diff --git a/README.md b/README.md index 325c0a27..94abc257 100644 --- a/README.md +++ b/README.md @@ -107,6 +107,8 @@ When you open a **Coder workspace**, startup runs automatically in the backgroun **Your next step:** run [`00_environment_check.ipynb`](implementations/getting_started/00_environment_check.ipynb) top to bottom. That notebook will confirm that startup succeeded. +**ADK web UI.** `adk web` binds to `localhost` *inside* the workspace, so a browser on your laptop cannot reach it until you tunnel. [Guide 5](guides/05-access-adk-web-via-ssh-tunnel.md) walks through that on macOS, Windows, and Linux. Skip it if you are running the repo locally. + On first boot, keys are verified against live services and your onboarding status is recorded. Workspace restarts reload keys without re-running the full test suite. **Local machine or troubleshooting** — fetch and verify keys manually: @@ -203,7 +205,7 @@ uv run pre-commit run --all-files ## Documentation - Per-implementation READMEs under [`implementations/`](implementations/) — the primary user surface. -- [`guides/`](guides/) — self-contained, step-by-step strategy guides for the most common build-phase tasks: onboarding a dataset, creating an experiment, customizing an agent's strategy, and auditing a result before you believe it. +- [`guides/`](guides/) — self-contained, step-by-step strategy guides for the most common build-phase tasks: onboarding a dataset, creating an experiment, customizing an agent's strategy, and auditing a result before you believe it. [Guide 5](guides/05-access-adk-web-via-ssh-tunnel.md) is the Coder companion for reaching `adk web` from your laptop. - [Architecture atlas](https://vectorinstitute.github.io/agentic-forecasting/architecture-atlas.html) ([source](docs/architecture-atlas.html)) — a self-contained visual atlas of the system architecture: the loop, the temporal fence, predictor families, the harness, agent anatomy, and how each reference implementation instantiates them. - [`aieng-forecasting/README.md`](aieng-forecasting/README.md) and [`aieng-forecasting/aieng/forecasting/methods/README.md`](aieng-forecasting/aieng/forecasting/methods/README.md) — the library and the method catalog. - [`planning-docs/roadmap.md`](planning-docs/roadmap.md) — architecture principles and extension ideas. diff --git a/docs/architecture-atlas.html b/docs/architecture-atlas.html index 683d1944..5f693506 100644 --- a/docs/architecture-atlas.html +++ b/docs/architecture-atlas.html @@ -349,7 +349,7 @@

Agentic Forecasting Atlas

  • 04The harness — backtest to develop, evaluate to commit
  • 05The agent — anatomy of an analyst
  • 06The implementations — five instantiations, one architecture
    • -
  • 07Build your own — two paths, four guides
    • +
  • 07Build your own — two paths, then the worksheets
    • @@ -850,6 +850,10 @@

    Where do you land?

    Want covariates / report-grounded prompting / judged categorical sp500 / food_price / boc README respectively, then Path A on that starter (toggles; guide 3's levers are the same AgentConfig fields). + + On Coder and want adk web in a local browser + guide 5 — tunnel localhost:8000 from the workspace to your laptop. Skip it if you are running locally. + diff --git a/guides/05-access-adk-web-via-ssh-tunnel.md b/guides/05-access-adk-web-via-ssh-tunnel.md new file mode 100644 index 00000000..dbaaf3e8 --- /dev/null +++ b/guides/05-access-adk-web-via-ssh-tunnel.md @@ -0,0 +1,158 @@ +# Guide 5 — Accessing the ADK Web UI via SSH tunneling + +**By the end of this guide** you will have your local browser talking to an ADK agent that is running inside your Coder workspace. The same steps work on **macOS, Windows, and Linux** — where a step differs, look for the OS callout. + +This is a Coder-environment companion, not a step on Path A or Path B. Skip it if you are running the repo on your own machine: `adk web` already binds to `localhost` there, and you can open the UI directly. + +**Prerequisites:** a Coder workspace with the repo ready (`uv sync` already done — the bootcamp image does this for you). You will also install the Coder CLI on your **personal computer**, not inside the workspace. + +--- + +## Prerequisites (inside your Coder workspace) + +1. Open your Coder workspace terminal. + +2. Start an ADK agent on port `8000`. From the **repository root**, for example: + +```bash +uv run adk web implementations/energy_oil_forecasting/ +``` + + Other common targets, also from the repo root: + +```bash +uv run adk web implementations/getting_started/concierge_agent +uv run adk web implementations/energy_oil_forecasting/starter_agent +``` + + From `implementations/energy_oil_forecasting/` the adaptive-agent notebooks use the shorter `uv run adk web adaptive_agent/`. Any of these is fine — the tunnel does not care which agent is serving. + +3. Keep this workspace terminal tab open and running. + +--- + +## Step 1 — Install the Coder CLI locally + +Install the Coder CLI on your **personal computer** (not inside the VM). Follow the official instructions, which cover macOS, Windows, and Linux: [https://coder.com/docs/install/cli](https://coder.com/docs/install/cli) + +Verify the install in a **fresh** terminal: + +```bash +coder version +``` + +--- + +## Step 2 — Authenticate and configure SSH + +Open a **new terminal** on your local computer. + +> **Windows:** Use **PowerShell** (or Windows Terminal), not the legacy `cmd` prompt. All commands below work unchanged in PowerShell. + +1. Log in to the platform: + +```bash +coder login https://platform.vectorinstitute.ai/ +``` + + Follow the prompt to paste your **access token** from the browser. The token is per-user and is shown on the login page that opens. + +2. Configure your local SSH settings: + +```bash +coder config-ssh +``` + + This links your local SSH client to Coder by adding host entries to your SSH config: + + - macOS / Linux: `~/.ssh/config` + - Windows: `%USERPROFILE%\.ssh\config` + + The dashboard shows a workspace name such as `ethan-fc-dev`. After `coder config-ssh`, the matching SSH host is `coder.ethan-fc-dev`. Option A below uses the SSH host (`coder.`); Option B uses the dashboard name (``, no `coder.` prefix). + +--- + +## Step 3 — Create the SSH tunnel + +You have two options. Option A is a raw `ssh` port-forward; Option B is simpler and identical across platforms. + +### Requirement: an SSH client + +- **macOS / Linux:** OpenSSH is preinstalled — nothing to do. +- **Windows:** Windows 10/11 ship an OpenSSH client, but it is occasionally not enabled. If `ssh` is "not recognized," enable it via **Settings → Apps → Optional Features → Add a feature → OpenSSH Client**, or use the `ssh` bundled with [Git for Windows](https://git-scm.com/download/win). Option B does not need a working `ssh` on your `PATH`. + +### Option A — Manual port forward with `ssh` + +Run in your local terminal, substituting your workspace name (the dashboard name, with the `coder.` prefix): + +```bash +ssh -L 8000:localhost:8000 coder. -N +``` + +Example: if the dashboard shows `ethan-fc-dev`, the host is `coder.ethan-fc-dev`. + +- The terminal will look **frozen or paused** — that means the tunnel is active. This is the same on PowerShell, macOS Terminal, and Linux. +- **Do not close this terminal window.** + +### Option B — Let Coder do it (recommended) + +Skip writing the `ssh -L` command by hand. Use the dashboard workspace name (no `coder.` prefix): + +```bash +coder port-forward --tcp 8000:8000 +``` + +This behaves identically on macOS, Windows, and Linux, and does not depend on your local SSH client. Leave it running. + +--- + +## Step 4 — Access the UI + +1. Open your local web browser. +2. Navigate to [http://localhost:8000](http://localhost:8000). +3. You can now fully interact with the ADK web interface. + +--- + +## Troubleshooting + +**"Address already in use" / port 8000 is taken.** Something else on your machine is using `8000`. Remap the *local* side of the tunnel to a free port, then browse to that port: + +```bash +# Option A +ssh -L 8080:localhost:8000 coder. -N +# Option B +coder port-forward --tcp 8080:8000 +``` + +Then open [http://localhost:8080](http://localhost:8080). The format is `LOCAL_PORT:localhost:REMOTE_PORT` (Option A) or `LOCAL_PORT:REMOTE_PORT` (Option B) — only change the first number. The agent in the workspace stays on `8000`. + +**`ssh: command not found` (Windows).** Enable the OpenSSH Client optional feature or use Git Bash (see Step 3), or switch to Option B. + +**`coder: command not found` after install.** Open a new terminal so the updated PATH is picked up, then re-run `coder version`. + +**Can't find your workspace name.** It's listed in the Coder dashboard, and `coder config-ssh` writes matching `Host coder.` entries into your SSH config (`~/.ssh/config` or `%USERPROFILE%\.ssh\config`). + +**Firewall prompt on first connect (Windows/macOS).** Allow access for the SSH/Coder client when prompted — it only opens a local loopback tunnel. + +**Page won't load / connection refused in the browser.** Confirm the ADK agent is still running in the workspace terminal (Prerequisites, step 2) and that the tunnel terminal is still open and "frozen." + +--- + +## Quick reference + +| Task | macOS / Linux | Windows | +| --- | --- | --- | +| Install CLI | See [official instructions](https://coder.com/docs/install/cli) | See [official instructions](https://coder.com/docs/install/cli) | +| Terminal to use | Terminal / any shell | PowerShell / Windows Terminal | +| SSH client | Preinstalled | OpenSSH Client feature or Git Bash | +| SSH config path | `~/.ssh/config` | `%USERPROFILE%\.ssh\config` | +| Tunnel (manual) | `ssh -L 8000:localhost:8000 coder. -N` | same | +| Tunnel (simple) | `coder port-forward --tcp 8000:8000` | same | +| Access URL | [http://localhost:8000](http://localhost:8000) | same | + +--- + +## Where to go next + +If you tunneled the energy starter or adaptive agent, **[guide 3](03-customize-agent-strategy.md)** is the lever map for changing how it thinks. The [repo concierge](../implementations/getting_started/99_repo_concierge.ipynb) is the other common `adk web` target — useful when you want to ask how the repo is put together rather than drive a forecaster. diff --git a/guides/README.md b/guides/README.md index 2bc7a19d..f864d3d6 100644 --- a/guides/README.md +++ b/guides/README.md @@ -1,8 +1,8 @@ # Bootcamp strategy guides -Step-by-step worksheets for the tasks you are most likely to take on during the build phase. Each one is self-contained: it states what you will have at the end, walks there in numbered steps with runnable code, and ends with a verification check and extension ideas. Every code snippet in these guides has been run against the repo as written. +Step-by-step worksheets for the tasks you are most likely to take on during the build phase. Each one is self-contained: it states what you will have at the end, walks there in numbered steps with runnable code, and ends with a verification check and extension ideas. Every code snippet in guides 1–4 has been run against the repo as written; guide 5 is Coder CLI / SSH setup on your laptop. -These are **guides, not reference implementations**. The reference implementations under [`implementations/`](../implementations/) show finished forecasting systems; these guides show you the *moves* — how to bring in data, stand up an experiment, and reshape an agent — using only machinery that already exists in the repo. +These are **guides, not reference implementations**. The reference implementations under [`implementations/`](../implementations/) show finished forecasting systems; these guides show you the *moves* — how to bring in data, stand up an experiment, and reshape an agent — using only machinery that already exists in the repo. [Guide 5](05-access-adk-web-via-ssh-tunnel.md) is the exception: it is Coder-environment setup (reaching `adk web` from your laptop), not a forecasting move. **Start with the [architecture atlas](https://vectorinstitute.github.io/agentic-forecasting/architecture-atlas.html)** — it's the map. Read it first if you're lost about how the pieces fit; then come back and pick a worksheet by path, not by number. @@ -18,7 +18,10 @@ These are **guides, not reference implementations**. The reference implementatio | 2 | [Creating a new experiment](02-create-an-experiment.md) | A backtest + evaluation setup on that series: a YAML spec, a predictor lineup in code, and a scored leaderboard | | 3 | [Customizing an analyst agent's strategy](03-customize-agent-strategy.md) | A map of every lever that changes how an agent behaves — persona, toolbelt, skills, search strategy — with a worked change for each | | 4 | [Auditing a result before you believe it](04-audit-your-results.md) | A finished backtest interrogated at four altitudes — inputs, model behaviour, score decomposition, noise floor — ending in a claim you can defend in a writeup | +| 5 | [Accessing the ADK Web UI via SSH tunneling](05-access-adk-web-via-ssh-tunnel.md) | Your local browser talking to `adk web` running inside a Coder workspace (macOS, Windows, Linux) | -**Prerequisites.** A working environment (`uv sync --dev` from the repo root — see the [main README](../README.md#setup)). Guides 1, 2, and 4 need nothing else; all three onboard, score, and audit against a small sample dataset (committed at [`assets/harbourview_lumber_spot.csv`](assets/harbourview_lumber_spot.csv)), entirely offline. Guide 3 exercises the agent stack, so run [`00_environment_check.ipynb`](../implementations/getting_started/00_environment_check.ipynb) first if you haven't. +**Coder / `adk web`.** Guides 1–4 assume you can already reach the repo. If you are on a Coder workspace and want the ADK browser UI on your laptop, that is [guide 5](05-access-adk-web-via-ssh-tunnel.md) — skip it on a local machine. + +**Prerequisites.** A working environment (`uv sync --dev` from the repo root — see the [main README](../README.md#setup)). Guides 1, 2, and 4 need nothing else; all three onboard, score, and audit against a small sample dataset (committed at [`assets/harbourview_lumber_spot.csv`](assets/harbourview_lumber_spot.csv)), entirely offline. Guide 3 exercises the agent stack, so run [`00_environment_check.ipynb`](../implementations/getting_started/00_environment_check.ipynb) first if you haven't. Guide 5 is Coder-only: it needs the [Coder CLI](https://coder.com/docs/install/cli) on your personal computer, not inside the workspace. **Conventions.** Guide 3 anchors on the [energy / WTI implementation](../implementations/energy_oil_forecasting/) — the most complete agent stack in the repo, built as a ToolSpec toolbelt — and guide 4 borrows its analysis helpers and committed artifacts. The sp500/food/BoC starters aren't built the same way: they expose `enable_search`/`enable_code_exec` toggles on the same underlying `AgentConfig` fields instead of a toolbelt. [Guide 3's opening section](03-customize-agent-strategy.md#which-starter-are-you-on) maps between the two, so guide 3's other patterns still transfer to whichever starter you're on. Paths in code are shown relative to the repo root; run snippets from the repo root unless a step says otherwise. diff --git a/implementations/README.md b/implementations/README.md index 7f861a2e..4dcc13f3 100644 --- a/implementations/README.md +++ b/implementations/README.md @@ -28,7 +28,7 @@ implementations/ `-- pyproject.toml # local workspace packaging ``` -YAML backtest and eval specs live under each use case in `specs/`. Each directory is independent; see its `README.md` for the walkthrough. For the build-phase moves — onboarding data, standing up an experiment, customizing an agent, auditing a result — see [`guides/`](../guides/). +YAML backtest and eval specs live under each use case in `specs/`. Each directory is independent; see its `README.md` for the walkthrough. For the build-phase moves — onboarding data, standing up an experiment, customizing an agent, auditing a result — see [`guides/`](../guides/). On a Coder workspace, [`guides/05-access-adk-web-via-ssh-tunnel.md`](../guides/05-access-adk-web-via-ssh-tunnel.md) is how you open `adk web` in a browser on your laptop. Every domain use case (all except `getting_started`) also ships a `starter_agent/` module and a `99_starter_agent.ipynb` — a fresh, hackable **starter agent** that is the consistent "build your own" entry point for that use case (toggleable news search + code execution, two lightweight tool-usage skills, an interactive cell, and one scored forecast). diff --git a/implementations/energy_oil_forecasting/05_adaptive_agent_training.ipynb b/implementations/energy_oil_forecasting/05_adaptive_agent_training.ipynb index bd1a08e2..a94e599e 100644 --- a/implementations/energy_oil_forecasting/05_adaptive_agent_training.ipynb +++ b/implementations/energy_oil_forecasting/05_adaptive_agent_training.ipynb @@ -460,7 +460,9 @@ "```\n", "\n", "Open `http://localhost:8000` in your browser. The agent has its full skill \n", - "set available: code execution, web search, and mutation tools.\n", + "set available: code execution, web search, and mutation tools. On a **Coder \n", + "workspace** that URL is inside the VM — [guide 5](../../guides/05-access-adk-web-via-ssh-tunnel.md) \n", + "tunnels it to your laptop.\n", "\n", "**Suggested conversation starters:**\n", "\n", diff --git a/implementations/energy_oil_forecasting/06_protected_eval.ipynb b/implementations/energy_oil_forecasting/06_protected_eval.ipynb index f39acaad..880d776f 100644 --- a/implementations/energy_oil_forecasting/06_protected_eval.ipynb +++ b/implementations/energy_oil_forecasting/06_protected_eval.ipynb @@ -538,7 +538,7 @@ " uv run adk web adaptive_agent/\n", "```\n", "\n", - "Open `http://localhost:8000`. See Notebook 5 for suggested conversation starters." + "Open `http://localhost:8000`. See Notebook 5 for suggested conversation starters. On a **Coder workspace** that URL is inside the VM — [guide 5](../../guides/05-access-adk-web-via-ssh-tunnel.md) tunnels it to your laptop." ], "id": "be42fd1a" } diff --git a/implementations/energy_oil_forecasting/analyst_agent/agent.py b/implementations/energy_oil_forecasting/analyst_agent/agent.py index 42055c55..4956b8e8 100644 --- a/implementations/energy_oil_forecasting/analyst_agent/agent.py +++ b/implementations/energy_oil_forecasting/analyst_agent/agent.py @@ -534,9 +534,7 @@ def build_wti_code_exec_config( return AgentConfig( name="wti_analyst_code", model=model, - instruction=( - _WTI_ANALYST_INSTRUCTION + _CONTEXT_RETRIEVAL_SUPPLEMENT + _CODE_EXEC_SKILLS_SUPPLEMENT - ), + instruction=(_WTI_ANALYST_INSTRUCTION + _CONTEXT_RETRIEVAL_SUPPLEMENT + _CODE_EXEC_SKILLS_SUPPLEMENT), max_output_tokens=max_output_tokens, context_retrieval=ContextRetrievalConfig( enabled=True, @@ -611,9 +609,7 @@ def build_wti_tool_config( return AgentConfig( name="wti_analyst_tool", model=model, - instruction=( - _WTI_ANALYST_INSTRUCTION + _CONTEXT_RETRIEVAL_SUPPLEMENT + _FORECAST_TOOL_SUPPLEMENT - ), + instruction=(_WTI_ANALYST_INSTRUCTION + _CONTEXT_RETRIEVAL_SUPPLEMENT + _FORECAST_TOOL_SUPPLEMENT), context_retrieval=ContextRetrievalConfig( enabled=True, instruction=_WTI_CONTEXT_RETRIEVAL_INSTRUCTION, diff --git a/implementations/getting_started/99_repo_concierge.ipynb b/implementations/getting_started/99_repo_concierge.ipynb index 8691ae4f..b3aa0527 100644 --- a/implementations/getting_started/99_repo_concierge.ipynb +++ b/implementations/getting_started/99_repo_concierge.ipynb @@ -206,7 +206,9 @@ "\n", "**Alternative:** `uv run adk web implementations/getting_started/concierge_agent`\n", "opens a browser UI (same agent). From `implementations/getting_started/`, you can\n", - "also use the shorter `uv run adk run concierge_agent`.\n", + "also use the shorter `uv run adk run concierge_agent`. On a **Coder workspace**,\n", + "that UI is inside the VM — [guide 5](../../guides/05-access-adk-web-via-ssh-tunnel.md)\n", + "tunnels it to your laptop.\n", "\n", "---\n", "\n", diff --git a/implementations/getting_started/README.md b/implementations/getting_started/README.md index 99e56389..0ab31572 100644 --- a/implementations/getting_started/README.md +++ b/implementations/getting_started/README.md @@ -154,7 +154,9 @@ and modules, and can quote snippets from the committed public-`main` catalog. ``` (`uv run adk web implementations/getting_started/concierge_agent` opens the same - agent in a browser.) + agent in a browser. On a **Coder workspace**, that UI is inside the VM — + [guide 5](../../guides/05-access-adk-web-via-ssh-tunnel.md) tunnels it to your + laptop.) From `implementations/getting_started/`, the shorter `uv run adk run concierge_agent` works too. diff --git a/pyproject.toml b/pyproject.toml index 50e9cbe6..509dbbcc 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -118,8 +118,9 @@ ignore = [ "scripts/**" = ["D1", "E402", "W505", "ERA001"] "implementations/**" = ["D1", "E402", "W505", "ERA001"] # Playground: throwaway exploration/diagnostics. Same relaxations as scripts, -# plus PLC0415 (optional deps are imported inside try/except by design). -"playground/**" = ["D1", "E402", "W505", "ERA001", "PLC0415"] +# plus PLC0415 (optional deps are imported inside try/except by design) and +# PLR0917 (archived notebooks predate the positional-arg cap). +"playground/**" = ["D1", "E402", "W505", "ERA001", "PLC0415", "PLR0917"] # Repo-root integration tests: optional extras imported inside try/except. "tests/**" = ["D1", "E402", "W505", "ERA001", "PLC0415"]