Files
earthtojakeandClaude Opus 5.5 523ae21341 Release 0.7.19: fix what day one of PostHog telemetry showed (Windows mesh export, cad_file and cad_screenshot failures, crash noise, failure reasons) (#586)
**This PR is the 0.7.19 release** (`scripts/release/bump-version.sh
patch`): merging it runs Publish Release. Its receiver changes under
`apps/api` deploy on the same merge through Deploy API, minutes before
PyPI has 0.7.19, so schema 4 is read before any client sends it.

Fixes for what PostHog's first day of telemetry showed (2026-10-08
00:14Z to about 21:40Z: about 209 installs and 59 crash reports). It
covers three bugs people are hitting, crash reports that were not
cadgen's bugs, and gaps in what the receiver lets us see. There is one
commit per fix.

## Bugs

**1. Builds that export a mesh crashed on Windows** (7 installs, all
Windows, about 26 crashes). `mesh_export.py` ran the Node exporter with
`text=True` and no encoding, so Windows read its UTF-8 output in the
local code page. The exporter's JSON report names every output path, so
any output folder whose name the code page cannot read (for example
`Рабочий стол` under cp1252, or most Chinese text under cp936) made
CPython's Windows output reader die quietly. `proc.stdout` came back
`None`, and `.splitlines()` raised an `AttributeError`. The exporter now
reads `utf-8` with `errors="replace"`, which keeps the JSON line intact.
The same fix goes into `run_node_builder`, whose input was also silently
empty under cp1252. ffmpeg, `gz sdf` and `doctor` now read `utf-8` with
`errors="backslashreplace"`, and doctor's child process is set to
`PYTHONIOENCODING=utf-8`. The tests force subprocess's default encoding
to cp1252, and both fail without the fix.

**2. `cad_file` failed on 48 of 49 calls on Windows** (5 of 6 installs).
Codex for Windows names a file opened from its file tree as
`openai/resource.path = "/C:/Users/…"`, read from the desktop bundle.
Python 3.13's `ntpath.isabs("/C:/…")` is False, so every call answered
"not an absolute path". The `file.resourceUri` alongside it is a
`codex-resource://` handle, so the fallback never helped. A new
`local_path` drops the slash before a drive on Windows, both for file
URIs and for plain paths, for `cad_file`, `cad_open` and `cad_show`.
This most likely also explains Antigravity's `cad_show` failures on
Windows (7 of 12). The Windows CI job now passes the path the way Codex
spells it.

**3. `cad_screenshot` failed on 30% of calls** (11 of 19 installs). The
most likely cause is an agent capturing straight after build, show or
open, while the view is still loading or has not synced yet. The view
refused with "Wait for the displayed model revision to finish loading",
"That viewer is not open" or "No CAD viewer with a model is open", or a
large model ran past the fixed 10 s wait.
- The page now waits until the view shows the requested model, loaded
and drawn (`CAPTURE_SETTLE_MS`, 20 s).
- The server waits for a view it just opened to sync (`OPENING_SECONDS`,
15 s) within one budget for the whole capture (`CAPTURE_SECONDS`, 40 s).
- The capture's reply still goes on its own call (`void answer(event)`),
so no view call is held open.

## Crash reports that were not cadgen's bugs
- **Windows viewer disconnects.** `ConnectionAbortedError` (WinError
10053) made up most of the crash volume: 23 installs. The viewer caught
only `BrokenPipeError` and `ConnectionResetError`, and the header write
had no guard. Every write to the socket now treats any `ConnectionError`
as the page having left.
- **A model's own mistakes.** A build123d name that does not exist,
raised through the `cadgen.build123d` re-export, and a non-string passed
to `srgb()`. Both now raise deliberately, so the existing rule counts
them as the person's error, and `srgb` raises a `TypeError` naming what
it was given.
- **Stopped workers.** A worker stopped by SIGTERM, SIGINT or SIGHUP (a
person quitting it, a logout) now counts as cancelled, not crashed.
SIGSEGV, SIGABRT and SIGKILL are still reported.

## Telemetry: what we can now see
- **Why a tool call failed.** There is a new `tool_failure {tool,
reason, count}` event in batch schema 4, which PostHog receives as
`tool_failed`. The reason is one word from a fixed list (`no_path`,
`relative_path`, `no_file`, `not_cad`, `no_view`, `wrong_view`,
`bad_request`, `timeout`, `view_error`, `too_large`, `no_viewer`, `bug`,
`other`), chosen where the call fails and never taken from a message. A
test checks that every `ToolFailed` and `NoAnswer` names one.
- **Rollout: the receiver goes first.** The API is its own Vercel
project now (#587) and deploys on merge to `main`, so merging this PR
puts the schema 4 receiver live before any release sends schema 4. A
refused batch is dropped, as before; there is no fallback in the client.
- **Refused batches are logged.** Each 400, 403 or 415 is one
`console.warn` line naming the rule that failed and the cadgen version.
Values, install ids and service messages are never logged. Vercel's
per-status counts need Observability Plus, so this is the only way to
see a refusal. The privacy policy says so.
- **Errors are logged by name**, for example `TimeoutError` instead of
`23`. A `/v1/forget` timed out at 17:02Z, and the client retries it.
- **`$session_id`** is now set, so error tracking can count sessions.
Our ids are UUIDv4, so PostHog's sessions table leaves them out; error
tracking should still read them, which needs checking after deploy.

Privacy policy, README and `apps/api/README.md` are updated where what
is sent or logged changed.

## Not in this PR
- **Deduplicating a resent batch.** The sender rebuilds a failed window
instead of resending it, and a batch has no id, so there is nothing
stable to dedupe on yet. It needs a per-batch id from the sender.
- **Dashboard totals.** PostHog's error-tracking "occurrences" counts
events, not each event's `count`; for the mesh-export crash that is 5
against 22. That is fixed on the dashboard side (t2c-analytics).
- **5 of 15 DXF builds failed.** DXF builds don't go through Node, so
the encoding fix doesn't cover them and they still need a look.

## Needs a real host
- Windows Codex: open a `.step` from the file tree; capture from a tab
hidden behind another tab.
- Claude Desktop: capture right after `cad_show` on a large STEP, or
while the card waits on Allow.
- Antigravity on Windows: confirm the path spelling it sends.

## Tests
Full suites on this branch, in a provisioned worktree (`.venv` from
`requirements-dev.txt`, `npm ci`, `bundle.sh --check`,
`CADGEN_DAEMON=0`): all pass.
- `scripts/test/test-python.sh --keep-going`: 2,774 tests in 8 groups,
OK.
- `scripts/test/test-js.sh`: every group passes (core, ui, web, mcp).
- `scripts/test/test-docs.sh`: receiver tests 30/30 and the rest 16/16.
- `scripts/test/test-global.sh`: 210 tests, OK (1 skipped).

Each new regression test was run against the old code, and each fails
there.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-09 01:35:24 +00:00

10 KiB

name, description, license
name description license
sdf SDFormat/SDF model and world authoring, validation, and simulator handoff. Use for `.sdf` files, SDFormat XML, models, worlds, links, joints, poses, frames, inertials, visual/collision geometry, mesh URIs, sensors, lights, physics, plugins, includes, Gazebo, static SDF review, or simulator-specific metadata. Do not use for signed-distance-field geometry. Open and visually review existing SDF files in CAD Viewer. MIT

SDF

Provenance: maintained in earthtojake/text-to-cad. Use the installed local skill files as the runtime source of truth; the repository link is only for provenance and release review.

Use this skill when the deliverable is an SDFormat document. SDFormat describes simulator and world behavior: models, worlds, frames, poses, links, joints, inertials, visuals, collisions, sensors, lights, physics, plugins, includes, and simulator metadata.

This skill is for SDFormat, not signed-distance-field geometry.

The .sdf file is the source of truth: author and edit the XML directly. There is no gen_sdf() contract.

Setup

Run cadgen through uv, so this skill's commands share one installation, and its warm build daemon, with the CAD app's server:

  • cadgen below means uvx --no-config --managed-python --python 3.13 --from cadgen==0.7.19 cadgen
  • python below means uvx --no-config --managed-python --python 3.13 --from cadgen==0.7.19 python

The first run downloads that installation and the first snapshot its headless browser; later runs reuse both.

Core rules

  1. Author .sdf XML directly and validate every created or modified file with cadgen sdf validate before reporting completion.
  2. Identify the target consumer before editing: Gazebo/libsdformat version, another simulator, visualization-only tooling, model package, or world handoff.
  3. Decide document kind: model-level SDF, world-level SDF, or model-in-world. Prefer model-level SDF for reusable robot/object exports.
  4. Use SI units unless the target explicitly requires otherwise: meters, kilograms, seconds, radians.
  5. Prefer version="1.12" for new outputs unless the target consumer constrains the version.
  6. Establish the design ledger before writing poses, frames, joint axes, mesh scales, inertials, sensors, or plugins, and keep it as a comment block at the top of the .sdf. Use references/design-ledger.md and references/llm-guardrails.md.
  7. Write relative_to / expressed_in explicitly on every nontrivial pose and axis. Implicit frame defaults are the top SDF failure mode. See references/frame-semantics.md.
  8. Do not infer spatial transforms from visual impression alone. Derive poses, axes, scale, mass, inertia, and frame names from upstream source data, drawings, simulator documentation, measured values, or explicit assumptions. Never freehand computed numbers — use formulas or a throwaway helper script (inertia tensors, unit conversions).
  9. When the robot already has a URDF, derive the SDF from it instead of re-authoring geometry; see references/interoperability.md.
  10. Regenerate upstream geometry, mesh, robot-description, render, topology, or package assets with their owning workflows before editing SDF that references them.
  11. After authoring, run available checks: bundled validation (which runs gz sdf --check itself whenever gz is on PATH), simulator load, joint motion, and plugin/sensor startup.
  12. Report assumptions, skipped checks, unresolved resource paths, and target-specific compatibility risks.

Scope

Use this skill for SDFormat outputs. Do not use it for signed-distance-field modeling, raw geometry generation, planning semantics, or to paper over incorrect upstream robot/source data unless the task is explicitly simulator-only.

Show the model

Show the user each file you create or change, and any they ask to see. Snapshots and validation don't replace this.

  • If your tools include cad_show (your host may prefix it), use it with the file's absolute path, and follow its description for when to call it again. cad_view reads what the user selected; cad_screenshot shows you what they see. Neither is a review of your own work.

  • Otherwise run the CAD Viewer, from any folder:

    cadgen viewer --host 127.0.0.1 --json --detach
    

    --detach returns once the viewer answers requests and leaves it running in the background: always pass it, since a foreground viewer never exits (and piping its output through tail can hide the URL for good). It starts this machine's one viewer, or reuses it. Read url from its one JSON line (never guess the port), and for each file return url?file=<its URL-encoded absolute path>. If it fails to launch, say so.

Review placement, resources and joints. The viewer does not execute simulator plugins or validate dynamics; keep simulator checks separate.

Workflow

  1. Locate the target .sdf and its consumers.
  2. Read or create the design ledger comment block.
  3. Read references/frame-semantics.md before editing any <pose>, <frame>, joint axis, relative_to, expressed_in, nested scope, sensor frame, or plugin frame.
  4. Author the XML directly, following the worked examples in references/examples.md.
  5. Validate the explicit target with cadgen sdf validate; treat bundled validation as a guardrail, not simulator proof.
  6. Run target-consumer smoke tests when available (references/smoke-tests.md).
  7. Show the result (Show the model). Static rendering does not execute SDF plugins or read file-authored motion metadata.
  8. Report checks run, checks skipped, and assumptions.

Commands

Run cadgen as Setup defines it. cadgen doctor <skill-dir> reports the installation in use and checks that it is the one this skill pins — docs drift silently on another. Validation itself needs nothing beyond the Python standard library; only snapshots need the browser. Use cadgen <verb> --help for the complete current interface.

cadgen sdf validate path/to/model.sdf
cadgen sdf validate path/to/model.sdf --strict
cadgen sdf validate path/to/model.sdf --json
cadgen sdf snapshot path/to/model.sdf review.png

The validator checks document shape, name scopes, pose/frame graphs, joints, geometry, mesh URIs, inertials, sensors, and plugins, and prints its findings plus a summary. One run validates ONE file: --strict treats warnings as failures and --json prints one line of {"ok", "path", "issues": [{"severity", "code", "message", "element", "hint"}], "summary"}, where element is the XML path. It exits nonzero if the target fails.

External checking is on by default:

cadgen sdf validate path/to/model.sdf --gz-check required
cadgen sdf validate path/to/model.sdf --gz-check never

gz sdf --check is target-consumer validation. --gz-check auto is the default: it runs when gz is on PATH, reporting gz_check_passed or the tool's own output as the error gz_check_failed, and otherwise notes info: gz_check_unavailable and carries on. An absent optional tool says nothing about the file, so it never fails a clean document and --strict does not change that. --gz-check required makes the tool mandatory — a missing gz is then an error — and --gz-check never skips it outright.

Required report shape

When finishing an SDF task, include a compact report:

Validated: path/to/model.sdf
Checks run:
- bundled SDF validation: passed
- gz sdf --check: skipped, gz not installed
- simulator load: skipped, target simulator unavailable
- viewer review: live link returned, or explicit launch failure
Assumptions:
- Assumed mesh units are meters.
- Assumed lidar frame is coincident with lidar_link.
Risks:
- Camera plugin filename was not verified in the target simulator environment.

Snapshot Tool

cadgen sdf snapshot renders the robot to a PNG still, using the same shared CLI and headless browser runtime every rendering skill uses — so a snapshot matches what the CAD Viewer shows.

cadgen sdf snapshot path/to/robot.sdf review.png

It accepts .sdf only (a format door, same TARGET [OUT] grammar as the rest). Pose the robot with --joint-values — {joint: degrees} JSON, joints you do not name staying at their defaults, where the CAD Viewer opens the robot (the "jointValues" job field is the same thing in a packet). The snapshot draws the robot with the viewer's own scene, so it shows what the viewer shows, and a link mesh that cannot be loaded fails it rather than leaving the link out. Robots are authored in metres and are framed on the robot scene scale automatically.

A normal snapshot uses the Solid preset and Light appearance; omitted groups inherit preset defaults. Pass --display render for the shared photographic scene. Inline display JSON and JSON files use grouped settings such as lighting, background, and floor; appearance is light (default) or dark. Projection and focal length belong in display.camera. Top-level --camera and --joint-values remain active in every display mode. The display modes are solid and render: edges, clip, exploded, the xray, hidden-line and wireframe modes and the hidden/off surface styles describe a STEP model's CAD edges, parts and solids, and are refused by name here.

Link meshes are resolved relative to the description, so they must be present: an unhydrated Git LFS pointer fails as "No link mesh loaded for robot". Run git lfs checkout <mesh dir> first.

The grammar is cadgen sdf snapshot TARGET [OUT] [flags], the same one every format door uses. Use cadgen sdf snapshot --help for the complete current interface — the flags a robot cannot act on are absent from it, not refused by it.

References

  • SDF workflow: references/sdf-workflow.md
  • Worked examples (golden skeletons): references/examples.md
  • LLM guardrails: references/llm-guardrails.md
  • Design ledger: references/design-ledger.md
  • Frame semantics: references/frame-semantics.md
  • Validation scope: references/validation.md
  • Smoke tests: references/smoke-tests.md
  • Interoperability notes (URDF-derived SDF, meshes, Gazebo): references/interoperability.md