[CI debug for #542] Stream Cloud Build logs to diagnose integration-test failure

.github/workflows/integration-test.yml

@@ -36,6 +36,8 @@ jobs:
 
       - name: Set up Google Cloud SDK
         uses: google-github-actions/setup-gcloud@v3
+        with:
+          install_components: beta
 
       - name: Submit integration tests to Cloud Build
         run: |
@@ -54,7 +56,7 @@ jobs:
           if [ "${{ github.ref }}" = "refs/heads/main" ]; then
             publish_base="true"
           fi
-          gcloud builds submit \
+          gcloud beta builds submit \
             --project="${{ vars.GCP_PROJECT_ID }}" \
             --region="${{ vars.GCP_REGION }}" \
             --gcs-source-staging-dir="${{ vars.GCP_CLOUD_BUILD_SOURCE_STAGING_DIR }}" \

.gitignore

@@ -100,6 +100,9 @@ workspace/.training_pending/
 # Per-skill persistent key-value stores (Skill.storage, skills/types.py).
 workspace/skill_storage/
 
+# Generated on-robot by the skill catalog (typed physical-skill refs).
+workspace/physical_skills/
+
 # recordings folder
 recordings/
 
@@ -116,7 +119,6 @@ data/.last_mode
 arm_wave/
 
 .vscode
-ros2_ws/src/brain/brain_client/innate/skills.pyi
 workspace/innate_skills/**/*.engine
 
 node_modules/

CLAUDE.md

@@ -1,86 +0,0 @@
-# CLAUDE.md
-
-Behavioral guidelines to reduce common LLM coding mistakes. Merge with project-specific instructions as needed.
-
-**Tradeoff:** These guidelines bias toward caution over speed. For trivial tasks, use judgment.
-
-## 1. Think Before Coding
-
-**Don't assume. Don't hide confusion. Surface tradeoffs.**
-
-Before implementing:
-
-- State your assumptions explicitly. If uncertain, ask.
-- If multiple interpretations exist, present them - don't pick silently.
-- If a simpler approach exists, say so. Push back when warranted.
-- If something is unclear, stop. Name what's confusing. Ask.
-
-## 2. Simplicity First
-
-**Minimum code that solves the problem. Nothing speculative.**
-
-- No features beyond what was asked.
-- No abstractions for single-use code.
-- No "flexibility" or "configurability" that wasn't requested.
-- No error handling for impossible scenarios.
-- If you write 200 lines and it could be 50, rewrite it.
-
-Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.
-
-## 3. Surgical Changes
-
-**Touch only what you must. Clean up only your own mess.**
-
-When editing existing code:
-
-- Don't "improve" adjacent code, comments, or formatting.
-- Don't refactor things that aren't broken.
-- Match existing style, even if you'd do it differently.
-- If you notice unrelated dead code, mention it - don't delete it.
-
-When your changes create orphans:
-
-- Remove imports/variables/functions that YOUR changes made unused.
-- Don't remove pre-existing dead code unless asked.
-
-The test: Every changed line should trace directly to the user's request.
-
-## 4. Goal-Driven Execution
-
-**Define success criteria. Loop until verified.**
-
-Transform tasks into verifiable goals:
-
-- "Add validation" → "Write tests for invalid inputs, then make them pass"
-- "Fix the bug" → "Write a test that reproduces it, then make it pass"
-- "Refactor X" → "Ensure tests pass before and after"
-
-For multi-step tasks, state a brief plan:
-
-```
-1. [Step] → verify: [check]
-2. [Step] → verify: [check]
-3. [Step] → verify: [check]
-```
-
-Strong success criteria let you loop independently. Weak criteria ("make it work") require constant clarification.
-
-## 5. Clean, Readable Code
-
-**You are a senior engineer. Write clean, maintainable code.**
-
-Clean code is understood without reading the comments:
-
-- Use clear, descriptive variable and function names that state intent.
-- Keep complexity low. Avoid deep, multiple-levels-of-nesting indentation.
-- Use early returns to flatten control flow instead of nesting.
-- Keep things simple, always.
-
-When handling errors, don't overdo it:
-
-- Avoid scattering multiple try/catch blocks where one is enough.
-- Catch errors at the level where you can actually do something about them.
-
----
-
-**These guidelines are working if:** fewer unnecessary changes in diffs, fewer rewrites due to overcomplication, and clarifying questions come before implementation rather than after mistakes.

README.md

@@ -135,6 +135,9 @@ You will find skills in two different directories:
 
 - **Built-in skills** — Located in `workspace/innate_skills/`.
 - **Your custom skills** — Stored in `workspace/custom_skills/`. Gitignored and yours to play with.
+- **Skill packs** — Any other folder dropped into `workspace/` loads as its own package (ids `<folder>/<name>`). A pack that lives elsewhere on disk is symlinked in (`ln -s /opt/team/skills workspace/team_skills`) and works the same, hot reload included.
+
+Helpers work like normal Python: any `.py` in your skills folder that doesn't define a `Skill` is just a module — `import` it, use relative imports inside subfolders, share across packages by bare name (`from innate_skills import arm_utils`). Device helpers are methods on the interfaces (`self.manipulation.go(...)`, `self.mobility.rotate_by(...)`); camera math and Gemini live under `innate` (`from innate import geometry, vision, gemini`).
 
 ### Skill definition
 
@@ -164,52 +167,27 @@ You will find skills in two different directories:
     <td width="50%" valign="top">
       <strong>Code skill</strong> — call the mobility interface to move forward.<br>
       Saved as <code>workspace/custom_skills/move_forward.py</code>:
-      <pre lang="python">from brain_client.skills.types import Interface, InterfaceType, Skill, SkillResult
-import time
+      <pre lang="python">from innate import Mobility, Skill, SkillReturn
 
 
 class MoveForward(Skill):
-    """Move the robot forward by a given distance."""
-
-    mobility = Interface(InterfaceType.MOBILITY)
-
-    def __init__(self, logger):
-        super().__init__(logger)
-        self._cancelled = False
+    """Move the robot forward by a given distance in meters."""
 
-    @property
-    def name(self):
-        return "move_forward"
-
-    def guidelines(self):
-        return "Move the robot forward by a given distance in meters."
-
-    def execute(self, distance_m: float = 0.5):
-        self._cancelled = False
-
-        if self.mobility is None:
-            return "Mobility interface not available", SkillResult.FAILURE
+    mobility: Mobility          # declare what you use; the runtime injects it
 
+    def execute(self, distance_m: float = 0.5) -> SkillReturn:
         speed = 0.2  # m/s
         duration = distance_m / speed
         self.mobility.send_cmd_vel(linear_x=speed, duration=duration)
-
-        stop_at = time.time() + duration
-        while True:
-            remaining = stop_at - time.time()
-            if not remaining > 0:
-                break
-            if self._cancelled:
-                self.mobility.send_cmd_vel(linear_x=0.0)
-                return "Move cancelled", SkillResult.CANCELLED
-            time.sleep(min(0.05, remaining))
-
-        return f"Moved forward {distance_m} m", SkillResult.SUCCESS
-
-    def cancel(self):
-        self._cancelled = True
-        return "Move cancelled"
+        self.sleep(duration)    # like time.sleep, but a Stop unwinds it
+        return f"Moved forward {distance_m} m"
 </pre>
+      The return value is the run's result message; call
+      <code>self.fail(message)</code> to end the run as a failure.
+      Cancellation is the framework's job: <code>self.sleep</code> (and every
+      blocking framework call) raises the moment a Stop lands, the base is
+      braked automatically, and the run reports CANCELLED — skills carry no
+      cancel code.
     </td>
   </tr>
 </table>
@@ -336,7 +314,7 @@ Input devices live in [`workspace/inputs/`](workspace/inputs/) and are pure Pyth
 import threading
 import time
 
-from brain_client.input_types import InputDevice
+from brain_client.inputs.types import InputDevice
 
 
 def read_thermometer_celsius() -> float:

ci/run_integration_tests.sh

@@ -83,7 +83,6 @@ fi
 echo "=== unit tests (fast, no ROS) ==="
 PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 python3 -m pytest -q \
   src/brain/brain_client/test/test_fake_cloud_selftest.py \
-  src/brain/brain_client/test/test_backwards_compat.py \
   src/brain/manipulation/test/test_config_validation.py
 
 echo "=== unit tests: webapp front door (aiohttp, no ROS) ==="

config/settings.yaml.template

@@ -167,12 +167,9 @@
 #     cmd_publish_hz: 50.0                  # cmd_vel republish rate during a move
 #     poll_period_sec: 0.02                 # action-loop poll interval
 
-# ── Extra agent / skill directories ───────────────────────────────────
-# Scan agents/skills from extra locations, on top of the built-in workspace/ dirs.
-# Absolute paths anywhere on the machine, scanned in place (never created); in a Docker/sim
-# setup, host paths must also be mounted into the container. (Read directly by the agent /
-# skill loaders — not a ROS node, so leave the section name as `script_paths`.)
-# script_paths:
-#   ros__parameters:
-#     extra_agent_dirs: ["/home/me/my-agents", "/opt/team/agents"]
-#     extra_skill_dirs: ["/home/me/my-skills"]
+# Extra agent/skill directories (script_paths.extra_agent_dirs / extra_skill_dirs)
+# were removed in 0.7: agents and skills load only from workspace/. To add a
+# skill pack, drop its directory into workspace/ — it is picked up (and
+# hot-reloaded) without a restart. A pack that must stay elsewhere on disk is
+# symlinked in instead, and behaves the same:
+#     ln -s /opt/team/skills ~/innate-os/workspace/team_skills

docs/INPUT_DEVICES.md

@@ -23,7 +23,7 @@ class HelloWorld(Directive):
 ```python
 # workspace/inputs/my_sensor_input.py
 
-from brain_client.input_types import InputDevice
+from brain_client.inputs.types import InputDevice
 import threading
 import time
 
@@ -215,9 +215,9 @@ innate-os/
 │       └── hello_world_directive.py  # Uses get_inputs()
 └── ros2_ws/src/brain/brain_client/
     └── brain_client/
-        ├── input_types.py            # Base class
-        ├── input_loader.py           # Auto-discovery
-        └── input_manager_node.py     # ROS bridge
+        ├── inputs/types.py           # Base class
+        ├── inputs/loader.py          # Auto-discovery
+        └── nodes/input_manager.py    # ROS bridge
 ```
 
 ## Summary

docs/PARAMETERS.md

@@ -54,7 +54,6 @@ To tune something, **uncomment a whole stanza** (the `node:`, `ros__parameters:`
 | `navigation_grid_localizer` | `max_score_threshold`, `max_range`, `auto_localize_timeout` | `0.3`, `12.0`, `30.0` |
 | `brain_client_node` | `cartesia_voice_id` (TTS voice), `vertical_fov`, `pose_image_interval`, `scan_stale_after_sec`, `send_depth`, `send_arm_camera_image`, `log_everything`, STT/transcribe models | see template |
 | `uninavid_node` (VLN) | `forward_speed`, `turn_speed`, `cmd_duration_sec`, `image_send_hz`, `consecutive_stops_to_complete`, `cmd_publish_hz`, `poll_period_sec` | `0.3` / `0.8`, rest see template |
-| `script_paths` | `extra_agent_dirs`, `extra_skill_dirs` (extra dirs scanned on top of `workspace/`) | `[]` |
 
 > **Driving caps vs the safety clamp.** `motion_control` is the *driving feel* cap: the
 > joystick, keyboard, and app drive joystick all ship the same `0.4` m/s / `1.0` rad/s

pyproject.toml

@@ -20,7 +20,7 @@ ignore = [
 ]
 
 [tool.ruff.lint.isort]
-known-first-party = ["mars_nav", "mars_control", "mars_cam", "mars_arm", "brain_client", "manipulation"]
+known-first-party = ["mars_nav", "mars_control", "mars_cam", "mars_arm", "brain_client", "manipulation", "workspace", "innate", "innate_proxy"]
 
 [tool.ruff.format]
 quote-style = "double"

pyrightconfig.json

@@ -0,0 +1,29 @@
+{
+  "pythonVersion": "3.10",
+  "pythonPlatform": "Linux",
+  // Scoped to the tree that is actually clean under standard mode: the skills
+  // API and the skills written against it. Unscoped, pyright checks all 291
+  // files in the repo and reports ~230 real diagnostics from code that was
+  // never typed — burying the ones a skill author needs to see. Add paths
+  // here as they get cleaned (test/ and the rest of ros2_ws are not yet).
+  "include": [
+    "ros2_ws/src/brain/brain_client/brain_client/skills",
+    "ros2_ws/src/brain/brain_client/innate",
+    "workspace/innate_skills"
+  ],
+  "extraPaths": [
+    "ros2_ws/src/brain/brain_client",
+    "ros2_ws/src/cloud/clients/proxy-client",
+    "ros2_ws/install/brain_client/local/lib/python3.10/dist-packages",
+    "ros2_ws/install/brain_messages/local/lib/python3.10/dist-packages",
+    "ros2_ws/install/innate_cloud_msgs/local/lib/python3.10/dist-packages",
+    "ros2_ws/install/mars_msgs/local/lib/python3.10/dist-packages",
+    "/opt/ros/humble/lib/python3.10/site-packages",
+    "/opt/ros/humble/local/lib/python3.10/dist-packages",
+    // System site-packages (pydantic, etc.). Needed when the language server
+    // has no usable interpreter and would otherwise only see the paths above.
+    "/usr/local/lib/python3.10/dist-packages",
+    "/usr/lib/python3/dist-packages",
+    "/usr/lib/python3.10/dist-packages"
+  ]
+}

ros2_ws/src/brain/brain_client/CMakeLists.txt

@@ -15,7 +15,7 @@ ament_python_install_package(${PROJECT_NAME}
   PACKAGE_DIR brain_client
 )
 
-# Public authoring namespace for skill files (`from innate.skills import ...`)
+# Public authoring namespace for skill files (`from innate import Skill, Mobility, ...`)
 ament_python_install_package(innate
   PACKAGE_DIR innate
 )

ros2_ws/src/brain/brain_client/brain_client/agents/initializer.py

@@ -8,16 +8,21 @@
 to keep the main brain_client_node.py clean and focused.
 """
 
-from typing import Any
-
 from brain_client.agents.loader import AgentLoader
+from brain_client.agents.types import Agent
 from brain_client.common.script_paths import (
     ensure_user_directories,
     get_agent_directories,
+    get_workspace_dir,
 )
+from brain_client.skills.physical_refs import render_refs, write_refs
+
+# roster types that are physical skills (data, no code class) — these are what
+# the generated physical_skills package covers
+_PHYSICAL_TYPES = frozenset({"learned", "replay", "eval", "physical"})
 
 
-def initialize_agents(logger, skills_dict: dict[str, Any] | None = None) -> tuple[dict[str, Any], Any | None]:
+def initialize_agents(logger, skills_dict: dict[str, dict] | None = None) -> tuple[dict[str, Agent], Agent | None]:
     """
     Initialize all agents using dynamic loading.
 
@@ -36,6 +41,13 @@ def initialize_agents(logger, skills_dict: dict[str, Any] | None = None) -> tupl
     # and, if present, ~/agents (in place — never moved).
     ensure_user_directories()
 
+    # Agent files may `from physical_skills import X`, so make sure the
+    # generated package matches this roster before importing them. The skills
+    # server writes it too (on every publish); write_refs content-compares, so
+    # whichever runs second is a no-op. Doing it here as well means agent
+    # loading never depends on the two processes' ordering.
+    _regenerate_physical_refs(logger, skills_dict)
+
     agents_directories = [str(p) for p in get_agent_directories()]
 
     # Load all agents dynamically from all directories
@@ -64,3 +76,13 @@ def initialize_agents(logger, skills_dict: dict[str, Any] | None = None) -> tupl
         logger.error("No agents loaded! This will cause issues.")
 
     return agents, default_agent
+
+
+def _regenerate_physical_refs(logger, skills_dict: dict[str, dict] | None) -> None:
+    """Write workspace/physical_skills/ from the roster metadata. Skipped when
+    no roster is available (nothing to generate from — an existing package is
+    left alone rather than emptied)."""
+    if not skills_dict:
+        return
+    entries = [meta for meta in skills_dict.values() if meta.get("type") in _PHYSICAL_TYPES]
+    write_refs(get_workspace_dir() / "physical_skills", render_refs(entries), logger)