Tailor agent skills to Spectrum 3847; disable AdvantageKit skills

.agents/skills/advantagekit-replay/scripts/pull_robot_wpilogs.py

@@ -32,7 +32,7 @@ def newest_remote_log(user: str, host: str, directory: str) -> str:
 def main() -> None:
     """Main."""
     parser = argparse.ArgumentParser()
-    parser.add_argument("--host", default="10.80.44.2", help="Robot IP")
+    parser.add_argument("--host", default="10.38.47.2", help="Robot IP")
     parser.add_argument("--user", default="admin", help="Robot SSH user")
     parser.add_argument("--remote-dir", default="/U", help="Robot log directory")
     parser.add_argument("--out", default="RobotLogs", help="Local output directory")

.agents/skills/advantagekit/references/common-issues.md

@@ -6,15 +6,15 @@ Replay works by re-running identical code with identical logged inputs. Any data
 
 ### Problem Sources
 
-| Source | Problem | Fix |
-| ------ | ------- | --- |
-| `Timer.getFPGATimestamp()` | Raw FPGA time — not deterministic | Use `Timer.getTimestamp()` |
-| NetworkTables / dashboard inputs | Values change outside the logging cycle | Log as inputs through IO layer |
-| YAGSL, Phoenix 6 swerve libraries | Bypass IO abstraction, call hardware directly | Use AKit's swerve template instead |
-| `Math.random()` / random number generators | Non-deterministic by nature | Log seed or generated values as inputs |
-| Iterating over `HashMap` / unordered collections | Iteration order varies across JVM runs | Use `LinkedHashMap` or sorted collections |
-| Filesystem reads | File contents may change | Log file data as inputs |
-| Driver Station data accessed before `Logger.start()` | Not yet deterministic | Defer all DS access until after `Logger.start()` |
+|                        Source                        |                    Problem                    |                       Fix                        |
+|------------------------------------------------------|-----------------------------------------------|--------------------------------------------------|
+| `Timer.getFPGATimestamp()`                           | Raw FPGA time — not deterministic             | Use `Timer.getTimestamp()`                       |
+| NetworkTables / dashboard inputs                     | Values change outside the logging cycle       | Log as inputs through IO layer                   |
+| YAGSL, Phoenix 6 swerve libraries                    | Bypass IO abstraction, call hardware directly | Use AKit's swerve template instead               |
+| `Math.random()` / random number generators           | Non-deterministic by nature                   | Log seed or generated values as inputs           |
+| Iterating over `HashMap` / unordered collections     | Iteration order varies across JVM runs        | Use `LinkedHashMap` or sorted collections        |
+| Filesystem reads                                     | File contents may change                      | Log file data as inputs                          |
+| Driver Station data accessed before `Logger.start()` | Not yet deterministic                         | Defer all DS access until after `Logger.start()` |
 
 ---
 

.agents/skills/advantagekit/references/output-logging.md

@@ -50,16 +50,16 @@ This guarantees the same logic that ran on the robot is used for replay, so odom
 
 All types support single values, 1D arrays, and 2D arrays unless noted.
 
-| Category | Types | Notes |
-| ---------- | ------- | ------- |
-| Primitives | `boolean`, `int`, `long`, `float`, `double`, `String` | — |
-| WPILib structs | `Translation2d/3d`, `Pose2d/3d`, `Rotation2d/3d`, `SwerveModuleState`, etc. | Preferred over protobuf — no delay |
-| Protobuf | Any WPILib protobuf type | **⚠️ First log can take >100ms** — log once while disabled |
-| Records | Custom `record` classes (fields: primitives, enums, structs, nested records) | **⚠️ Same first-log delay as protobuf** — log while disabled; no array fields |
-| Enums | Any enum | Logged as `name()` string |
-| Colors | WPILib `Color` | Logged as hex triplet string |
-| Mechanisms | `LoggedMechanism2d` | Output only |
-| Suppliers | `BooleanSupplier`, `IntSupplier`, `LongSupplier`, `DoubleSupplier` | Output only |
+|    Category    |                                    Types                                     |                                     Notes                                     |
+|----------------|------------------------------------------------------------------------------|-------------------------------------------------------------------------------|
+| Primitives     | `boolean`, `int`, `long`, `float`, `double`, `String`                        | —                                                                             |
+| WPILib structs | `Translation2d/3d`, `Pose2d/3d`, `Rotation2d/3d`, `SwerveModuleState`, etc.  | Preferred over protobuf — no delay                                            |
+| Protobuf       | Any WPILib protobuf type                                                     | **⚠️ First log can take >100ms** — log once while disabled                    |
+| Records        | Custom `record` classes (fields: primitives, enums, structs, nested records) | **⚠️ Same first-log delay as protobuf** — log while disabled; no array fields |
+| Enums          | Any enum                                                                     | Logged as `name()` string                                                     |
+| Colors         | WPILib `Color`                                                               | Logged as hex triplet string                                                  |
+| Mechanisms     | `LoggedMechanism2d`                                                          | Output only                                                                   |
+| Suppliers      | `BooleanSupplier`, `IntSupplier`, `LongSupplier`, `DoubleSupplier`           | Output only                                                                   |
 
 **Protobuf / Record first-use pattern** — call once during `disabledInit()` to absorb the delay:
 

.agents/skills/advantagekit/references/recording-inputs.md

@@ -45,12 +45,14 @@ All [supported types](https://docs.advantagekit.org/data-flow/supported-types) a
 Two approaches — pick one consistently:
 
 **Naming convention** (simpler):
+
 ```java
 public double velocityRadPerSec = 0.0;
 public double positionMeters = 0.0;
 ```
 
 **Measure objects** (type-safe):
+
 ```java
 public Distance position = Meters.of(0.0);
 public LinearVelocity velocity = MetersPerSecond.of(0.0);
@@ -88,12 +90,12 @@ Direct NetworkTables access (`SmartDashboard.getNumber()`, `SmartDashboard.getSt
 
 ### Logged Replacements
 
-| Instead of | Use |
-|-----------|-----|
-| `SendableChooser<T>` | `LoggedDashboardChooser<T>` |
-| `SmartDashboard.getNumber()` | `LoggedNetworkNumber` |
-| `SmartDashboard.getString()` | `LoggedNetworkString` |
-| `SmartDashboard.getBoolean()` | `LoggedNetworkBoolean` |
+|          Instead of           |             Use             |
+|-------------------------------|-----------------------------|
+| `SendableChooser<T>`          | `LoggedDashboardChooser<T>` |
+| `SmartDashboard.getNumber()`  | `LoggedNetworkNumber`       |
+| `SmartDashboard.getString()`  | `LoggedNetworkString`       |
+| `SmartDashboard.getBoolean()` | `LoggedNetworkBoolean`      |
 
 ### Auto Chooser Example
 

.agents/skills/advantagescope/SKILL.md

@@ -1,37 +1,32 @@
 ---
+
 name: advantagescope
-description: "Use for AdvantageScope visualization automation: building/running the Team 8044 AdvantageScope fork, opening WPILOGs with layouts, using opt-in agent control/export hooks, capturing 2D/3D field screenshots or frame sequences, and falling back to vanilla AdvantageScope/manual inspection when hooks are unavailable."
+description: "Use for AdvantageScope visualization automation: building/running our AdvantageScope fork (Spectrum 3847) or vanilla AdvantageScope, opening WPILOGs/DataLogs with layouts, using opt-in agent control/export hooks, capturing 2D/3D field screenshots or frame sequences, and falling back to vanilla AdvantageScope/manual inspection when hooks are unavailable."
 metadata:
-  short-description: Automate AdvantageScope visualization
----
+short-description: Automate AdvantageScope visualization
+--------------------------------------------------------
 
 # AdvantageScope
 
 ## Core Rules
 
-- This skill is the robot-repo consumer workflow for AdvantageScope visualization. It should help agents discover robot logs/assets, generate layouts, locate the Team 8044 fork, and export previews.
-- Keep fork maintenance details in the Team 8044 AdvantageScope repo's `$advantagescope-fork` skill. Use that skill when modifying, rebasing, packaging, or debugging fork internals.
+- This skill is the robot-repo consumer workflow for AdvantageScope visualization. It should help agents discover robot logs/assets, generate layouts, locate our AdvantageScope fork or vanilla AdvantageScope, and export previews.
+- Keep fork maintenance details in the AdvantageScope repo's `$advantagescope-fork` skill (we maintain our own fork). Use that skill when modifying, rebasing, packaging, or debugging fork internals.
 - Do not hard-code season-specific field ids, robot model names, autos, or layouts. Discover custom assets from `AScope_Assets` and use explicit task arguments.
 - Do not automate by screen clicks or fixed coordinates. Use local agent control/export hooks when available.
 - If only vanilla AdvantageScope is available, open the log/layout manually and use WPILOG parsing for objective results.
 - When a new repeatable AdvantageScope workflow is discovered, treat updating this skill as part of the work. Add the durable instructions and, when the workflow is more than a couple commands, add or improve a generic helper script so future agents can use the app immediately without rediscovering the path.
 
 ## Build And Runtime
 
-- Use the Team 8044 fork for agent exports:
-  ```sh
-  https://github.com/Team8044/AdvantageScope
-  ```
-- When changing the fork itself, switch to the fork checkout and use its repo-local `$advantagescope-fork` skill.
-- Prefer an explicit `--fork <checkout>` argument or `ADVANTAGESCOPE_FORK=<checkout>` when a checkout is already known.
-- If no checkout is known, discover local git repos whose remotes include `Team8044/AdvantageScope`; common search roots include the current repo's parents and `~/Documents/GitHub`.
-- If the fork is not available locally, ask the user before cloning/building it because that requires network, disk, and dependency changes.
+- Prefer our AdvantageScope fork for agent exports. If the fork is not checked out locally, ask the user before cloning/building it because that requires network, disk, and dependency changes; otherwise fall back to vanilla AdvantageScope plus WPILOG summaries.
 - `export_preview.py` can also use `ADVANTAGESCOPE_EXECUTABLE` or `--advantagescope <executable>` when a packaged/built fork executable is already available.
 - Do not copy robot assets into the fork. Custom robot assets should stay in this robot repo's `AScope_Assets` folder and be loaded through AdvantageScope's user assets feature.
 
 ## Agent Hooks
 
 - Before exporting an auto preview, get the WPILOG from `$wpilib-sim`'s permanent workflow:
+
   ```sh
   python3 .agents/skills/wpilib-sim/scripts/run_auto_sim.py \
     --repo . \
@@ -40,7 +35,9 @@ metadata:
     --duration <auto-duration> \
     --buffer 1
   ```
+
   Then compute the auto window from logged DriverStation state:
+
   ```sh
   python3 .agents/skills/wpilib-sim/scripts/find_wpilog_window.py \
     --repo . \
@@ -50,6 +47,7 @@ metadata:
     --duration <auto-duration>
   ```
 - Preferred export command from a built fork:
+
   ```sh
   ./node_modules/.bin/electron bundles/main.js \
     --agent-export \
@@ -64,9 +62,10 @@ metadata:
     --agent-headless
   ```
 - For auto previews, use a real-time export window anchored to the logged auto start instead of the start of the file. Prefer the first timestamp where `/DriverStation/Autonomous` and `/DriverStation/Enabled` are both true, then export `--start <auto-start> --end <auto-start + auto-duration> --fps 30`.
-- Realtime export records the active 2D/3D field canvas while AdvantageScope playback runs at 1x. A current Team 8044 fork waits for selected-renderer readiness before recording, so 3D field assets should be loaded in frame 0 without a fixed delay. A preview should take roughly the requested video duration plus small startup/transcode overhead.
+- Realtime export records the active 2D/3D field canvas while AdvantageScope playback runs at 1x. A current fork waits for selected-renderer readiness before recording, so 3D field assets should be loaded in frame 0 without a fixed delay. A preview should take roughly the requested video duration plus small startup/transcode overhead.
 - For agent exports, prefer `--agent-headless` instead of Chromium `--headless`. It keeps AdvantageScope windows hidden while preserving the normal renderer/WebGL lifecycle needed for 3D recording.
 - The skill includes helper scripts:
+
   ```sh
   python3 .agents/skills/advantagescope/scripts/discover_ascope_assets.py \
     --repo .
@@ -93,10 +92,11 @@ metadata:
     --fps 30
   ```
 - Agent realtime export should produce `preview.webm` directly from the renderer. It should also produce `preview.mp4` when bundled FFmpeg conversion succeeds.
-- Use `--agent-export-mode multiview` for synced hub + satellite realtime videos. The Team 8044 fork records the selected hub canvas and the first satellite canvas at the same time, then composes a side-by-side `preview.mp4` with bundled FFmpeg.
+- Use `--agent-export-mode multiview` for synced hub + satellite realtime videos. The fork records the selected hub canvas and the first satellite canvas at the same time, then composes a side-by-side `preview.mp4` with bundled FFmpeg.
 - Use `--agent-export-mode frames` only for debugging deterministic frame stepping; normal auto previews must not use the frame-by-frame path.
 - Use `--agent-allow-frame-only` only with frame export debugging when no encoder is available.
 - Optional control mode:
+
   ```sh
   ./node_modules/.bin/electron bundles/main.js --agent-control --agent-port 0
   ```
@@ -112,9 +112,10 @@ metadata:
 
 - Use Line Graph tab type `1` for static plotted data from WPILOGs.
 - A Line Graph controller state uses `leftSources`, `rightSources`, and `discreteSources`. Numeric source entries should use `type: "smooth"`, `logType: "Number"`, `visible: true`, and `options` with `color` and `size`. Discrete boolean overlays can use `type: "graph"` with `logType: "Boolean"`.
-- Prefer `capture_layout.py` for graph screenshots. It launches the Team 8044 fork in `--agent-control --agent-headless`, waits for `liveStatus.ready=true`, optionally sends `setTime`, sends `capture`, and closes cleanly.
+- Prefer `capture_layout.py` for graph screenshots. It launches the fork in `--agent-control --agent-headless`, waits for `liveStatus.ready=true`, optionally sends `setTime`, sends `capture`, and closes cleanly.
 - Pair graph screenshots with `$wpilib-sim` topic summaries from `read_wpilog_values.py` so users get both the visual trend and concrete min/max values.
 - Fast path for current-style graph requests:
+
   ```sh
   python3 .agents/skills/wpilib-sim/scripts/list_wpilog_topics.py \
     --repo . \
@@ -148,22 +149,23 @@ metadata:
   - hub tab 1: Line Graph controller tab,
   - satellite window: attached to the Line Graph controller UUID.
 - Use `scripts/generate_multiview_3d_joysticks_layout.py` for 3D Field + Joysticks layouts. The helper can select the field tab or joystick tab with `--selected-tab field|joysticks` and can select a specific joystick layout for a port, for example `--joystick-port 0 --joystick-layout "Xbox Controller (White)"`.
-- For game pieces, add `--game-piece-topic <Pose3d[] topic>` and `--game-piece-variant <variant>`. For MapleSim 2026 fuel, use `/RealOutputs/FieldSimulation/Fuel` and variant `Fuel`.
-- For projectile or path overlays, add `--trajectory-topic <Pose3d[] topic>`. For MapleSim fuel shots, use `/RealOutputs/FieldSimulation/FuelShotTrajectory`; the helper adds it as a 3D Field `trajectory` source with a bold orange line by default.
-- For robot sources that include Z/pitch/roll, pass `--topic-type Pose3d` to `scripts/generate_field3d_layout.py`. For MapleSim bump traversal, use `/RealOutputs/FieldSimulation/RobotPose3d` with `--topic-type Pose3d`.
+- For game pieces, add `--game-piece-topic <Translation3d[] topic>` and `--game-piece-variant <variant>`. For fuel, use `/Robot/Sim/Fuel/Positions` (all pieces) or `/Robot/Sim/Fuel/InFlight` (airborne only) and variant `Fuel`.
+- For projectile or path overlays, add `--trajectory-topic <Translation3d[] topic>`. For our fuel shots, use `/Robot/Sim/Fuel/LastShotArc`; the helper adds it as a 3D Field `trajectory` source with a bold orange line by default.
+- For robot sources that include Z/pitch/roll, pass `--topic-type Pose3d` to `scripts/generate_field3d_layout.py`. For MapleSim bump traversal, use `Sim/RobotPose3d` (logged under `/Robot/` by DogLog) with `--topic-type Pose3d`.
 - Use explicit field and robot model arguments from the task or asset discovery. Do not bake season-specific asset ids into the skill or helper defaults.
 - Export with `export_preview.py --mode multiview --headless`. Expected outputs are `view-0-hub.webm`, `view-1-satellite.webm`, `preview.mp4`, and `manifest.json`.
-- Joysticks multiview is supported by the Team 8044 fork's renderer-owned canvas capture path. Validate manifests report the joystick view with `sourceMode: "renderer-canvas"` and `sourceName: "JoysticksRenderer"`.
+- Joystick multiview is supported by the fork's renderer-owned canvas capture path. Validate manifests report the joystick view with `sourceMode: "renderer-canvas"` and `sourceName: "JoysticksRenderer"`.
 - Fast path for a 3D teleop preview plus live Xbox joystick pane:
+
   ```sh
   python3 .agents/skills/advantagescope/scripts/generate_multiview_3d_joysticks_layout.py \
     --out artifacts/auto-previews/layouts/teleop-3d-joysticks.json \
     --field "<field-id>" \
     --robot "<robot-model>" \
-    --pose-topic /RealOutputs/FieldSimulation/RobotPosition \
-    --game-piece-topic /RealOutputs/FieldSimulation/Fuel \
+    --pose-topic /Robot/Sim/RobotPose3d \
+    --game-piece-topic /Robot/Sim/Fuel/Positions \
     --game-piece-variant Fuel \
-    --trajectory-topic /RealOutputs/FieldSimulation/FuelShotTrajectory \
+    --trajectory-topic /Robot/Sim/Fuel/LastShotArc \
     --joystick-port 0 \
     --joystick-layout "Xbox Controller (White)" \
     --selected-tab field
@@ -180,6 +182,7 @@ metadata:
     --mode multiview
   ```
 - Fast path for a 3D auto preview synced with a four-topic current graph:
+
   ```sh
   python3 .agents/skills/advantagescope/scripts/generate_multiview_layout.py \
     --out artifacts/auto-previews/layouts/multiview-3d-drive-current.json \
@@ -231,4 +234,4 @@ metadata:
 2. Discover custom assets with `discover_ascope_assets.py`, generate a layout with explicit field/model arguments, and export with `export_preview.py`.
 3. Check `manifest.json` for the requested mode, selected renderer readiness, no missing sources/assets, valid video output, and no main/renderer error. Multiview exports should include per-view webm files plus composed `preview.mp4`.
 4. Confirm wall-clock export time is close to the requested video duration, then inspect frame 0 and a mid-run frame for nonblank output, expected field asset, and visible robot pose. If using an older fork without renderer readiness, use visual inspection fallback and only add a short explicit delay as a last resort.
-5. If the fork code itself must change, move to the Team 8044 AdvantageScope checkout and use `$advantagescope-fork`.
+5. If the fork code itself must change, move to the AdvantageScope checkout and use `$advantagescope-fork`.