Repository navigation
v1.1.92
Added
-
Seven more cookbook examples for everyday requests a design worker had nothing close to (scan-to-make #734):
desk_clamp_hook(a rounded S-frame headphone hook that clamps an 18 to 30 mm desk with a printed thumbscrew),desk_edge_cable_holder(a five-slot stick-on cable rail with sprung arms),drawer_divider(slotted strips that cross in half-laps, sized from the drawer),organizer_tray(a tapered bin with round corners and a scooped floor),plant_pot(a revolved pot with a ribbed saucer),shelf_bracket(a 6 x 8 in bracket with a curved brace) andbroom_grip(a one-screw spring clip for a handle). Each builds valid throughcad_runin the confined runtime on Linux, and its renders were looked at and judged against the bought product it cites; none has been printed. The cookbook now holds nineteen. -
Every cookbook example has a "Use for:" line naming the requests it suits.
cad_examplesreturns it asuse_forin the list and in each record, andcad_designlists the cookbook by it (about 2.5 KB for the nineteen), so a worker chooses by purpose. -
cad_runreturnsnext, beside the render paths: read the perspective view before the next change, and say what it shows is wrong against what is being made. -
cad_runreturnsprinted: whatever the script wrote withprint(), up to 4000 characters (build.logkeeps the rest). A design can report a number it computed, such as a spring's strain, without making a run fail to read it. It is the script's own words, not a measurement by the engine.broom_gripanddesk_edge_cable_holderprint their arm strain this way. -
X1C can print and upload over the native route with
connection_mode: "bambu_native"(Bambu Studio's networking plug-in, afternpm run build:nativeon macOS). An X1C on firmware 01.12.00.00, with LAN Only and Developer Mode on and a ready, writable SD card, answers every direct FTPSSTORwith553 Could not create file., soprint_3mfandupload_filecannot reach it, while Bambu Studio prints normally. The X1C default stayslan_mqtt_ftpsand is not redirected; an X1C upload that fails with 553 now says to retry withbambu_native. The native route keeps the shared inspection, AMS mapping and fresh printer-state checks. Native controls stay X2D-only, socancel_print,pause_printandresume_printnamedbambu_nativefor another model now use direct MQTT instead of failing with "The selected printer is not an X2D." Checked with mocked transport regressions and one physical X1C print on macOS (firmware 01.12.00.00, AMS 2 Pro slot, job stored on the SD card); Linux, other X1C firmware and X1E are untested, and X1E stays excluded.
Changed
cad_designstarts a new design from the product, in order: name two or three products a person would buy and say which the design follows; choose from the cookbook by purpose and read the example whose use is nearest; only then choose the mechanism. The concept, with the product it is modelled on, goes in the script's opening docstring, as in the cookbook, so the choice can be seen. In a test with Sonnet 5.5 the worker chose a clamp and a screw first and never listed the cookbook, and built a bracket with the mechanism bolted on.cad_designrequires a look after every run: read the perspective view before the nextcad_runor any other change, and before the final run list what the views show is wrong against the named product.cad_check_fitwaits for the engine's execution slot up to itstimeout_ms, ascad_rundoes, and returnsqueue_wait_ms. It refused withcad-operation-busywhenever another CAD operation was running, so a worker that sent four checks at once lost three. A wait that outlasts the timeout ends withcad-queue-timeout;cad-operation-busyis no longer returned.
Fixed
-
Accept a P1S-sliced job on a printer that freshly identifies itself as a P1P, following Bambu Studio's compatibility table while retaining the P1P's model, nozzle, material, thermal and readiness checks. Contributor inspection and hardware evidence is distinct from mocked MCP dispatch validation.
-
Align direct P1S calibration booleans with the shared Studio option resolver, preserve explicit choices, and report requested/resolved/transmitted options and unsupported overrides without guessing SDK wire fields. Payload regressions use fake transports with network blocked; this parity work does not establish a cure for scan-to-make #605 or physical print success.
-
cad_runno longer reports a printed screw in its nut as touching. The exact B-rep distance could call a shaft in its threaded hole "inside" the other solid and return 0 where the tessellations were apart and the solids shared no volume (desk_clamp_hookclamped at a 31 mm opening, 0.28 mm clear). In that case the surfaces are measured instead. Real contact and real overlap still read 0. -
Require fresh ready external storage for native X1C printing unless the fresh raw report advertises internal printing through Studio's
fun2bit 0; recheck the selected storage at dispatch and reject unsupported explicit/environment eMMC requests. Validate storage changes, material mismatch and byte-verified MQTT resume with fake transports. Linux hardware remains unverified. -
Accept an unambiguous single external slot reported as 255 on single-nozzle X1C firmware 01.12 while retaining material checks, explicit mapping validation and dual-nozzle selection.
Thanks to @ianhyzy for #140 (integrated by #152) and @geooff for #128 (integrated by #153). Contributor-reported hardware results and mocked maintainer validation are distinguished in CONTRIBUTORS.md.