SpawnBackground must run claude in the task's work_dir (a bg session begins there and keys its transcript/CLAUDE.md to it), not flow's cwd.

If the captured short id isn't in the agents registry, capture failed — surface an error rather than recording a phantom id.

TestSpawnTabBasicNoEnv verifies the two-call argv sequence with no env vars: 1) zellij action new-tab --name <title> --cwd <cwd> 2) zellij action writ

TestSpawnTabBasicNoEnv verifies the two-call argv sequence with no env vars: 1. RunnerOutput: kitty @ launch --type=tab --tab-title=<t> --cwd=<cwd> (

TestSpawnTabBasicNoEnv — happy path with no env vars. WriteScript is called once with the expected body shape, OpenURL with the correct URL-encoded wa

TestSpawnTabEnvVarsSorted verifies env vars are emitted alphabetically, each value shell-quoted, all space-separated, before the command. This matches

TestSpawnTabEnvVarsSorted verifies env vars are emitted alphabetically, each value shell-quoted, all space-separated, before the command. This matches

TestSpawnTabEnvVarsSorted — env vars sorted alphabetically and shell-quoted into the `exec env …` line. Exact parity with the zellij/iterm contract.

TestSpawnTabFlattensEmbeddedNewlines verifies that any `\n` inside `command` is replaced with a space before write-chars. Without this, zellij types e

TestSpawnTabFlattensEmbeddedNewlines — embedded `\n` in command must be replaced with a space before send-text, same reasoning as zellij (PTY would in

TestSpawnTabLaunchEmptyOutput — if kitty @ launch returns empty stdout, SpawnTab must fail rather than send-text to id:"" (which kitty would interpret

TestSpawnTabLaunchErrorShortCircuits — an error from `kitty @ launch` must be wrapped (with the remote-control hint) and send-text must NOT be attempt

TestSpawnTabNoEnvVars covers the env-prefix branch when envVars is nil — the line should still cd and run the command, just with no VAR=value assignme

TestSpawnTabNoEnvVars covers the env-prefix branch when envVars is nil — the line should still cd and run the command, just with no VAR=value assignme

TestSpawnTabNoTitleOmitsOSC2 — empty title skips the OSC 2 line entirely (no stray printf in the script).

TestSpawnTabPassesThroughUnknownErrors verifies that non-Accessibility errors from osascript are returned unchanged — only the specific permission-den

TestSpawnTabPropagatesNewTabError verifies an error from the new-tab call is returned and write-chars is NOT attempted.

TestSpawnTabPropagatesOpenURLError — when OpenURL fails, Runner is never invoked and removeScript cleans up the orphaned temp file.

TestSpawnTabPropagatesRunnerError — when Runner fails, removeScript cleans up the orphaned temp file and the error is returned unwrapped (for non-Acce

TestSpawnTabPropagatesWriteScriptError — when WriteScript fails, OpenURL and Runner are never invoked.

TestSpawnTabRoutesToGhostty asserts the ghostty Runner is the one called when Detect() resolves to BackendGhostty.

TestSpawnTabRoutesToITerm asserts the iterm Runner is the one called when Detect() resolves to BackendITerm.

TestSpawnTabRoutesToKitty asserts the kitty Runner+RunnerOutput pair is invoked when Detect() resolves to BackendKitty.

TestSpawnTabRoutesToTerminal asserts the terminal Runner is the one called when Detect() resolves to BackendTerminal.

TestSpawnTabRoutesToWarp asserts the warp Runner is the one called when Detect() resolves to BackendWarp.

TestSpawnTabRoutesToZellij asserts the zellij Runner is the one called when Detect() resolves to BackendZellij.

TestSpawnTabScriptShape verifies the AppleScript emitted to osascript targets Terminal.app, embeds env-var assignments before the command, uses cmd-T

TestSpawnTabScriptShape verifies the AppleScript emitted to osascript targets Ghostty, embeds env-var assignments before the command, sets the tab tit

TestSpawnTabSetsTitleViaOSC2 — when title is non-empty, the OSC 2 escape sequence is emitted as a printf line before cd.

TestSpawnTabWrapsAccessibilityError verifies that when osascript fails with a macOS Accessibility-denied error pattern, SpawnTab returns a wrapped err

TestSpawnTabWrapsAccessibilityError — when Runner returns one of the documented Accessibility-denied error patterns, the returned error string names "

TestStatsE2E exercises `flow stats` against a real (but temp) flow root: empty stats succeed, --card writes a file, and a garbage --since value return

TestTable_AlignedColumns verifies that the Table renderer pads columns to the widest visible cell across all rows.

TestTable_AlignedWithANSI verifies that colored cells align against uncolored cells of the same visible width — the bug that motivated dropping text/t

The targeted owner-mutation helpers must touch ONLY their own columns — never clobber in-flight tick bookkeeping (the dispatch/finish race fix).

Parser-level tests live in internal/harness/claude/transcript_test.go (the claude jsonl format is owned by the harness impl). The tests here exercise

TestTranscriptRetrospectiveBindFullOutput pins the bug this task fixes: a retrospective `flow do --here` bind stamps session_started at BIND time, whi

TestWriteCardPNG verifies that writeCardPNG writes a valid PNG file with sane dimensions that can be re-decoded.

UninstallSkill removes the skill directory entirely (SKILL.md plus VERSION sidecar plus anything else dropped alongside).

runSkipPermissions is the default SkipPermissionsRunner — execs `claude -p <prompt> --dangerously-skip-permissions`. Stdout/stderr are discarded becau