TestMigrationHarnessSurvivesSessionInvariantRebuild pins the fix for a bug where DBs upgrading from a pre-session-invariant version (e.g. v0.1.0-alpha

Regression guard: with FLOW_TERM unset (non-bg mode), flow show / list must NOT query the background-agent registry — no `claude agents` subprocess, n

TestOpenDBConcurrentDoesNotBusy pins that two parallel OpenDB calls on the same path don't race during schema setup. Without busy_timeout applied at o

TestOpenDBOnPreMigrationDB simulates an existing user upgrading from a pre-feat/playbooks flow.db: tasks table exists but lacks the kind and playbook_

TestOverrideBeatsEnv confirms the test escape hatch: setting Override pins the backend regardless of env vars, so individual tests can pin the dispatc

Each tick is a FRESH, sessionless run: the detached child's env must not carry CLAUDE_CODE_SESSION_ID from the parent, or the tick would resume the pa

A tick that finishes BEFORE the scheduler records its pid must not have its result clobbered. The scheduler dispatch write must touch only the running

An owner whose tick_pid looks alive (pid reuse after a reboot) but whose tick started long ago must NOT be skipped forever (gap 6) — the staleness cap

After dispatch advances next_wake_at into the future, an immediate second scheduler pass must not re-dispatch the same owner.

tick-due must skip (not crash, not launch) an owner with an unparseable --every — a defensive path since CreateOwner / direct writes don't validate th

A hand-triggered `flow owner tick --auto` must not stack a second tick on top of one already running (e.g. a scheduled tick, or an event-triggered tic

Tick prompts must reference the ABSOLUTE owner dir, so a headless run (cwd=work_dir) reads/writes the real charter+journal under $FLOW_ROOT instead of

`claude agents --json --all` also returns interactive (non-background) sessions. parseBackgroundAgents must drop them — flow's bg surfaces (bg_status,

parseSince unit tests exercise the date logic directly without needing to stand up a DB. Kept here next to the command that uses it.

TestReconcileAutoRunDeadPid: a 'running' row whose supervisor pid is no longer alive is reconciled to 'dead' on read.

TestReconcileAutoRunLivePidStaysRunning: a live pid keeps 'running'.

TestRenderCardPNG verifies that renderCardPNG returns a non-nil image with positive dimensions for both automation-present and automation-absent Stats

TestRendersEntriesWithoutTimestamp pins that entries lacking a `timestamp` field still render — the renderer never inspects timestamps, so a missing o

TestRendersWholeTranscript pins the no-cutoff contract: every user/assistant entry is rendered regardless of its timestamp, so a retrospective `--here

TestResolveTranscriptPathGlobFallback simulates the bg/worktree gotcha: the session ran with cwd under a git worktree but its jsonl is keyed under the

TestResolveTranscriptPathNotFound: neither deterministic nor glob hits.

TestResolveTranscriptPathPrefersDeterministic: when the cwd-encoded path exists, it wins — no glob needed.

ResumeBackground resumes the OLD id under --bg but, because --bg mints a fresh id, must return the NEW captured agent (history inherited).

TestScriptBodyExecReplaces — the final non-empty line is exec, so bash is replaced by the command process. No orphan bash, clean ^C.

TestScriptBodySelfDeletes — first non-shebang line is rm -- "$0". This is what makes the temp script disappear ~immediately on successful execution.

TestShellQuote is a sanity check on the local helper — same contract as iterm.ShellQuote.

TestShellQuote is a sanity check on the local helper — same contract as iterm.ShellQuote and terminal.ShellQuote.

TestShellQuote — same contract as iterm.ShellQuote / terminal.ShellQuote.

TestShellQuote — same contract as iterm/terminal/zellij.

TestShellQuote — same contract as iterm.ShellQuote / terminal.ShellQuote zellij.ShellQuote.

TestShellQuoteParity makes sure the re-exported helper matches every backend's implementation. All backends quote identically.

TestSkillEmphasizesCloseOutValue pins §4.7's framing of `flow done` as the load-bearing moment that persists the session's learnings — it triggers the

TestSkillHasAccessibilityErrorRecipe pins the §4.4 recipe for handling the macOS Accessibility error from the Terminal.app backend: name Terminal defi

TestSkillHasExplicitInvocationSection pins the §1a behavior: when the skill is invoked without a trigger phrase, it should describe its capabilities a

TestSkillHasUpgradeWorkflow pins the §4.15 upgrade procedure: the skill must know how to walk the user through replacing the binary per the README at

TestSkillInstallIsIdempotent verifies a second install --force does not duplicate either hook entry. Past regressions append duplicates silently; pin

TestSkillInstallPreservesUnrelatedHooks pins the safety property: installing flow's UserPromptSubmit hook MUST add flow's entry while leaving other co

TestSkillInstallSkipHook leaves settings.json untouched when --skip-hook.

TestSkillInstallWritesBothHooks verifies install wires up BOTH the SessionStart hook and the UserPromptSubmit drift/close-out hook into ~/.claude/sett

TestSkillNoCliCoachingInUserFacingLabels pins the rule that the skill must not put literal `flow ...` invocations inside AskUserQuestion option labels

TestSkillUninstallRemovesBothHooks verifies uninstall strips both the SessionStart and UserPromptSubmit entries, leaving an empty (or absent) hooks ma