Wires the earlier runScript/stopScript deeplinks to actual CLI
tooling and adds the socket query needed to discover script UUIDs.

- Add "scripts" socket query resource keyed on worktreeID that
  returns each ScriptDefinition as id / kind / name / displayName
  plus a running marker derived from runningScriptsByWorktreeID.
- Extend `supacode worktree run` and `stop` with `--script/-s`
  so callers can target a specific script by UUID while the
  zero-argument form keeps firing the bare run / stop deeplinks
  for backward compatibility.
- Introduce `supacode worktree script list` (via a sibling
  WorktreeScriptCommand to satisfy SwiftLint nesting) that prints
  tab-separated id / kind / displayName rows and underlines any
  currently running script.
- Validate `--script` arguments as UUIDs at the CLI layer so a
  typo surfaces a helpful error instead of a generic deeplink
  parse failure.

Review follow-ups covering cross-worktree correctness, silent
success paths, the `-s` flag collision, and in-app docs.

- Resolve the target script directly from the worktree's
  @SharedReader RepositorySettings in runScript/stopScript
  deeplink handlers so cross-worktree CLI calls no longer miss
  a freshly loaded selection's scripts.
- Validate empty command, already-running, and not-running
  cases inline in the deeplink effects and surface them as
  alerts. Previously these hit silent `.none` guards in
  runNamedScript / stopScript and misled the CLI into ok:true.
- Normalize trailing slashes in the `scripts` socket query so
  `supacode worktree script list -w <path>` matches worktree
  IDs with or without a trailing slash, matching resolveWorktreeID.
- Rename the CLI short flag from `-s` to `-c` under `worktree
  run`/`stop` so it stops colliding with `-s` = `--surface`
  across the rest of the CLI. Long form `--script` is unchanged.
- Update CLIReferenceView, DeeplinkReferenceView, and the
  supacode-cli skill content with the new subcommand, option,
  and deeplink URLs.
- Extend the test suite with confirm-accepted round trip,
  socket-source responseFD storage, empty command, already
  running, and not-running-stop cases, and rewrite the existing
  script deeplink tests to seed scripts via @shared repository
  settings now that the reducer reads them from there.

Phase 3 verification caught a narrow but real silent-success
path in the script deeplink effects: if the target worktree is
removed between the initial `handleWorktreeDeeplink` check and
the confirmation-accept re-dispatch, both runScript and
stopScript handlers would return `.none` without setting an
alert, so the outer response flow would report `ok:true` to the
CLI even though nothing ran.

- Set a "Worktree not found" alert in the worktree-missing guard
  of both runScriptDeeplinkEffect and stopScriptDeeplinkEffect
  so the alert-diff check drives the socket response to ok:false.
- Extract a shared worktreeNotFoundAlert helper.
- Add stopScriptSocketDeeplinkSendsErrorWhenNotRunning as a
  regression guard exercising the socket responseFD path for
  stop failures, which had no coverage.

Final polish pass after Phase 3 self-review.

- Collapse the inline "Worktree not found" AlertState block in
  handleWorktreeDeeplink into the worktreeNotFoundAlert helper
  introduced for the script deeplink effects.
- Sanitize tab, newline, and carriage return characters in the
  `worktree script list` output so a script name containing
  those characters cannot corrupt the tab-separated columns
  when the output is piped into xargs/cut/awk.
- Expand CLISkillContent's shared/codex sections to document
  the new `worktree script list` subcommand, the `-c`
  `--script` flag, and the tab-separated list output format
  so agents reading the skill pick up the new surface.