aria-code 4.2.0__tar.gz → 4.4.0__tar.gz

{aria_code-4.2.0/aria_code.egg-info → aria_code-4.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aria-code
-Version: 4.2.0
+Version: 4.4.0
 Summary: AI-powered financial terminal — local LLM + multi-source market data
 License: BUSL-1.1
 Project-URL: Homepage, https://github.com/artherahq/aria-code
@@ -51,11 +51,16 @@ Provides-Extra: files
 Requires-Dist: pdfplumber>=0.11.0; extra == "files"
 Requires-Dist: pypdf>=4.3.0; extra == "files"
 Requires-Dist: python-docx>=1.1.2; extra == "files"
+Requires-Dist: python-pptx>=1.0.2; extra == "files"
 Requires-Dist: openpyxl>=3.1.5; extra == "files"
 Requires-Dist: beautifulsoup4>=4.12.3; extra == "files"
 Requires-Dist: Pillow>=10.4.0; extra == "files"
 Provides-Extra: video
 Requires-Dist: opencv-python>=4.8.0; extra == "video"
+Provides-Extra: video-analysis
+Requires-Dist: faster-whisper>=1.0.0; extra == "video-analysis"
+Provides-Extra: adk
+Requires-Dist: google-adk>=2.6.3; extra == "adk"
 Provides-Extra: ibkr
 Requires-Dist: ib_insync>=0.9.86; extra == "ibkr"
 Provides-Extra: futu
@@ -76,6 +81,11 @@ Requires-Dist: longbridge>=1.0.22; extra == "brokers"
 Requires-Dist: alpaca-py>=0.35.0; extra == "brokers"
 Provides-Extra: backtest
 Requires-Dist: vectorbt>=0.26.2; extra == "backtest"
+Provides-Extra: image-gen
+Requires-Dist: torch>=2.2.0; extra == "image-gen"
+Requires-Dist: diffusers>=0.30.0; extra == "image-gen"
+Requires-Dist: transformers>=4.40.0; extra == "image-gen"
+Requires-Dist: accelerate>=0.30.0; extra == "image-gen"
 Provides-Extra: web
 Requires-Dist: ddgs>=9.0; extra == "web"
 Provides-Extra: browser
@@ -100,6 +110,7 @@ Requires-Dist: duckdb>=0.10.3; extra == "full"
 Requires-Dist: pdfplumber>=0.11.0; extra == "full"
 Requires-Dist: pypdf>=4.3.0; extra == "full"
 Requires-Dist: python-docx>=1.1.2; extra == "full"
+Requires-Dist: python-pptx>=1.0.2; extra == "full"
 Requires-Dist: openpyxl>=3.1.5; extra == "full"
 Requires-Dist: beautifulsoup4>=4.12.3; extra == "full"
 Requires-Dist: Pillow>=10.4.0; extra == "full"
@@ -112,6 +123,7 @@ Requires-Dist: duckdb>=0.10.3; extra == "all"
 Requires-Dist: pdfplumber>=0.11.0; extra == "all"
 Requires-Dist: pypdf>=4.3.0; extra == "all"
 Requires-Dist: python-docx>=1.1.2; extra == "all"
+Requires-Dist: python-pptx>=1.0.2; extra == "all"
 Requires-Dist: openpyxl>=3.1.5; extra == "all"
 Requires-Dist: beautifulsoup4>=4.12.3; extra == "all"
 Requires-Dist: Pillow>=10.4.0; extra == "all"
@@ -223,6 +235,7 @@ $ aria-code
 | **83 commands** | Consolidated from ~150 — removed all LLM-replaceable commands; natural language handles the rest |
 | **LLM routing fix** | Model now knows it can call real-time data tools instead of saying "I don't have live data" |
 | **Unified agent loop** | CLI, SDK and self-host backend now run the *same* tested runtime (`runtime.run_turn`) — consistent provider routing, cloud→local fallback and tool execution on every surface |
+| **Warehouse ERP agents** | Read-only carrier-sync, inbound-exception and inventory-health analysis with a dedicated operational signal scheme; see [data contract](docs/warehouse-erp-agents.md) |
 | **One-command install** | `pip install aria-code` (PyPI) or `npm install -g @artheras/aria-code` — verified clean on Python 3.10 / 3.11 / 3.12 |
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
@@ -315,6 +328,28 @@ mindmap
 
 ## 🚀 Quick Start
 
+### Option 0: Download the binary (no Python/Node install needed)
+
+Every tagged release publishes standalone binaries for macOS (arm64), Windows (x64), and Linux (x64) — download from the [Releases page](https://github.com/artherahq/aria-code/releases), no Python or Node.js required:
+
+```bash
+# macOS / Linux
+chmod +x aria-code-macos-arm64   # or aria-code-linux-x64
+./aria-code-macos-arm64
+```
+```powershell
+# Windows
+.\aria-code-windows-x64.exe
+```
+
+> The macOS build is signed + notarized only once `MACOS_CERTIFICATE_P12_BASE64` and the Apple notarization secrets are configured in repo Settings → Secrets (see `.github/workflows/build-native-binaries.yml`) — without them, CI still builds and smoke-tests successfully but produces an **unsigned** binary that macOS Gatekeeper will refuse to run. Until those secrets are added, get a signed macOS build by running `scripts/build_native_binary.sh --notarize` locally on a machine with the real Developer ID cert imported. Windows/Linux binaries need no signing step and run as-is.
+
+Each release also publishes a standalone **`aria-code-mcp-*`** binary — the MCP server (see [Expose aria-code to other MCP clients](#expose-aria-code-to-other-mcp-clients) below) with no Python install needed:
+
+```bash
+claude mcp add aria-code -- /path/to/aria-code-mcp-macos-arm64   # or -linux-x64 / -windows-x64.exe
+```
+
 ### Option 1: Bootstrap (fresh Mac / Linux — recommended)
 
 No Node.js, Python, or Xcode required. One command handles everything:
@@ -937,6 +972,75 @@ Connect any [Model Context Protocol](https://modelcontextprotocol.io) server:
 /mcp tools     # All available MCP tools
 ```
 
+### Expose aria-code to other MCP clients
+
+aria-code can also run *as* an MCP server, so Claude Code, Codex, or any
+other MCP client can call into it directly:
+
+- `aria.market.quote` / `aria.agent.team` / `aria.artifacts.list` — live quote + technicals, multi-agent research, generated artifacts (all read-only)
+- `aria.skill.list` / `aria.skill.get` — discover and fetch the installed portable [SKILL.md](https://github.com/artherahq/aria-skills) workflows: reusable expert playbooks for UI design systems, anti-generic-AI-look design critique, trading-app patterns, equity research, backtest validation, and more. `aria.skill.list` takes an optional `query` to rank by relevance to your task; `aria.skill.get` returns the full workflow text for **you** to follow — skills are instruction documents, not executable tools, so nothing runs server-side. Each result carries an `integrity` field (`verified` for signed catalog skills, `unlocked` for local drop-ins) so you can see a skill's provenance before following it
+- `aria.report.generate` — fetch data and generate a full HTML research report artifact for a symbol
+- `aria.backtest.run` — run a historical strategy simulation (`buy_hold` or `sma_cross`) against real price history
+- `aria.broker.positions` / `aria.broker.list_previews` — read account/positions and recent order previews for a configured broker (US/CN/HK/UK — see `/broker`)
+- `aria.broker.preview_order` — build a risk-checked order preview. **Never places a live trade by itself** — execution requires either a human running `/trade confirm <preview_id>` in the aria-code terminal, or `aria.broker.confirm_order` below (off by default per broker)
+- `aria.broker.confirm_order` — **execute a previously built order preview, for real.** Off by default for every broker. Refuses unconditionally unless (1) a human already ran `/trade allow-chat-confirm <broker_id>` at the aria-code terminal keyboard — retyping the exact broker id, not y/yes — to opt that specific broker into chat-confirmed execution (this opt-in can never be triggered from MCP/chat, only from the terminal), **and** (2) the call itself passes `confirmed: true`. Executions via this path are tagged `source: "chat_mcp"` in the trade audit log (`~/.aria-code/trade_audit.jsonl`) to distinguish them from terminal-confirmed trades. Run `/trade disallow-chat-confirm <broker_id>` at any time to turn it back off
+- `aria.report.chart` — candlestick + MA20/50 chart PNG for a symbol
+- `aria.report.indicator_chart` — candlestick + volume + RSI(14) + MACD(12,26,9) multi-panel chart PNG for a symbol
+- `aria.report.comparison_chart` — normalized (base=100) % return comparison chart PNG across 2+ symbols
+- `aria.report.allocation_chart` — pie chart PNG of a broker account's position weights by market value
+- `aria.report.pdf` / `aria.report.docx` / `aria.report.pptx` — render a Markdown report to PDF, an editable Word document, or a slide deck (one slide per heading) — all chart/report tools save to your artifacts folder
+- `aria.report.canva_design` / `aria.report.canva_upload_asset` — fill a Canva brand template with data (and optionally an uploaded image asset) and export a design draft (needs `/canva connect <client_id> <client_secret>` once — register an app at [canva.com/developers](https://www.canva.com/developers/) and have a brand template ready first; Canva has no API for creating a design from scratch, only autofilling an existing template)
+- `aria.figma.read_file` / `aria.figma.comments` — read a Figma file's page/frame structure (depth-limited summary) and its comments (needs `/apikey set figma <personal_access_token>` once). Read-only — Figma has no public API for writing/creating designs from a script, only its in-app Plugin API can do that
+- `aria.report.estimate_image_cost` — get an illustrative per-image cost estimate (USD) for OpenAI's `gpt-image-1` at a given size/quality, before generating. Free.
+- `aria.report.generate_image` / `aria.report.edit_image` — generate a new image or transform an existing local photo via OpenAI's `gpt-image-1` (needs `/apikey set openai sk-...`). `edit_image` accepts an optional `mask_path` for inpainting — a mask PNG where transparent areas mark what to edit, opaque areas are preserved untouched. **Real per-call cost, billed the instant it succeeds** — both hard-refuse without `confirmed: true`, and calling `aria.report.estimate_image_cost` first is strongly recommended
+- `aria.report.generate_image_local` / `aria.report.edit_image_local` — the same, entirely locally via a self-hosted SDXL-Turbo (no API key, no per-call cost; needs the optional `image_gen` extra and ~4GB of weights on first use)
+- `aria.video.probe` / `aria.video.trim` / `aria.video.concat` / `aria.video.overlay_text` / `aria.video.overlay_audio` / `aria.video.convert` / `aria.video.change_speed` — deterministic local video editing via `ffmpeg` (no AI, no API key)
+- `aria.video.transcribe` / `aria.video.detect_scenes` — local AI-assisted analysis: speech transcription (`faster-whisper`, needs the optional `video_analysis` extra) and scene-cut detection (`opencv`, needs the optional `video` extra) — both produce editing *decisions*, not new pixels
+- `aria.video.generate_estimate` / `aria.video.generate_submit` / `aria.video.generate_status` — cloud AI text-to-video via Kling or Runway (`provider: "kling" | "runway"`). **Real per-request cost, billed the instant submission succeeds** — `generate_submit` hard-refuses without `confirmed: true`, and calling `generate_estimate` first is strongly recommended. Needs `/apikey set kling <access_key>:<secret_key>` or `/apikey set runway <key>`. No local/free option exists for this layer the way SDXL-Turbo covers images — see `video_editor.py`/`video_analysis.py` for the free local-only editing/analysis layers
+
+```bash
+python3 aria_mcp_server.py          # dev, from the repo root
+# or, once installed: aria-code-mcp
+```
+
+Register with Claude Code:
+
+```bash
+claude mcp add aria-code -- python3 /path/to/aria-code/aria_mcp_server.py
+```
+
+Register with Codex (`~/.codex/config.toml`):
+
+```toml
+[mcp_servers.aria_code]
+command = "python3"
+args = ["/path/to/aria-code/aria_mcp_server.py"]
+```
+
+#### Want deeper A-share quant tools too? Register a second server, don't wait for us to add them here
+
+This server intentionally does **not** re-expose Arthera's `quant_engine` quant
+tools (factor calculation, Bayesian-HMM regime detection, northbound-flow,
+Kelly position sizing, options pricing, pairs-trading stats, and a much
+deeper `run_backtest` than the one above — T+1 settlement, limit-up/down
+filtering, `sma_cross`/`rsi_mean_revert`/`momentum` strategies, CN_A or US
+market). That's a separate project (Arthera) with its own MCP server —
+`packages/quant_engine/mcp_server.py` — speaking the identical MCP
+2024-11-05 stdio JSON-RPC protocol, so any client here can register it
+as a second, independent server rather than this one proxying it:
+
+```bash
+claude mcp add arthera-quant -- python3 /path/to/Arthera/packages/quant_engine/mcp_server.py
+```
+
+We keep these two separate on purpose, not as a gap to close later: a few of
+`quant_engine`'s tools (`get_ai_signal`, `get_predictions`,
+`get_market_insights`) call Arthera's own internal Alibaba Cloud service and
+simply won't work without that access, so folding them into this server
+would mean shipping tools that error out for anyone without Arthera's
+private infrastructure. If you don't have access to that repo, everything
+above this section still works standalone.
+
 ---
 
 ## ⚙️ Configuration
@@ -1021,7 +1125,7 @@ converts to Apache 2.0 four years after release.
 
 <p align="center">
   <a href="https://arthera.finance">Website</a> ·
-  <a href="https://github.com/Cinsoul/Arthera">Full Platform</a> ·
+  <a href="https://github.com/artherahq/aria-code">Project Repository</a> ·
   <a href="https://github.com/artherahq/aria-code/issues">Issues</a> ·
   <a href="https://github.com/artherahq/aria-code/discussions">Discussions</a>
 </p>

{aria_code-4.2.0 → aria_code-4.4.0}/README.md

@@ -90,6 +90,7 @@ $ aria-code
 | **83 commands** | Consolidated from ~150 — removed all LLM-replaceable commands; natural language handles the rest |
 | **LLM routing fix** | Model now knows it can call real-time data tools instead of saying "I don't have live data" |
 | **Unified agent loop** | CLI, SDK and self-host backend now run the *same* tested runtime (`runtime.run_turn`) — consistent provider routing, cloud→local fallback and tool execution on every surface |
+| **Warehouse ERP agents** | Read-only carrier-sync, inbound-exception and inventory-health analysis with a dedicated operational signal scheme; see [data contract](docs/warehouse-erp-agents.md) |
 | **One-command install** | `pip install aria-code` (PyPI) or `npm install -g @artheras/aria-code` — verified clean on Python 3.10 / 3.11 / 3.12 |
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
@@ -182,6 +183,28 @@ mindmap
 
 ## 🚀 Quick Start
 
+### Option 0: Download the binary (no Python/Node install needed)
+
+Every tagged release publishes standalone binaries for macOS (arm64), Windows (x64), and Linux (x64) — download from the [Releases page](https://github.com/artherahq/aria-code/releases), no Python or Node.js required:
+
+```bash
+# macOS / Linux
+chmod +x aria-code-macos-arm64   # or aria-code-linux-x64
+./aria-code-macos-arm64
+```
+```powershell
+# Windows
+.\aria-code-windows-x64.exe
+```
+
+> The macOS build is signed + notarized only once `MACOS_CERTIFICATE_P12_BASE64` and the Apple notarization secrets are configured in repo Settings → Secrets (see `.github/workflows/build-native-binaries.yml`) — without them, CI still builds and smoke-tests successfully but produces an **unsigned** binary that macOS Gatekeeper will refuse to run. Until those secrets are added, get a signed macOS build by running `scripts/build_native_binary.sh --notarize` locally on a machine with the real Developer ID cert imported. Windows/Linux binaries need no signing step and run as-is.
+
+Each release also publishes a standalone **`aria-code-mcp-*`** binary — the MCP server (see [Expose aria-code to other MCP clients](#expose-aria-code-to-other-mcp-clients) below) with no Python install needed:
+
+```bash
+claude mcp add aria-code -- /path/to/aria-code-mcp-macos-arm64   # or -linux-x64 / -windows-x64.exe
+```
+
 ### Option 1: Bootstrap (fresh Mac / Linux — recommended)
 
 No Node.js, Python, or Xcode required. One command handles everything:
@@ -804,6 +827,75 @@ Connect any [Model Context Protocol](https://modelcontextprotocol.io) server:
 /mcp tools     # All available MCP tools
 ```
 
+### Expose aria-code to other MCP clients
+
+aria-code can also run *as* an MCP server, so Claude Code, Codex, or any
+other MCP client can call into it directly:
+
+- `aria.market.quote` / `aria.agent.team` / `aria.artifacts.list` — live quote + technicals, multi-agent research, generated artifacts (all read-only)
+- `aria.skill.list` / `aria.skill.get` — discover and fetch the installed portable [SKILL.md](https://github.com/artherahq/aria-skills) workflows: reusable expert playbooks for UI design systems, anti-generic-AI-look design critique, trading-app patterns, equity research, backtest validation, and more. `aria.skill.list` takes an optional `query` to rank by relevance to your task; `aria.skill.get` returns the full workflow text for **you** to follow — skills are instruction documents, not executable tools, so nothing runs server-side. Each result carries an `integrity` field (`verified` for signed catalog skills, `unlocked` for local drop-ins) so you can see a skill's provenance before following it
+- `aria.report.generate` — fetch data and generate a full HTML research report artifact for a symbol
+- `aria.backtest.run` — run a historical strategy simulation (`buy_hold` or `sma_cross`) against real price history
+- `aria.broker.positions` / `aria.broker.list_previews` — read account/positions and recent order previews for a configured broker (US/CN/HK/UK — see `/broker`)
+- `aria.broker.preview_order` — build a risk-checked order preview. **Never places a live trade by itself** — execution requires either a human running `/trade confirm <preview_id>` in the aria-code terminal, or `aria.broker.confirm_order` below (off by default per broker)
+- `aria.broker.confirm_order` — **execute a previously built order preview, for real.** Off by default for every broker. Refuses unconditionally unless (1) a human already ran `/trade allow-chat-confirm <broker_id>` at the aria-code terminal keyboard — retyping the exact broker id, not y/yes — to opt that specific broker into chat-confirmed execution (this opt-in can never be triggered from MCP/chat, only from the terminal), **and** (2) the call itself passes `confirmed: true`. Executions via this path are tagged `source: "chat_mcp"` in the trade audit log (`~/.aria-code/trade_audit.jsonl`) to distinguish them from terminal-confirmed trades. Run `/trade disallow-chat-confirm <broker_id>` at any time to turn it back off
+- `aria.report.chart` — candlestick + MA20/50 chart PNG for a symbol
+- `aria.report.indicator_chart` — candlestick + volume + RSI(14) + MACD(12,26,9) multi-panel chart PNG for a symbol
+- `aria.report.comparison_chart` — normalized (base=100) % return comparison chart PNG across 2+ symbols
+- `aria.report.allocation_chart` — pie chart PNG of a broker account's position weights by market value
+- `aria.report.pdf` / `aria.report.docx` / `aria.report.pptx` — render a Markdown report to PDF, an editable Word document, or a slide deck (one slide per heading) — all chart/report tools save to your artifacts folder
+- `aria.report.canva_design` / `aria.report.canva_upload_asset` — fill a Canva brand template with data (and optionally an uploaded image asset) and export a design draft (needs `/canva connect <client_id> <client_secret>` once — register an app at [canva.com/developers](https://www.canva.com/developers/) and have a brand template ready first; Canva has no API for creating a design from scratch, only autofilling an existing template)
+- `aria.figma.read_file` / `aria.figma.comments` — read a Figma file's page/frame structure (depth-limited summary) and its comments (needs `/apikey set figma <personal_access_token>` once). Read-only — Figma has no public API for writing/creating designs from a script, only its in-app Plugin API can do that
+- `aria.report.estimate_image_cost` — get an illustrative per-image cost estimate (USD) for OpenAI's `gpt-image-1` at a given size/quality, before generating. Free.
+- `aria.report.generate_image` / `aria.report.edit_image` — generate a new image or transform an existing local photo via OpenAI's `gpt-image-1` (needs `/apikey set openai sk-...`). `edit_image` accepts an optional `mask_path` for inpainting — a mask PNG where transparent areas mark what to edit, opaque areas are preserved untouched. **Real per-call cost, billed the instant it succeeds** — both hard-refuse without `confirmed: true`, and calling `aria.report.estimate_image_cost` first is strongly recommended
+- `aria.report.generate_image_local` / `aria.report.edit_image_local` — the same, entirely locally via a self-hosted SDXL-Turbo (no API key, no per-call cost; needs the optional `image_gen` extra and ~4GB of weights on first use)
+- `aria.video.probe` / `aria.video.trim` / `aria.video.concat` / `aria.video.overlay_text` / `aria.video.overlay_audio` / `aria.video.convert` / `aria.video.change_speed` — deterministic local video editing via `ffmpeg` (no AI, no API key)
+- `aria.video.transcribe` / `aria.video.detect_scenes` — local AI-assisted analysis: speech transcription (`faster-whisper`, needs the optional `video_analysis` extra) and scene-cut detection (`opencv`, needs the optional `video` extra) — both produce editing *decisions*, not new pixels
+- `aria.video.generate_estimate` / `aria.video.generate_submit` / `aria.video.generate_status` — cloud AI text-to-video via Kling or Runway (`provider: "kling" | "runway"`). **Real per-request cost, billed the instant submission succeeds** — `generate_submit` hard-refuses without `confirmed: true`, and calling `generate_estimate` first is strongly recommended. Needs `/apikey set kling <access_key>:<secret_key>` or `/apikey set runway <key>`. No local/free option exists for this layer the way SDXL-Turbo covers images — see `video_editor.py`/`video_analysis.py` for the free local-only editing/analysis layers
+
+```bash
+python3 aria_mcp_server.py          # dev, from the repo root
+# or, once installed: aria-code-mcp
+```
+
+Register with Claude Code:
+
+```bash
+claude mcp add aria-code -- python3 /path/to/aria-code/aria_mcp_server.py
+```
+
+Register with Codex (`~/.codex/config.toml`):
+
+```toml
+[mcp_servers.aria_code]
+command = "python3"
+args = ["/path/to/aria-code/aria_mcp_server.py"]
+```
+
+#### Want deeper A-share quant tools too? Register a second server, don't wait for us to add them here
+
+This server intentionally does **not** re-expose Arthera's `quant_engine` quant
+tools (factor calculation, Bayesian-HMM regime detection, northbound-flow,
+Kelly position sizing, options pricing, pairs-trading stats, and a much
+deeper `run_backtest` than the one above — T+1 settlement, limit-up/down
+filtering, `sma_cross`/`rsi_mean_revert`/`momentum` strategies, CN_A or US
+market). That's a separate project (Arthera) with its own MCP server —
+`packages/quant_engine/mcp_server.py` — speaking the identical MCP
+2024-11-05 stdio JSON-RPC protocol, so any client here can register it
+as a second, independent server rather than this one proxying it:
+
+```bash
+claude mcp add arthera-quant -- python3 /path/to/Arthera/packages/quant_engine/mcp_server.py
+```
+
+We keep these two separate on purpose, not as a gap to close later: a few of
+`quant_engine`'s tools (`get_ai_signal`, `get_predictions`,
+`get_market_insights`) call Arthera's own internal Alibaba Cloud service and
+simply won't work without that access, so folding them into this server
+would mean shipping tools that error out for anyone without Arthera's
+private infrastructure. If you don't have access to that repo, everything
+above this section still works standalone.
+
 ---
 
 ## ⚙️ Configuration
@@ -888,7 +980,7 @@ converts to Apache 2.0 four years after release.
 
 <p align="center">
   <a href="https://arthera.finance">Website</a> ·
-  <a href="https://github.com/Cinsoul/Arthera">Full Platform</a> ·
+  <a href="https://github.com/artherahq/aria-code">Project Repository</a> ·
   <a href="https://github.com/artherahq/aria-code/issues">Issues</a> ·
   <a href="https://github.com/artherahq/aria-code/discussions">Discussions</a>
 </p>

aria_code-4.4.0/adk_apps/__init__.py

@@ -0,0 +1 @@
+"""ADK application packages shipped as optional Aria Code integrations."""

aria_code-4.4.0/adk_apps/aria_code_review/__init__.py

@@ -0,0 +1 @@
+"""Google ADK entrypoint package for Aria's read-only code reviewer."""

aria_code-4.4.0/adk_apps/aria_code_review/agent.py

@@ -0,0 +1,32 @@
+"""Google ADK entrypoint: a bounded reviewer for pasted code."""
+
+from __future__ import annotations
+
+import os
+
+from google.adk.agents.llm_agent import Agent
+
+from packages.adk_bridge import CodeReviewTools
+
+
+_tools = CodeReviewTools()
+
+root_agent = Agent(
+    model=os.getenv("ARIA_ADK_MODEL", "gemini-3.5-flash"),
+    name="aria_code_review",
+    description="A read-only code reviewer for user-submitted source text.",
+    instruction="""You are Aria Code Review, a careful software reviewer.
+
+Ask the user to paste the relevant source code when it has not been supplied.
+Use review_code for deterministic findings and preserve its severity, line,
+rule, and limitations. Never claim to have opened a local path, accessed a
+repository, run a command, changed a file, or executed tests: this ADK agent
+has none of those capabilities.
+
+Explain only findings supported by the submitted code. Separate deterministic
+findings from suggestions, and recommend a test or validation step where
+appropriate. Do not expose credentials appearing in source; tell the user to
+revoke and rotate any exposed credential instead.
+""",
+    tools=[_tools.review_code],
+)

aria_code-4.4.0/adk_apps/aria_research/__init__.py

@@ -0,0 +1 @@
+"""Read-only Aria market-research agent for Google ADK."""

aria_code-4.4.0/adk_apps/aria_research/agent.py

@@ -0,0 +1,31 @@
+"""Google ADK entrypoint: ``adk web adk_apps`` discovers ``aria_research``."""
+
+from __future__ import annotations
+
+import os
+
+from google.adk.agents.llm_agent import Agent
+
+from packages.adk_bridge import MarketResearchTools
+
+
+_tools = MarketResearchTools()
+
+root_agent = Agent(
+    model=os.getenv("ARIA_ADK_MODEL", "gemini-3.5-flash"),
+    name="aria_research",
+    description="A read-only market-research assistant for Aria Code and Arthera.",
+    instruction="""You are Aria Research, a careful financial-research assistant.
+
+Use get_market_snapshot for live market facts and explicitly identify its data
+quality, timestamp, warnings, and source limitations. Use get_market_data_health
+when a data result is missing, stale, or conflicting. Do not invent prices,
+returns, news, citations, positions, or tool output.
+
+This agent is research-only. Never claim to execute, recommend, place, modify,
+or schedule trades. Do not request credentials. State clearly that your output
+is research assistance, not investment advice. For quantitative conclusions,
+separate observed facts from hypotheses and mention material uncertainty.
+""",
+    tools=[_tools.get_market_snapshot, _tools.get_market_data_health],
+)

{aria_code-4.2.0 → aria_code-4.4.0}/agents/base.py

@@ -10,7 +10,7 @@ from __future__ import annotations
 import logging
 from abc import ABC, abstractmethod
 from dataclasses import dataclass, field
-from typing import Any, Callable, Dict, List, Optional
+from typing import Any, Callable, Dict, List, Optional, Generator, AsyncGenerator
 
 logger = logging.getLogger(__name__)
 
@@ -67,12 +67,15 @@ class BaseAgent(ABC):
     name:        str = "base"
     description: str = "基础 Agent"
 
-    # Post-training IPO facts injected into every agent's system prompt
-    _POST_TRAINING_FACTS = (
-        "\n\n## ⚡ Critical post-training facts (override training memory)\n"
-        "- **SPCX** = SpaceX (Space Exploration Technologies Corp.) — IPO'd June 12, 2026 "
-        "on NASDAQ. It is NOT an ETF. ⛔ Never say 'SPCX is a China ETF' or 'SPCX is private'.\n"
-        "- **2026 FIFA World Cup** opened June 11, 2026 in USA/Canada/Mexico.\n"
+    # Time-sensitive claims must come from a tool result with a timestamp, not
+    # from a hard-coded prompt.  This is appended to every agent because a
+    # wrong IPO/symbol/event claim can otherwise contaminate all specialisms.
+    _TIME_SENSITIVE_FACT_POLICY = (
+        "\n\n## Time-sensitive fact policy\n"
+        "- Do not treat examples, aliases, model memory, or internal notes as current facts.\n"
+        "- Verify ticker identity, IPO/listing status, prices, corporate events, sports results, "
+        "and dates with an appropriate tool or source before stating them as facts.\n"
+        "- If a current fact cannot be verified, say that it is unverified instead of guessing.\n"
     )
 
     # Language rule injected per-call based on detected user language
@@ -86,14 +89,21 @@ class BaseAgent(ABC):
         llm_provider=None,         # BaseLLMProvider 实例（可选，None 则用模板生成）
         data_router=None,          # DataRouter 实例（可选）
         on_token: Optional[Callable[[str], None]] = None,  # 流式 token 回调
+        on_thought: Optional[Callable[[str], None]] = None, # 思考过程回调
+        on_tool_start: Optional[Callable[[str, Dict], None]] = None, # 工具调用开始回调
+        on_tool_end: Optional[Callable[[str, Any], None]] = None, # 工具调用结束回调
         config: Optional[Dict] = None,
         lang: str = "zh",          # user language: "zh" | "en"
     ):
         self.llm      = llm_provider
         self.data     = data_router
         self.on_token = on_token
+        self.on_thought = on_thought
+        self.on_tool_start = on_tool_start
+        self.on_tool_end = on_tool_end
         self.config   = config or {}
         self.lang     = lang
+        self.memory   = []         # Short-term conversation history for multi-turn context
 
     async def fetch_data(self, symbol: str) -> Dict[str, Any]:
         """
@@ -127,43 +137,123 @@ class BaseAgent(ABC):
             )
         return ""
 
+    async def _execute_tool(self, tool_name: str, tool_args: Dict[str, Any]) -> str:
+        """
+        执行本地工具或 MCP 工具（可被子类重写扩展）。
+        实际项目中可将 aria-skills 暴露为工具。
+        """
+        if self.on_tool_start:
+            self.on_tool_start(tool_name, tool_args)
+
+        result = None
+        try:
+            # 路由到数据层或其他 MCP Client 执行
+            if self.data and hasattr(self.data, "execute_tool"):
+                result = await self.data.execute_tool(tool_name, tool_args)
+            else:
+                result = f"Error: Tool {tool_name} not implemented or data router unavailable."
+        except Exception as e:
+            result = f"Error executing {tool_name}: {e}"
+
+        if self.on_tool_end:
+            self.on_tool_end(tool_name, result)
+
+        import json
+        return json.dumps(result, ensure_ascii=False) if not isinstance(result, str) else result
+
     async def _call_llm(
         self,
         system: str,
         user: str,
         max_tokens: int = 800,
         quote: Optional[Dict[str, Any]] = None,
+        max_tool_loops: int = 5,
     ) -> str:
-        """调用 LLM 生成分析文本（无 LLM 时返回空字符串）"""
+        """调用 LLM 生成分析文本，内置 Tool Calling 的执行循环"""
         if not self.llm:
             return ""
-        # Inject language rule + post-training facts + data guard into system prompt
+        # Inject language rule + time-sensitive fact policy + data guard.
         _lang_rule = self._LANG_RULES.get(self.lang, self._LANG_RULES["zh"])
         _data_warn = self._data_guard(quote or {})
-        system = system + self._POST_TRAINING_FACTS + _lang_rule + _data_warn
+        system = system + self._TIME_SENSITIVE_FACT_POLICY + _lang_rule + _data_warn
         from providers.llm.base import Message
-        messages = [
-            Message(role="system", content=system),
-            Message(role="user",   content=user),
-        ]
+
+        # 自动注入 DAG pipeline 中的 upstream context
+        if hasattr(self, "_current_data") and self._current_data:
+            upstream = self._current_data.get("upstream_context")
+            if upstream:
+                user += "\n\n## ⬆️ 团队/上游 Agent 提供的背景上下文 (Upstream Insights)\n"
+                user += "请在分析中参考以下来自其他专家 Agent 的核心观点：\n"
+                user += "\n".join(upstream)
+
+        # 融入历史记忆
+        if not self.memory:
+            messages = [
+                Message(role="system", content=system),
+                Message(role="user",   content=user),
+            ]
+        else:
+            messages = list(self.memory)
+            messages.append(Message(role="user", content=user))
+
         full_text = ""
-        try:
-            async for event in self.llm.stream(
-                messages, max_tokens=max_tokens
-            ):
-                t = event.get("type")
-                if t == "token":
-                    tok = event.get("text", "")
-                    full_text += tok
-                    if self.on_token:
-                        self.on_token(tok)
-                elif t == "error":
-                    logger.warning(f"[{self.name}] LLM 错误: {event.get('message')}")
-                    break
-        except Exception as e:
-            logger.warning(f"[{self.name}] LLM 调用失败: {e}")
+        loop_count = 0
+
+        while loop_count < max_tool_loops:
+            loop_count += 1
+            current_text = ""
+            tool_calls = []
+
+            try:
+                async for event in self.llm.stream(
+                    messages, max_tokens=max_tokens
+                ):
+                    t = event.get("type")
+                    if t == "token":
+                        tok = event.get("text", "")
+                        current_text += tok
+                        full_text += tok
+                        if self.on_token:
+                            self.on_token(tok)
+                    elif t == "thought" and self.on_thought:
+                        self.on_thought(event.get("text", ""))
+                    elif t == "tool_call":
+                        # 记录工具调用意图
+                        tool_calls.append({
+                            "name": event.get("name"),
+                            "args": event.get("args", {})
+                        })
+                    elif t == "tool_start" and self.on_tool_start:
+                        self.on_tool_start(event.get("tool_name", ""), event.get("tool_args", {}))
+                    elif t == "tool_end" and self.on_tool_end:
+                        self.on_tool_end(event.get("tool_name", ""), event.get("tool_result", None))
+                    elif t == "error":
+                        logger.warning(f"[{self.name}] LLM 错误: {event.get('message')}")
+                        break
+            except Exception as e:
+                logger.warning(f"[{self.name}] LLM 调用失败: {e}")
+                break
+
+            # 将当前 Assistant 的回复加入消息流
+            messages.append(Message(role="assistant", content=current_text))
+
+            # 如果没有工具调用，说明 LLM 完成了最终回答，跳出循环
+            if not tool_calls:
+                break
+
+            # 有工具调用，则逐一执行，并把结果喂回 LLM
+            for tc in tool_calls:
+                tool_res = await self._execute_tool(tc["name"], tc["args"])
+                # 注意：实际 provider 中 Message 结构可能需要 tool_call_id
+                # 这里假设通用 Message 能够携带 role="tool"
+                messages.append(Message(role="tool", name=tc["name"], content=tool_res))
+
+        # 更新短期记忆
+        self.memory = messages
+
         return full_text.strip()
 
+
     @abstractmethod
     async def analyze(self, symbol: str, data: Dict[str, Any]) -> AgentResult:
         """