In `@src/config/index.ts`:
- Around line 645-653: The current beads tracker warning uses a single message
that claims an "interactive epic selection" will be shown even in headless mode;
update the branch where config.tracker.plugin is 'beads'|'beads-bv'|'beads-rust'
and config.epicId is falsy to choose the warning based on config.showTui: if
config.showTui is true, keep the existing interactive-selection wording,
otherwise replace it with a headless-appropriate message (e.g., "No epic ID
specified for beads tracker; running headless so no interactive epic selection
available") so the warnings array reflects the actual runtime mode; modify the
check around config.epicId and push the appropriate string accordingly.

In `@src/setup/wizard.ts`:
- Around line 60-83: In formatTrackerUnavailableReason (PluginDetection),
restrict the beads-specific heuristics so they only run for beads trackers:
check plugin.id for beads family (e.g., plugin.id === 'beads' or plugin.id ===
'beads-rust' or plugin.id.startsWith('beads')) before evaluating plugin.error
for "directory not found" or "not available"; for the binary branch derive cli
via plugin.id === 'beads-rust' ? 'br' : 'bd' as before, otherwise skip those
branches and fall back to returning plugin.error or `${plugin.description} (not
detected)` to avoid misattributing non‑beads tracker errors.
- Around line 104-129: Wrap the per-tracker sequence (calling
instance.initialize(), instance.isReady(), and the optional instance.detect())
in a try/catch/finally so any thrown errors are caught, instance.dispose() is
always called in finally, and a detection entry is still pushed to detections
with available: false and error set to the caught error message (e.g.,
"unavailable" or error.toString()) and version left undefined; specifically
update the block around instance.initialize, instance.isReady, the
instanceAsDetectable.detect call, and the final detections.push to ensure errors
don't abort the wizard and dispose() always runs.

In `@src/setup/wizard.test.ts`:
- Around line 507-643: Tests rely on real environment checks so they can flake
when bd/br CLIs or a .beads folder exist; make tracker availability
deterministic by mocking the detection layer before calling runSetupWizard: stub
the function(s) used to detect beads trackers (e.g., detectBeadsTrackers or any
isCliAvailable/isTrackerInstalled helper and any fs.existsSync check for
".beads") to return "unavailable" for beads, and return available for "json" (or
mock the tracker registry/getAvailableTrackers used by runSetupWizard) so
capturedTrackerChoices always reflect the expected availability in these tests.

In `@src/setup/wizard.ts`:
- Around line 116-118: runSetupWizard currently initializes the detector with an
empty config so detection falls back to process.cwd(); change it to pass the
wizard's options.cwd into tracker setup: thread the cwd from runSetupWizard's
options into detectTrackerPlugins and call instance.initialize({ cwd }) before
detection (and/or call instanceAsDetectable.detect with the cwd if detect
accepts options) so tracker detection uses the intended project directory;
update references to detectTrackerPlugins, instance.initialize, and
instanceAsDetectable.detect to accept and propagate the cwd.

In @.github/workflows/ci.yml:
- Around line 154-162: Add an explicit guard around the coverage merge step:
check whether any files match the glob coverage-parts/*.lcov before invoking
lcov (using the existing MERGE_ARGS loop), and if none are found either skip the
lcov run or fail with a clear error message so lcov is not invoked with no
--add-tracefile args; also include "empty" in the lcov --ignore-errors list
(i.e., --ignore-errors inconsistent,empty) and remove or avoid the 2>/dev/null
redirection during development so lcov errors are visible.

In @.github/workflows/ci.yml:
- Around line 172-175: The printed label "Coverage breakdown by batch (%
Funcs):" does not match the awk extraction (the pipeline grep "^All files"
coverage-output.txt | awk '{print "  " $4}'), which currently pulls the % Stmts
column; fix by either changing the label to "% Stmts" or modifying the awk
parsing to extract the % Funcs value (e.g., split on the '|' delimiter and trim
to the correct field) so the label and extracted column match; update the line
containing the exact strings "Coverage breakdown by batch (% Funcs):" and the
grep/awk pipeline accordingly.