0.140.0: One local study can test phone, tablet and desktop, and every run ends with findings

humanish 0.140.0 lets a local browser study give each participant its own device. One study with
mobile, tablet and desktop participants runs them side by side on your machine, with mobile
emulation for the phone, as hosted desktops already did; humanish runtime setup installs the new
local runtime. A study with a Codex participant is now analyzed by the same Codex account on every
target, so a run ends with findings without an OpenAI key; the default OpenAI analysis cap is $10;
a skipped analysis names the command that runs it; and run --json carries the findings.

Added

• A local browser study (execution.target: local) can size each participant's screen:
  execution.desktop.device, participants[].device and execution.desktop.resolution from 500
  to 1920 pixels wide and 480 to 1200 high, and execution.desktop.fidelity.mobileEmulation gives
  phone participants a mobile viewport, pixel ratio, touch and user agent, as on hosted desktops.
  One study can run mobile, tablet and desktop participants side by side. This needs the
  runtime this release pins (runtime-2026.10.21.1 for amd64, runtime-2026.10.21.2 for arm64);
  humanish runtime setup installs it. On an older runtime, doctor --study and run refuse a
  device study before any VM starts and say so (#1916, #1918).

Changed

• A study that declares no review.analysis gets its analyst from its participant. A Codex
  local-agent participant's run is now analyzed by the same Codex account on a hosted desktop
  too, so a study run on a signed-in Codex ends with findings and needs no OPENAI_API_KEY.
  Before, only a local browser study did this. Other participants, Claude Code included, keep the
  OpenAI analyst. doctor --study and study check show the analyst the run will use. When a
  default Codex analyst cannot start, for example with a keyring-only Codex login, the analysis is
  recorded as failed with its reason and the run still passes.
• The default cap on the OpenAI analysis's expected cost is $10, up from $3. A one-request
  analysis is admitted at the evidence limits, and 17 to 40 participants are admitted from $4.60
  to $7.05 with no evidence. Set review.analysis.maxCostUsd to choose another cap.
• run --json and watch --json say what the analysis left. automaticAnalysis.message says what
  happened and automaticAnalysis.nextCommand is the command that gets findings, or humanish review --run <id> when there are some. A run whose analysis left findings has findings: the
  count, the top five (title, headline, impact, confidence, participant ids), the review
  command and the Observer page that opens on the findings.
• The humanish skill tells agents to read and report a live run's findings next to the run
  commands, and no longer suggests copying review.analysis: false from the scripted demo. The
  humanish/README.md that init writes says a live run ends with findings and that humanish review --run latest shows them.

Fixes

• An analysis skipped or refused after a live run names the next command in the human output, on a
  next: line. Without OPENAI_API_KEY, the command is humanish analyze --run <id> --provider codex when Codex is signed in with a ChatGPT account, and the message names
  review.analysis.provider: codex as the study setting; otherwise it says to set the key.
  humanish review gives the same next step. Before, the run printed analysis: skipped because OPENAI_API_KEY is not set with no command, and review suggested only --max-cost 3.