Repository navigation
fix(muse): Muse Code 1.4.8's argument hint, skill suggestions and skill rows in the slash palette - #448
Conversation
…uggestions verifies Muse 1.4.8 paints an argument hint after a complete command, in the bottom rule's colour, and lists built-in skills under a command they prefix. The draft read now drops a trailing [ ] run in the rule's colour, and accepts several suggestion rows when the first, bold one is the typed command; a moved highlight still fails the verify. Four live captures (2026-10-15, visible screen, cwd swapped for the sandbox path). Canary on muse 1.4.8: idle, drafts 15/15, narrow 3/3, start-exit 4/4, bar 3/3 with the model, effort and resume pickers on screen. Co-Authored-By: Claude <noreply@anthropic.com>
Muse 1.4.8 lists skills in the / palette with tags between the command and the gap (' /plan · built-in · default skill On an explicit…'), so the one-suggestion rule saw no suggestion and the draft read polluted. Tag runs may now follow the command. One live capture. Live readback: /plan verifies; canary muse: idle, drafts 15/15, start-exit 2/2, bar 3/3.
Co-Authored-By: Claude <noreply@anthropic.com>
|
Thanks @GGGODLIN, this lands in 1.22.0. Your two commits are kept with their hashes, and the changelog credits you. I labelled your five captures for the screen corpus. One question: how were muse--palette-hint-moved.txt and the Muse resume-picker -moved files from #446 made, a real Down press on the pane or an edit of the first capture? The README rows for them say "after Down". |
Merge AltanS/collie v1.22.0 (f2f4ef6) with the Send Protocol, Surface readers, hook holds, durable sessions and Themes v2. Keep fork 2.5.0 and the Gaddi brand, updater, descriptions and recap, retell, reply deliverables, local links, attention switchers, operator commands and skills, image editing, Taiwanese speech conversion, tablet layout, host clipboard and Island integration. Take upstream's integrated AltanS#446, AltanS#447, AltanS#448 and AltanS#452 captures and readers, including the opencode wrapped-body allowance and agy inline confirmation guard. Preserve the pending AltanS#449/AltanS#453 bars, Grok and Muse marks, Muse catalog and bundled skills, OMP paste support, and the 9dde2b2 Tasks dock in Chat and Terminal. Compose quoting with growing durable replies and keep the cross-host switcher's address beside its new badge. Route account-switch keys through the upstream key gate. Keep historical fork changelog sections unchanged; merge upstream 1.22.0 into Unreleased without duplicating already shipped fixes. Adopt upstream's Claude/opencode verification ledger as directed, without widening Send profile claims. Combine the existing Windows skip budgets by named tests. Root/web typechecks, lint, build and both unit suites pass; real-agent and dev-lane PWA acceptance remain with the coordinating session. No push, release or flake.lock change. Co-Authored-By: Claude <noreply@anthropic.com>
|
Both |
Two Muse Code 1.4.8 slash-palette shapes the draft read got wrong, one commit each. In both, a slash command sent from the phone did not verify, so the send was refused.
fix(muse)argument hint and skill suggestions❯ /model [<model-id>]), in the bottom rule's colour, and lists built-in skills under a command they prefix (/resumeoverresume-claude,resume-codex). The hint read as typed text, and more than one suggestion row read as an unknown popup[ ]run in the rule's colour, and accepts several suggestion rows when the first, bold one is the typed command; a moved highlight still fails the verify, since Enter runs the highlighted rowfix(muse)built-in skill row/palette with tags between the command and the gap (/plan · built-in · default skill On an explicit…), so the one-suggestion rule saw no suggestion and the draft read was pollutedChecked against a real agent on the fork build that carries these commits along with the fork's other changes, not on this branch over
mainalone: Muse Code 1.4.8 in a throwaway Herdr session on 2026-10-15, with the harness canary through drivers that are not part of this PR: idle, drafts 15/15, narrow 3/3, start-exit 4/4./planread back live as a verified draft; the captures pin/modelunder its hint,/resumeover its skill suggestions, and the same with the highlight moved off/resume, which does not verify.Captures. Five, each with its fixtures README row:
muse--palette-hint.txt,muse--palette-hint-skills.txt,muse--palette-hint-moved.txt,muse--palette-typed-arg.txt,muse--palette-skill.txt, all visible-screen reads. The statusline's shortened cwd became/private/tmp/collie-muse-sandboxat the same width and its model label was renamedmini. Nothing was submitted while capturing.No screenshots: nothing new is drawn; the change is which drafts verify.
Base branch — see CONTRIBUTING.md:
main— there is one line of development, andmainis itChecks
bun run typecheckat the root and inweb/exited 0;bun run lintexited 0;bun test ./bridge ./cli ./scripts8539 pass, 70 skip, 0 fail;cd web && bun run test416 of 416 files, 26210 tests pass. With each commit's fix taken out and only its tests left, onmain8629ecb: 2 failures for the hint and suggestion rows, 1 for the skill row (its tests need the first commit under them). Not run:cd web && bun run e2e, the Windows suites, and a live agent run on this branch overmainalone.- **A slash command sent to Muse Code 1.4.8 verifies under its argument hint and skill suggestions.** The grey hint read as typed text, and a command with skill suggestions under it read as an unknown popup, so the send was refused.- **A Muse Code built-in skill row in the `/` palette parses.** Its tags between the command and the description hid the suggestion, so `/plan` sent from the phone did not verify.