0.3.4

Motifcode 0.3.4 improves terminal rendering and protects against unsafe checkpoint resumes. All three fixes are included in the npm CLI.

Fixes

• Reduce footer flicker. Unchanged footer frames are no longer cleared and repainted, including hidden reasoning updates. Height changes still invalidate the layout correctly. #29, fixes #27. Thanks @wooksong!
• Block uncertain writes when resuming. A checkpoint left during a write now requires inspecting the working tree before continuing, including older checkpoints that incorrectly recorded the write as read-only. Interactive /resume, --continue, and print-mode resume use the same protection; a completed, durably recorded write can resume normally. #33, fixes #7.
• Keep live terminal output out of scrollback. Footer redraws now reserve and clip terminal space correctly so partially streamed output and transient input do not accumulate as permanent history. The fix also handles height/width changes, narrow windows, Unicode character widths, and tab alignment. #34, fixes #28.

If your terminal renders East Asian ambiguous characters as two columns, set MOTIF_AMBIGUOUS_WIDTH=2; the default remains one column.

Update

npm install -g motifcode@0.3.4
motif version

Expected version: 0.3.4. The motifcode command is also available.

Validation

The combined functional changes passed 770 automated tests, including checkpoint safety, legacy journal handling, installed-CLI behavior, and terminal viewport regressions. The actual CLI was also exercised through a PTY and terminal emulator with a local HTTP fixture. These are offline regression checks, not a live-model speed or quality benchmark.

CI passed all five jobs on release commit 4570d635a6776c12ce281ddeed3aa607ed00a924. The release workflow passed a frozen install, typecheck, build, full test suite, and tool-schema lint, then published with npm provenance.

The published npm package also passed all 13 registry and installed-CLI/PTY smoke checks, with its bundle SHA-256 matching the release build and its provenance verified against the release tag, commit and workflow.

Scope

This is a patch release. The separate interactive task-feedback proposal #36 is not included. MCP, Claude/Codex skill interoperability, and serving-performance investigations remain separate work; this release does not claim faster model inference.

Full changelog: v0.3.3...v0.3.4