The useful idea is confirmation before drawing
EditaPlot covers 41 documented Origin plotting routes, but its best idea is smaller than that catalog: an agent should not guess what a scientific column means. The skill sorts columns into evidence, visible support, calculation or validation data, retained data, and unresolved data. A changed source file or mapping invalidates the confirmation. For XRD refinement data, that means Observed can become points while weight or axis-control columns stay out of the plot unless the user says otherwise.
That restraint makes EditaPlot more credible than a prompt that says "make this publication ready." The README explicitly refuses to calculate fits, remove outliers, assign phases, or infer conclusions. Reference images are treated as visual briefs rather than data sources. A PNG cannot smuggle its labels or fitted values into the result, and an unsupported style request is rejected or left at the template default. Scientists still have to own the scientific claim.
Origin 2026 is a hard dependency, not an export option
The complete path supports physical Windows 10/11 x64 and Origin or OriginPro 2026 product number 10.30. EditaPlot starts a dedicated Origin instance, runs a smoke check, renders, exports OPJU, PNG, PDF, and TIF, then reads objects back and calls for a human visual check. Python previews do not count as completed Origin work. That is a sensible acceptance rule for editable figures, but it makes Origin the center of the system.
Mac, Linux, WSL, Wine, CrossOver, Parallels, and other virtual machines are outside the V1 support boundary. Python must be 64-bit version 3.10, 3.11, or 3.12. The launcher can create a project environment, yet it will not install or modify Origin. A team without the exact commercial desktop application cannot use the main result path, regardless of how easily the Python portion installs.
What happened when we ran it
Our run at commit cc12ad4 installed 35 packages in 30 seconds and occupied 37 MB. The build completed in 9 seconds. Our sandbox was an unprivileged Debian container with 3 CPUs, 8 GB of RAM, Python 3.12, and no secrets. The checkout held 677 files, about 68,740 lines of source, and used 10.5 MB. Pip-audit reported 0 known vulnerabilities.
Pytest failed in 20 seconds. Of 291 tests, 238 passed, 19 failed, and 34 ended in collection or setup errors. The log tail showed a missing page.width mapping, README release-gate assertions, and a missing star-trend workflow file. Several semantic workflow cases could not import modules named nu or matplotlib. Those messages establish what failed. They do not establish why those modules or files were absent.
Our test method exercises repository mechanics in a fresh Debian container, not the advertised Origin render on physical Windows. We did not produce an OPJU file or judge image quality. The Linux suite result still matters because the repository includes those tests and 53 cases did not reach a pass. A buyer should reproduce the suite on the intended Windows host and require a real smoke, render, export, and object-readback cycle before trusting the workflow.
The permission model is narrow but still operational work
EditaPlot documents one exact escalation point: a sandboxed Codex process must receive approval to run the chosen origin-smoke or render command in the signed-in Windows session. The project says normal use does not need administrator rights, whole-drive access, DCOM edits, registry changes, or firewall changes. If approval is denied, the process stops before Origin COM and must not describe the figure as complete.
Outputs go into a new timestamped folder beside the CSV, TXT, XLS, or XLSX input rather than overwriting the source. Network access is mainly for the initial clone, updates, and locked Python packages. Medical data supplied to Codex remains subject to the account and organization's retention policy, and the README says EditaPlot does not promise automatic PHI detection. That boundary deserves a written lab procedure before anyone attaches patient material.
Recent code does not cancel the failed suite
GitHub showed 124 stars, 0 open issues and pull requests, and a last push on September 20, 2026. Release v0.1.3 was published the same day. Those dates show recent maintenance, while an empty queue says little about field use or response quality. The release notes are also unusually rough: their body ends with an instruction to click the Publish release button, which reads like authoring residue in a public artifact.
EditaPlot earns a controlled trial for a lab already standardized on OriginPro 2026. The confirmation contract is careful, the bilingual documentation is detailed, and 238 tests passed in our container. The failed 291-test run blocks a broader recommendation. Put one representative dataset through the full Windows path, inspect every exported artifact, and keep the scientist who owns the claim in the approval loop.