TestDetectFlowTermInvalidFallsThrough — an unrecognized FLOW_TERM value is silently ignored and TERM_PROGRAM detection takes over.

TestDetectFlowTermOverride — FLOW_TERM with a valid backend value wins over TERM_PROGRAM but loses to ZELLIJ and kitty. Iterates over every valid back

TestDetectFromEnv verifies the TERM_PROGRAM → backend mapping. The Override knob and the ZELLIJ / kitty / FLOW_TERM checks have higher precedence and

TestDetectKitty verifies $KITTY_WINDOW_ID and $TERM=xterm-kitty both route to BackendKitty, and that kitty beats TERM_PROGRAM (kitty does not set TERM

TestDetectKittyBeatsFlowTerm — kitty's per-window markers beat FLOW_TERM, matching ZELLIJ's behavior. Rationale: if the user is inside kitty, that's w

TestDetectZellij verifies the ZELLIJ env var beats every other signal. zellij sets ZELLIJ in every shell it spawns, so its presence means the user is

A row whose next_wake_at can't be parsed must be skipped, not crash the whole scheduler pass — DueOwners is defensive against a corrupt value.

TestE2EFullRoundtrip exercises the full command surface in the order a user would hit it for a realistic session: init, add project, add task under th

TestEncodeCwd pins the empirical rule derived from ~/.claude/projects/* vs. the original cwd recorded in each dir's *.jsonl. `/`, `.`, and `_` each ma

TestEscapeAppleScriptString covers the embedded helper directly so regressions in the script-path escape don't slip through.

TestFocusSessionAcrossOSWindows — the match scan walks every OS window, not just the first.

TestFocusSessionEmptyID short-circuits without touching ps.

TestFocusSessionEmptyID short-circuits without invoking zellij.

TestFocusSessionEmptyID short-circuits without invoking kitty.

TestFocusSessionEmptyID short-circuits on empty UUID without touching ps.

TestFocusSessionFocusError surfaces focus-pane-id errors as a backend failure (distinct from "no match found").

TestFocusSessionFocusError — a focus-window failure surfaces as a backend error, not as "no match".

TestFocusSessionListPanesError surfaces zellij CLI errors.

TestFocusSessionLsError surfaces a `kitty @ ls` failure as a backend error (distinct from "no match found"), wrapped with the command context.

TestFocusSessionMalformedJSON — protect against zellij output drift.

TestFocusSessionMalformedJSON — guard against kitty output drift.

TestFocusSessionMatchesAndFocuses verifies the happy path drives osascript with the right tty against Terminal.app's dictionary.

TestFocusSessionMatchesAndFocuses verifies the happy path: list-panes JSON contains a non-plugin pane whose pane_command runs claude with the target U

TestFocusSessionMatchesAndFocuses verifies the happy path: a foreground process under a kitty window runs claude with the target UUID, and FocusSessio

TestFocusSessionMatchesAndFocuses confirms the happy path: ps returns a claude row carrying the UUID, FocusSession extracts the tty, drives osascript

TestFocusSessionNoMatch — list-panes contains no claude with this UUID.

TestFocusSessionNoMatch — no foreground process under any window runs claude with the target UUID; Runner must not be called.

TestFocusSessionNoMatchInPS returns (false, nil) without touching osascript.

TestFocusSessionOsascriptError surfaces script-runtime failures distinct from "no match".

TestFocusSessionOsascriptError surfaces the error too — distinguishes "the focus mechanism broke" from "no match found" so the caller can log it.

TestFocusSessionPSError surfaces ps failures.

TestFocusSessionPSError surfaces the error to the caller instead of silently treating it as "no match". The caller can then decide whether to fall thr

TestFocusSessionResumeFlag — covers --resume in pane_command.

TestFocusSessionResumeFlag covers `--resume <uuid>` in argv.

TestFocusSessionResumeFlag covers the --resume variant alongside --session-id, since both forms appear depending on whether the tab is in bootstrap or

TestFocusSessionRoutesToITerm — when Detect() resolves to iTerm, FocusSession invokes the iterm backend.

TestFocusSessionRoutesToKitty — Override=Kitty hits the kitty backend. The kitty backend uses `kitty @ ls` JSON rather than ps/tty, so the stub return

TestFocusSessionRoutesToTerminal — Override=Terminal hits the terminal backend.

TestFocusSessionRoutesToZellij — Override=Zellij hits the zellij backend.

TestFocusSessionScriptMissReturnsFalse — ps found a match but the AppleScript walked all tabs without finding the tty.

TestFocusSessionScriptMissReturnsFalse covers the case where ps found a match but iTerm2's AppleScript walked all sessions without finding the tty (e.

TestFocusSessionSkipsNoControllingTTY ignores claude rows whose tty is "??".

TestFocusSessionSkipsNoControllingTTY ignores claude rows whose tty column is "??" — a backgrounded claude with no controlling terminal has nothing fo

TestFocusSessionSkipsPluginPanes — plugin panes (tab-bar, status-bar) have null pane_command and must not match.

TestFocusSessionSkipsWindowsWithoutClaude — windows whose foreground_processes do not include the substring `claude` are skipped without running the r

TestFocusSessionUUIDCaseInsensitive — UUID match is case-insensitive.

TestFocusSessionUUIDCaseInsensitive — UUID match is case-insensitive, matching the iterm / zellij behaviour.

TestFocusSessionUUIDCaseInsensitive verifies the UUID match is case-insensitive (matches the existing liveClaudeSessions normalization behavior).

TestHarnessForSpawn_AllPathsLandOnClaudeToday smoke-tests every branch of harnessForSpawn under the single-harness registry (claude is the only adapte

TestHarnessForTask covers the column → adapter lookup. NULL and empty resolve to claude+nil error (back-compat). Unknown non-empty names resolve to ni

TestHookSessionStartRequiresSkillInvocation pins the invariant that the injected additionalContext explicitly instructs the session to invoke the flow

TestHookSessionStartUnboundEmitsAmbientHint pins the contract for ad-hoc sessions (no task carries the current $CLAUDE_CODE_SESSION_ID): the hook must

TestHookUserPromptSubmitBoundEmitsAnchor pins the bound-session contract: when the current $CLAUDE_CODE_SESSION_ID belongs to a task, the hook injects

TestHookUserPromptSubmitUnboundIsNoOp pins the unbound contract: with no task carrying $CLAUDE_CODE_SESSION_ID (var unset, or set but unmatched), the

TestIsAccessibilityDenied unit-tests the pattern matcher in isolation.

TestIsBackground verifies the bg-mode selector. $FLOW_TERM=bg means "spawn as a terminal-free background agent" — distinct from the terminal backends

TestLatestReleaseCacheTTLBoundary verifies the cache is treated as stale once it exceeds versionCacheTTL.

TestLatestReleaseFetchesAndCaches verifies the happy path: no cache → HTTP fetch succeeds → tag returned and cache written.

TestLatestReleaseHonorsCache verifies a fresh cache short-circuits the HTTP path: even if the network is rigged to fail, a fresh cache returns the cac

TestLatestReleaseSilentOnNetworkFailure pins the silent-failure contract: every error path returns "" so the SessionStart hook never blocks or surface

TestLatestReleaseSilentOnNon200 covers rate-limited responses.

TestLaunchCmd_PreservesByteIdentity verifies the LaunchCmd output is byte-identical to the legacy command string that the pre-refactor do.go was produ

TestLiveSessionsIncludesBackgroundAgents: the [live] detection used by `flow list` must count background agents (which don't show in the ps scan) as l

TestLiveSessions_BareClaude — a bare `claude` invocation (no --session-id, no --resume) contributes no UUID. Detection requires the flag.

TestLiveSessions_PSError surfaces the ps failure to the caller. Callers in app/ swallow this silently — that's their policy. The harness layer just re

TestLiveSessions_ParsesPSOutput verifies the ps-grep heuristic against a representative process list. Lines without "claude" are skipped; lines with t

TestLoadCacheCorrupt verifies the binding constraint that LoadCache never errors: a corrupt cache file must yield an empty, usable cache (not nil, non

TestMain defaults claude.BGCommandRunner to an empty background-agent registry for the whole app test package. Without this, every show / list / do ov

TestMigrationAddsAutoRunColumns verifies the autonomous-run bookkeeping columns land on the tasks table and round-trip through ScanTask.

TestMigrationDeduplicatesSessionIDs simulates Anshul's bug: a DB where two non-archived tasks share the same session_id (could happen via the now-remo