live execution graph" width="100%">
+  <img src="docs/assets/execweave-launch-demo-v5-x.gif" alt="ExecWeave Live execution graph" width="100%">
 </p>
 
+<!-- execweave-demo:start -->
+## この Demo を再現する
+
+上のスクリーンショットは、実際の ExecWeave v0.6.3 live session です。この workload は execution graph の挙動が分かりやすくなるよう、複数の Python modules、JSON/CSV files、tests、file inspection、外向き HTTP requests を意図的に発生させます。
+
+ローカルの Agent CLI を ExecWeave の下で実行します。例：
+
+```bash
+execweave live --open -- claude
+```
+
+その後、次の workload prompt を Agent に貼り付けます：
+
+```text
+Create a small Python project in ./execweave-demo with 8 modules,
+generate sample JSON and CSV data, run the program and tests,
+inspect the generated files, and fetch example.com plus the GitHub API.
+```
+
+同じ workload は `codex`、`gemini`、`cursor`、`opencode` でも利用できます。node、edge、event、process、endpoint の数は OS、Agent version、environment によって変わります。ExecWeave が記録するのは実際に観測された runtime evidence であり、上図は固定の期待 graph ではなく一つの実行例です。
+<!-- execweave-demo:end -->
+
 ## Installation
 
 ExecWeave は PyPI で正式公開されています。最新の公開 release は次でインストールできます。
@@ -62,39 +84,39 @@ execweave live --open -- python my_agent.py
 execweave live --open -- ollama serve
 ```
 
-`execweave live` は、起動した command tree の process / file / network evidence をリアルタイム表示します。Agent semantic hooks、model-runtime API metadata、inference-gateway routing metadata は **現在 Live Viewer に自動注入されません**。
+`execweave live` は起動した command tree の process/file/network evidence をリアルタイム表示します。v0.6.4 では設定済み Claude/Codex/Gemini/Cursor hooks と OpenCode plugin が run 固有の live sidecar へ自動送信され、ExecWeave 配下で起動した Ollama、llama.cpp、vLLM server にはローカル model-catalog probe が自動実行されます。
 
 #### Live capability matrix
 
 | Integration | Direct OS-runtime live | Specialized metadata | Auto in Live Viewer |
 | --- | --- | --- | --- |
-| Claude Code | Yes | `execweave-claude-record` / hooks | No |
-| OpenAI Codex | Yes | `execweave-codex-record` / hooks | No |
-| Gemini CLI | Yes | `execweave-gemini-record` / hooks | No |
-| Cursor | Yes | `execweave-cursor-record` / hooks | No |
-| OpenCode | Yes | `execweave-opencode-record` / plugin | No |
-| Ollama | Yes, ExecWeave から起動する場合（例: `ollama serve`） | `execweave-model-runtime event/probe --runtime ollama` | No |
-| llama.cpp | Yes, ローカル server を ExecWeave から起動する場合 | `execweave-model-runtime event/probe --runtime llamacpp` | No |
-| vLLM | Yes, ローカル server を ExecWeave から起動する場合 | `execweave-model-runtime event/probe --runtime vllm` | No |
-| LM Studio | ExecWeave から起動したローカル process のみ。既存 server には attach しません | `execweave-model-runtime event/probe --runtime lmstudio` | No |
-| LiteLLM Proxy | Yes, ローカル proxy を ExecWeave から起動する場合 | `execweave-inference-gateway event --gateway litellm` | No |
+| Claude Code | Yes | `execweave-claude-record` / hooks | Yes（設定済み hook/plugin） |
+| OpenAI Codex | Yes | `execweave-codex-record` / hooks | Yes（設定済み hook/plugin） |
+| Gemini CLI | Yes | `execweave-gemini-record` / hooks | Yes（設定済み hook/plugin） |
+| Cursor | Yes | `execweave-cursor-record` / hooks | Yes（設定済み hook/plugin） |
+| OpenCode | Yes | `execweave-opencode-record` / plugin | Yes（設定済み hook/plugin） |
+| Ollama | Yes, ExecWeave から起動する場合（例: `ollama serve`） | `execweave-model-runtime event/probe --runtime ollama` | Yes（自動ローカル probe） |
+| llama.cpp | Yes, ローカル server を ExecWeave から起動する場合 | `execweave-model-runtime event/probe --runtime llamacpp` | Yes（自動ローカル probe） |
+| vLLM | Yes, ローカル server を ExecWeave から起動する場合 | `execweave-model-runtime event/probe --runtime vllm` | Yes（自動ローカル probe） |
+| LM Studio | ExecWeave から起動したローカル process のみ。既存 server には attach しません | `execweave-model-runtime event/probe --runtime lmstudio` | Yes（明示した `--port` で起動成功後に自動 probe） |
+| LiteLLM Proxy | Yes, ローカル proxy を ExecWeave から起動する場合 | `execweave-inference-gateway event --gateway litellm` / configured callback | Yes（設定済み callback） |
 | OpenRouter | Remote service 自体は direct live 不可。ローカル client/Agent を live してください | `execweave-inference-gateway event/generation --gateway openrouter` | No |
 
 Ollama がすでにバックグラウンドで動作中なら、`execweave-model-runtime probe --runtime ollama` で loaded-model state を snapshot できます。OpenRouter では `live` がローカル client と network activity を観測し、gateway routing/usage metadata は別の evidence layer のままです。
 
-<!-- v0.6.3-live -->
-### v0.6.3 ライブ可観測性
+<!-- v0.6.4-live -->
+### v0.6.4 ライブ可観測性
 
-同じ live session をブラウザまたは Terminal で確認できます：
+`top` は Agent を元の Terminal で対話可能なままにし、dashboard を別の Terminal ウィンドウで開きます：
 
 ```bash
-execweave top -- codex          # Terminal dashboard
-execweave top --open -- codex   # Terminal + Web Viewer
+execweave top -- codex
+execweave top --open -- codex
 ```
 
-Live 更新は増分 snapshot/delta と有界履歴を使用し、graph 全体の再構築・再送を繰り返しません。Live と standalone Viewer は、設定を保持する Dark/Light 切り替えに対応します。Linux では非常に大きな recursive filesystem scope を事前に確認し、inotify watch 容量が不足する場合は自動的に polling へフォールバックするため、inotify watch exhaustion で session 全体が停止しません。
+Live 更新は増分 snapshot/delta と有界履歴を使用し、Live/standalone Viewer は永続的な Dark/Light theme をサポートします。Linux の大規模 recursive filesystem scope は事前確認され、必要なら inotify から polling へ自動フォールバックします。
 
-`live` は汎用 OS-runtime view であり integration whitelist ではありません。Agent semantic、model-runtime、gateway metadata は分離された evidence layer のままで、v0.6.3 では Live Viewer に自動注入されません。
+v0.6.4 では各 live run に共有 specialized-evidence sidecar があります。設定済み Claude/Codex/Gemini/Cursor hooks と OpenCode plugin は同じ Live Graph に自動反映され、ExecWeave 配下で起動した Ollama、llama.cpp、vLLM server は loopback API から model catalog を自動 probe します。live specialized events は provisional であり、command 終了後は canonical runtime + semantic merge から final graph を再構築します。存在しない evidence は推測しません。
 
 `execweave-scalability` で graph scalability benchmark を再現でき、CI は 10k、100k、1M synthetic events を検証します。
 
@@ -217,7 +239,7 @@ Security findings は evidence limit を維持し、possible sensitive-file →
 
 ## Current status
 
-ExecWeave `main` は現在 **v0.6.3**。Runtime collection、execution graph、5 種の Agent/IDE integration、OpenRouter/LiteLLM、Ollama/llama.cpp/vLLM/LM Studio、exact Gateway ↔ Runtime identity、公開済み PyPI packaging、reference overhead benchmark、cross-platform command-launcher compatibility、large-graph browser safety guards、incremental Live JSONL tail/cache、cross-platform CI が baseline に含まれます。
+ExecWeave `main` は現在 **v0.6.4**。Runtime collection、execution graph、5 種の Agent/IDE integration、OpenRouter/LiteLLM、Ollama/llama.cpp/vLLM/LM Studio、exact Gateway ↔ Runtime identity、公開済み PyPI packaging、reference overhead benchmark、cross-platform command-launcher compatibility、large-graph browser safety guards、incremental Live JSONL tail/cache、cross-platform CI が baseline に含まれます。
 
 ## Privacy
 
@@ -235,4 +257,4 @@ ExecWeave は local-first です。Runtime events、semantic sidecars、graphs
 
 ## License
 
-[`LICENSE`](LICENSE) を参照してください。
+[`LICENSE`](LICENSE) を参照してください。

{execweave-0.6.3 → execweave-0.6.4}/README.ko.md

@@ -20,9 +20,31 @@ ExecWeave는 AI 에이전트 활동을 인터랙티브 execution graph로 변환
 > **Event is ground truth. The graph is a materialized view.**
 
 <p align="center">
-  <img src="docs/assets/execweave-viewer.png" alt="ExecWeave Live execution graph" width="100%">
+  <img src="docs/assets/execweave-launch-demo-v5-x.gif" alt="ExecWeave Live execution graph" width="100%">
 </p>
 
+<!-- execweave-demo:start -->
+## 이 Demo 재현하기
+
+위 스크린샷은 실제 ExecWeave v0.6.3 live session입니다. 이 workload는 execution graph가 유용하게 보이도록 여러 Python modules, JSON/CSV files, tests, file inspection, 외부 HTTP requests를 의도적으로 발생시킵니다.
+
+로컬 Agent CLI를 ExecWeave 아래에서 실행합니다. 예:
+
+```bash
+execweave live --open -- claude
+```
+
+그 다음 아래 workload prompt를 Agent에 붙여 넣습니다:
+
+```text
+Create a small Python project in ./execweave-demo with 8 modules,
+generate sample JSON and CSV data, run the program and tests,
+inspect the generated files, and fetch example.com plus the GitHub API.
+```
+
+같은 workload는 `codex`, `gemini`, `cursor`, `opencode`에서도 사용할 수 있습니다. 실제 node, edge, event, process, endpoint 수는 OS, Agent version, environment에 따라 달라집니다. ExecWeave는 실제로 관측된 runtime evidence를 기록하며, 위 그림은 고정된 예상 graph가 아니라 하나의 실제 실행 예입니다.
+<!-- execweave-demo:end -->
+
 ## Installation
 
 ExecWeave는 PyPI에 정식 공개되어 있습니다. 최신 공개 release 설치:
@@ -62,39 +84,39 @@ execweave live --open -- python my_agent.py
 execweave live --open -- ollama serve
 ```
 
-`execweave live`는 자신이 시작한 command tree의 process, file, network evidence를 실시간으로 보여 줍니다. Agent semantic hooks, model-runtime API metadata, inference-gateway routing metadata는 **현재 Live Viewer에 자동으로 주입되지 않습니다**.
+`execweave live`는 실행한 command tree의 process/file/network evidence를 실시간으로 표시합니다. v0.6.4부터 설정된 Claude/Codex/Gemini/Cursor hooks와 OpenCode plugin은 run 전용 live sidecar로 자동 전송되며, ExecWeave 아래에서 실행한 Ollama, llama.cpp, vLLM server에는 로컬 model-catalog probe가 자동 적용됩니다.
 
 #### Live capability matrix
 
 | Integration | Direct OS-runtime live | Specialized metadata | Auto in Live Viewer |
 | --- | --- | --- | --- |
-| Claude Code | Yes | `execweave-claude-record` / hooks | No |
-| OpenAI Codex | Yes | `execweave-codex-record` / hooks | No |
-| Gemini CLI | Yes | `execweave-gemini-record` / hooks | No |
-| Cursor | Yes | `execweave-cursor-record` / hooks | No |
-| OpenCode | Yes | `execweave-opencode-record` / plugin | No |
-| Ollama | Yes, ExecWeave가 시작한 경우(예: `ollama serve`) | `execweave-model-runtime event/probe --runtime ollama` | No |
-| llama.cpp | Yes, 로컬 server를 ExecWeave가 시작한 경우 | `execweave-model-runtime event/probe --runtime llamacpp` | No |
-| vLLM | Yes, 로컬 server를 ExecWeave가 시작한 경우 | `execweave-model-runtime event/probe --runtime vllm` | No |
-| LM Studio | ExecWeave가 시작한 로컬 process만 가능하며 기존 server에는 attach하지 않음 | `execweave-model-runtime event/probe --runtime lmstudio` | No |
-| LiteLLM Proxy | Yes, 로컬 proxy를 ExecWeave가 시작한 경우 | `execweave-inference-gateway event --gateway litellm` | No |
+| Claude Code | Yes | `execweave-claude-record` / hooks | Yes (설정된 hook/plugin) |
+| OpenAI Codex | Yes | `execweave-codex-record` / hooks | Yes (설정된 hook/plugin) |
+| Gemini CLI | Yes | `execweave-gemini-record` / hooks | Yes (설정된 hook/plugin) |
+| Cursor | Yes | `execweave-cursor-record` / hooks | Yes (설정된 hook/plugin) |
+| OpenCode | Yes | `execweave-opencode-record` / plugin | Yes (설정된 hook/plugin) |
+| Ollama | Yes, ExecWeave가 시작한 경우(예: `ollama serve`) | `execweave-model-runtime event/probe --runtime ollama` | Yes (자동 로컬 probe) |
+| llama.cpp | Yes, 로컬 server를 ExecWeave가 시작한 경우 | `execweave-model-runtime event/probe --runtime llamacpp` | Yes (자동 로컬 probe) |
+| vLLM | Yes, 로컬 server를 ExecWeave가 시작한 경우 | `execweave-model-runtime event/probe --runtime vllm` | Yes (자동 로컬 probe) |
+| LM Studio | ExecWeave가 시작한 로컬 process만 가능하며 기존 server에는 attach하지 않음 | `execweave-model-runtime event/probe --runtime lmstudio` | Yes (명시적 `--port`로 시작 성공 후 자동 probe) |
+| LiteLLM Proxy | Yes, 로컬 proxy를 ExecWeave가 시작한 경우 | `execweave-inference-gateway event --gateway litellm` / configured callback | Yes (설정된 callback) |
 | OpenRouter | Remote service 자체는 direct live 불가. 로컬 client/Agent를 live해야 함 | `execweave-inference-gateway event/generation --gateway openrouter` | No |
 
 Ollama가 이미 백그라운드에서 실행 중이면 `execweave-model-runtime probe --runtime ollama`로 loaded-model state를 snapshot할 수 있습니다. OpenRouter에서는 `live`가 로컬 client와 network activity를 관측하고 gateway routing/usage metadata는 별도 evidence layer로 유지됩니다.
 
-<!-- v0.6.3-live -->
-### v0.6.3 라이브 관측성
+<!-- v0.6.4-live -->
+### v0.6.4 라이브 관측성
 
-동일한 live session을 브라우저 또는 Terminal에서 확인할 수 있습니다:
+`top`은 Agent를 기존 Terminal에서 계속 대화형으로 유지하고 dashboard를 별도 Terminal 창에서 엽니다:
 
 ```bash
-execweave top -- codex          # Terminal dashboard
-execweave top --open -- codex   # Terminal + Web Viewer
+execweave top -- codex
+execweave top --open -- codex
 ```
 
-Live 업데이트는 증분 snapshot/delta와 제한된 이력을 사용하므로 전체 graph를 반복해서 재구성하고 전송하지 않습니다. Live 및 standalone Viewer는 선택을 기억하는 Dark/Light 전환을 지원합니다. Linux에서는 매우 큰 recursive filesystem scope를 사전 점검하고 inotify watch 용량이 부족하면 자동으로 polling으로 전환하므로 inotify watch exhaustion 때문에 session 전체가 중단되지 않습니다.
+Live 업데이트는 증분 snapshot/delta와 제한된 이력을 사용하며 Live/standalone Viewer는 지속되는 Dark/Light theme를 지원합니다. Linux의 매우 큰 recursive filesystem scope는 사전 점검 후 필요하면 inotify에서 polling으로 자동 전환됩니다.
 
-`live`는 일반 OS-runtime view이며 integration whitelist가 아닙니다. Agent semantic, model-runtime, gateway metadata는 분리된 evidence layer로 유지되며 v0.6.3에서는 Live Viewer에 자동으로 주입되지 않습니다.
+v0.6.4에서는 각 live run에 공유 specialized-evidence sidecar가 생성됩니다. 설정된 Claude/Codex/Gemini/Cursor hooks와 OpenCode plugin은 같은 Live Graph로 자동 유입되고, ExecWeave 아래에서 실행한 Ollama, llama.cpp, vLLM server는 loopback API로 model catalog를 자동 probe합니다. live specialized events는 provisional이며 command 종료 후 canonical runtime + semantic merge에서 final graph를 다시 만듭니다. 관찰되지 않은 evidence는 추측하지 않습니다.
 
 `execweave-scalability`로 graph scalability benchmark를 재현할 수 있으며 CI는 10k, 100k, 1M synthetic events를 검증합니다.
 
@@ -217,7 +239,7 @@ Security findings는 evidence limit를 유지하며 possible sensitive-file →
 
 ## Current status
 
-ExecWeave `main`은 현재 **v0.6.3**입니다. Runtime collection, execution graph, 5개 Agent/IDE integration, OpenRouter/LiteLLM, Ollama/llama.cpp/vLLM/LM Studio, exact Gateway ↔ Runtime identity, 공개된 PyPI packaging, reference overhead benchmark, cross-platform command-launcher compatibility, large-graph browser safety guards, incremental Live JSONL tail/cache, cross-platform CI가 baseline에 포함됩니다.
+ExecWeave `main`은 현재 **v0.6.4**입니다. Runtime collection, execution graph, 5개 Agent/IDE integration, OpenRouter/LiteLLM, Ollama/llama.cpp/vLLM/LM Studio, exact Gateway ↔ Runtime identity, 공개된 PyPI packaging, reference overhead benchmark, cross-platform command-launcher compatibility, large-graph browser safety guards, incremental Live JSONL tail/cache, cross-platform CI가 baseline에 포함됩니다.
 
 ## Privacy
 
@@ -235,4 +257,4 @@ ExecWeave는 local-first입니다. Runtime events, semantic sidecars, graphs, re
 
 ## License
 
-[`LICENSE`](LICENSE)를 참조하세요.
+[`LICENSE`](LICENSE)를 참조하세요.