Repository navigation
fix: make lich close read a name the way lich send does - #227
Merged
Merged
Conversation
findSession matched a card's label and a session's roster name in the same pass, so a workspace where one string is both — a label here, another session's roster name there — answered "names 2 sessions" and refused to close anything. `lich send` resolves the same collision: it tries the labels first and only falls back to the roster names when none matched, and the comment here already claimed to do that. Two passes, and the two surfaces agree about which session a name addresses. The check the removal path skips is now documented rather than silent: a status git cannot read costs the sentence written for a dirty checkout, never the safety, because `git worktree remove` without --force refuses one anyway.
internal/cli declared its own "lich" constant beside relay.MCPServerName, which exists precisely because more than one package needs it and the MCP server already depends on that package. One copy, so a rename cannot leave half the product answering to the old name.
Three comments had drifted from what they describe, and the release notes carried a heading twice: - agentplugin.Installed claimed to shell out to a CLI and to be asked once per errand. It reads a file for Claude Code and opencode, runs a CLI for Codex and Crush, and the relay asks HasTools before delivering and Installed after — two uncached questions per errand, not one. - delegate-prompt's list claimed to mirror providers.AcceptsMCPServer. Since the plugin started carrying lich's operations into opencode and Crush that is no longer the same question, and the menu keeps the fixed list on purpose: what the plugin reports is true of an install, not of a session started before it, and the command works in all four. The comment says that now. Its test's name said "where the sender has none", which was the same stale claim; the assertions are untouched. - CHANGELOG's [Unreleased] had two "Added" headings, and the release job copies the whole section verbatim — so the published notes would have shown both. Merged, entries and their order unchanged.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Six of the seven medium findings from the
[Unreleased]changelog review. The relay split (internal/relay/relay.gois 992 lines against the project's 800 ceiling) is deliberately left out — it is a large no-logic diff and it collides with work in flight on that file, so it comes as its own PR.What changes
lich closeresolves a name the waylich senddoes.findSessionmatched a card's label and a session's roster name in one pass. A workspace where one string is both — a label on one card, another session's roster name — answered"foo" names 2 sessionsand closed nothing, whilelich send fooresolved it fine:relay.resolvetries the labels first and only falls back to the roster names when none matched, giving the label the tie on purpose. Two surfaces of one product disagreeing about which session a name addresses. Now the same two passes, with a test pinning the collision.The skipped
WorktreeDirtyerror is documented instead of silent. No data is at stake —git worktree removewithout--forcerefuses a dirty checkout on its own — so a status git cannot read costs lich's own wording and nothing else. The comment says so.One constant for the MCP server name.
internal/clideclared its own"lich"besiderelay.MCPServerName, which exists precisely because more than one package needs it.Three comments that had stopped being true:
agentplugin.Installedclaimed to shell out to a CLI and to be asked once per errand. It reads a file for Claude Code and opencode, runs a CLI for Codex and Crush, and the relay asksHasToolsbefore delivering plusInstalledafter — two uncached questions per errand. Collapsing them into one would change the relay'sPluginsinterface to save a call nobody has measured, so the comment is what changes.delegate-prompt's kind list claimed to mirrorproviders.AcceptsMCPServer. Since feat: give opencode and Crush sessions lich's own operations #222 that is no longer the same question: the plugin carries lich's operations into opencode and Crush too. The menu keeps the fixed list deliberately — what the plugin reports is true of an install, not of a session started before it, and the shell command works in all four — and the comment now says that rather than pointing at a mirror that moved. No behaviour change, so no changelog entry; its test's name carried the same stale claim and was reworded, assertions untouched.CHANGELOG.md's[Unreleased]had two### Addedheadings. The release job copies the section verbatim, so the published notes would have shown both. Merged; entries and their order unchanged.Test plan
gofmt -l .clean,go vet ./...cleango test ./...green (new:TestCloseResolvesALabelBeforeAnotherSessionsRosterName)biome check .,vitest run(802 tests),tsc --noEmit,vite buildall green