Repository navigation
Sync fork with upstream smtg-ai/claude-squad (1.0.17 → 1.0.20) - #1
Draft
BippleDops wants to merge 12 commits into
Draft
BippleDops wants to merge 12 commits into
BippleDops wants to merge 12 commits into
Conversation
## Summary Fixed completion scripts and error messages to use the actual invoked binary name instead of hardcoding 'claude-squad'. - Fish, bash, and zsh completion scripts now work with custom binary names (e.g., `--name cs`) - Error messages and help text now reference the actual binary name - Completion generation dynamically extracts the binary name from `os.Args[0]` ## Problem When users installed claude-squad with a custom name (e.g., `install.sh --name cs`), the generated completion scripts still referenced `claude-squad`, making them non-functional. ## Solution Modified `main.go` to: 1. Extract the actual binary name from `os.Args[0]` using `filepath.Base()` 2. Dynamically set `rootCmd.Use` to this binary name before Cobra generates completions 3. Update error messages and version output to use the actual binary name This ensures all three completion shells (fish, bash, zsh) work correctly regardless of how the binary is named.
## Summary - Add Shift+J / Shift+K keybindings to move instances up/down in the list - New order is persisted to state.json immediately ## Test plan - [ ] Press `K` (Shift+K) to move selected instance up - [ ] Press `J` (Shift+J) to move selected instance down - [ ] Verify boundary cases (top/bottom) are no-ops - [ ] Restart `cs` and verify reordered list is preserved
…smtg-ai#285) ## Summary - Move `SetStatus(Paused)` before `combineErrors` check in `Pause()` so the state transition always happens after worktree removal, even when non-critical errors occurred - Skip already-paused instances in `metadataUpdateDoneMsg` handler to prevent the polling loop from overwriting `Paused` back to `Running`/`Ready` Fixes smtg-ai#284 ## Test plan - [ ] Start a session, press `c` to checkout — should pause normally - [ ] Simulate IsDirty failure (e.g., externally delete worktree dir, then press `c`) — should still transition to Paused - [ ] Verify metadata polling doesn't revert Paused status back to active --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes smtg-ai#280 ## Summary The 500ms metadata tick in `tickUpdateMetadataCmd` calls `ComputeDiff()` for every active instance and stores each full diff string in `DiffStats.Content`, but the diff pane only ever renders one instance at a time. With N sessions this keeps N copies of full git diff output resident for zero UI benefit. This PR keeps the full diff only for the currently-selected instance; every other instance gets a lightweight `git diff --numstat` summary that populates only `Added`/`Removed`. The list view's `+X/-Y` counters are unaffected because they only need the totals. ## Changes - `session/git/diff.go`: add `DiffNumstat()` which runs `git diff --numstat <base>` and returns a `*DiffStats` with `Content` empty. Parsing is extracted into `parseNumstat` so it's unit-testable. - `session/instance.go`: add `ComputeDiffNumstat()` mirroring `ComputeDiff()`. - `app/app.go`: `tickUpdateMetadataCmd` now takes the selected instance and dispatches full-diff vs numstat accordingly. Both call sites pass `m.list.GetSelectedInstance()`. - `session/git/diff_test.go`: unit tests for `parseNumstat` covering empty output, binary files (`-\t-`), multi-file sums, tabs-in-path, and trailing newlines. ## Trade-offs Switching the selected instance incurs up to one tick (~500ms) before the diff pane repopulates with the full content — previously, the full diff was always resident for every instance so the switch felt instant. This is an intentional trade for bounded memory. ## Test plan - [x] `go test ./...` passes locally - [x] `gofmt -w .` clean - [x] Manually verify the diff tab renders correctly for the selected instance - [x] Manually verify preview / terminal tabs behave unchanged --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
…ry (smtg-ai#302) ## Summary - `github.com/go-git/go-git/v5` was declared as a direct dependency in `go.mod` but is not imported anywhere in the codebase — all git operations use `exec.Command` to shell out to the system `git` binary. - `go-git`'s transitive dependency `github.com/go-git/go-billy/v5` was pinned at `v5.6.2`, which is affected by a security advisory requiring `v5.9.0` or later. - Running `go mod tidy` removes `go-git` and its full transitive closure (including `go-billy`, `go-winio`, `ProtonMail/go-crypto`, `cloudflare/circl`, `golang.org/x/crypto`, `golang.org/x/net`, and others), resolving the advisory by eliminating the dependency entirely. ## Changes - `go.mod`: removed `go-git/go-git` direct dep and 14 indirect deps that came with it - `go.sum`: updated accordingly ## Test plan - [ ] `go build ./...` passes - [ ] `go vet ./...` passes - [ ] `gofmt -l .` reports no files - [ ] Existing tests pass (`go test ./...`) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
## Problem
Sessions created from a **pre-existing branch** never get a base commit
recorded, which makes the TUI log an error roughly twice a second for
the lifetime of the session:
```
WARNING app.go:253: could not update diff stats: git command failed:
fatal: ambiguous argument '': unknown revision or path not in the working tree.
Use '--' to separate paths from revisions, like this:
'git <command> [<revision>...] -- [<file>...]'
(exit status 128)
```
`setupNewWorktree` resolves `HEAD` and assigns `g.baseCommitSHA`, but
`setupFromExistingBranch` returns without ever setting it — on both the
local-branch and remote-tracking paths:
```go
// Create a new worktree from the existing local branch
if _, err := g.runGitCommand(g.repoPath, "worktree", "add", g.worktreePath, g.branchName); err != nil {
return fmt.Errorf("failed to create worktree from branch %s: %w", g.branchName, err)
}
return nil // baseCommitSHA is still ""
```
`diff.go` then runs `git diff ""`, which exits 128 on every refresh:
```go
content, err := g.runGitCommand(g.worktreePath, "--no-pager", "diff", g.GetBaseCommitSHA())
```
The instance persists to `state.json` with `"base_commit_sha": ""`, so
restarting does not clear it — the session is permanently stuck
producing failed subprocess calls, and its diff tab stays empty.
## Reproduction
1. In a repo with an existing branch, create a session and select that
branch as the source (the `is_existing_branch` path added in smtg-ai#262).
2. `cat ~/.claude-squad/state.json` — the instance shows
`"base_commit_sha": ""`.
3. Watch the log at the path printed by `cs debug`; the warning above
repeats continuously.
## Fix
Both paths in `setupFromExistingBranch` now resolve the worktree `HEAD`
and store it, so "base" means the same thing it already means for new
branches: the commit the session started from.
## Testing
Added `TestSetupFromExistingBranch_RecordsBaseCommit`, which builds a
real repo, runs `Setup()` with `isExistingBranch: true`, and asserts the
recorded SHA matches the branch tip.
Verified it fails without the production change:
```
--- FAIL: TestSetupFromExistingBranch_RecordsBaseCommit
worktree_ops_test.go:109: GetBaseCommitSHA() = "", want "b93ed4f66fba0eb41fd3ef9475c686e4671635ad"
```
`go test ./...` passes and `gofmt -l .` is clean.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude <noreply@anthropic.com>
…ai#322) Fixes smtg-ai#216. ## The bug When the tmux server goes away — reboot, crash, `tmux kill-server` — every session goes with it, but the instances in `state.json` are still recorded as `Running`. On the next start `cs` tries to attach to sessions that no longer exist. The problem is that `tmux attach-session` against a missing session **forks successfully** and then exits 1 on its own. So `pty.Start` returns no error and `TmuxSession.Restore` reported success while attached to nothing: ```go func (t *TmuxSession) Restore() error { ptmx, err := t.ptyFactory.Start(exec.Command("tmux", "attach-session", "-t", t.sanitizedName)) if err != nil { // only fires if the *fork* fails return fmt.Errorf("error opening PTY: %w", err) } ... ``` The instance then sits there `Running` with every `capture-pane` against it failing, spamming the TUI and the log: ``` ERROR tmux.go:238: error capturing pane content in status monitor: error capturing pane content: exit status 1 ERROR app.go:987: error capturing pane content: exit status 1 ``` The preview pane stays blank and there is no way to get the session back — `r` is gated on `Paused`, so the only exits are killing the instance or `cs reset`. Several reports in smtg-ai#216 match this shape ("happens when I come back to sessions after a few days"); other reports in that thread look like a different, new-session failure, which this PR does not touch. ## The fix Three changes: **1. `Restore` checks that the session exists** and returns a new `ErrSessionNotFound` sentinel when it does not, so callers can tell "session is gone" apart from "PTY failed". **2. Restoring an instance whose session is gone parks it as `Paused`** instead of leaving it `Running`-but-broken. The worktree and branch are still on disk, so the existing `Resume` path rebuilds the session from the same `Program`. Returning an error from `Start` here would be worse than useless: `Storage.LoadInstances` aborts on the first `FromInstanceData` failure, so a single dead session would hide *every* other instance. **3. `Resume` no longer runs `gitWorktree.Setup` when the worktree is still valid on disk.** `setupFromExistingBranch` does `worktree remove -f` + `os.RemoveAll` before re-adding, which is correct after a normal `Pause` (the directory is already gone), but would throw away uncommitted work in an instance paused by the new recovery path above. This is the change that makes recovery non-destructive. ## Verification Reproduced against a real `state.json` with five instances whose tmux server had died in a reboot, loading them through `Storage.LoadInstances` before and after: ``` before: photos tagger status=Running previewErr=error capturing pane content: exit status 1 after: photos tagger status=Paused previewErr=<nil> ``` All five then resumed from the TUI with `r` into working sessions, worktrees and uncommitted contents intact. Four tests added — `Restore` returning `ErrSessionNotFound` vs. attaching normally, and `Start(false)` parking a dead-session instance as `Paused` vs. restoring a live one as `Running`. `gofmt`, `go vet` and the full suite are clean. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
1 task
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
9 participants
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.
Summary
Brings this fork's
mainup to date withsmtg-ai/claude-squadmain(ce1ffb4, "Bump version to 1.0.20"). The fork was 0 commits ahead / 12 behind; this branch is exactlyupstream/main, so a "Rebase and merge" is a pure fast-forward with no merge commit.Equivalent alternative: use GitHub's Sync fork button on the repo page instead of merging this PR.
Verification
go build ./...— passgo vet ./...— passgo test ./...— pass (app, config, session, session/git, session/tmux, ui)Related
cursor/paperclip-cs-serve-rebased-65a7re-applies the fork's two owner commits (cs serveHTTP+SSE API, OpenTelemetry tracing) on top of this sync; it is based on this branch so it shows only those two commits.Before merging
mainidentical to upstream.