Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@docs/reference/troubleshooting.md`:
- Around line 453-456: Replace the fenced CLI code block tagged as ```bash with
a ```console block and add the prompt prefixes: change the two lines "openshell
term <sandbox-name>" and "tail -f /tmp/gateway.log" to include a leading "$ "
prompt so they read "$ openshell term <sandbox-name>" and "$ tail -f
/tmp/gateway.log"; ensure the fence opener becomes ```console and the fence
closer remains ```.

In `@src/lib/inventory-commands.ts`:
- Around line 129-130: The overlap warning message currently instructs users to
run the non-existent command "nemoclaw remove <sandbox>"; update the string to
instruct "Run `nemoclaw destroy <sandbox>`..." so the remediation command
matches the CLI; locate the string "Run `nemoclaw remove <sandbox>` on whichever
sandbox should stop polling, or rerun onboarding with the channel disabled." in
src/lib/inventory-commands.ts and replace "remove" with "destroy" (and adjust
any corresponding unit tests or snapshots that assert this message).

In `@src/lib/messaging-conflict.ts`:
- Around line 61-68: The current loop calls providerExists(providerName) and
swallows probe errors but still calls registry.updateSandbox(entry.name, {
messagingChannels: discovered }), which allows a transient probe failure to
persist an empty/partial list; instead, track if any providerExists call throws
(e.g., set a boolean probeError flag during the catch), and only call
registry.updateSandbox(entry.name, { messagingChannels: discovered }) when no
probe errors occurred (or when you have verified all probes completed
successfully); reference the existing symbols providerExists, discovered,
registry.updateSandbox, and entry.name and add a brief log when skipping the
update due to probe errors.

In `@src/nemoclaw.ts`:
- Around line 1062-1069: backfillAndFindOverlaps currently lets exceptions
bubble up from backfillMessagingChannels or findAllOverlaps which can break the
non-critical `nemoclaw status` command; wrap the calls in a try/catch in the
backfillAndFindOverlaps function, call backfillMessagingChannels(registry, {
providerExists: providerExistsInGateway }) and then findAllOverlaps(registry)
inside the try, and on any exception log or warn (using the existing logger) and
return an empty array so status remains usable; keep references to
backfillMessagingChannels, findAllOverlaps, registry, and
providerExistsInGateway so reviewers can locate the change.

---

Outside diff comments:
In `@src/lib/onboard.ts`:
- Around line 3193-3201: The code is persisting the raw UI selection
(enabledChannels) into registry.registerSandbox instead of the token-backed,
filtered list that actually got configured (e.g.,
activeMessagingChannels/activeMessagingBridgeSet); update the call to
registry.registerSandbox to set messagingChannels to the filtered active list
(ensure you use the same variable used for the bridge attachment logic, e.g.,
activeMessagingChannels or activeMessagingBridgeSet) and also pass that same
filtered list into the earlier conflict/status check so both persistence and
checks reflect the real configured channels rather than the raw UI selections.

---

Nitpick comments:
In `@docs/reference/troubleshooting.md`:
- Around line 449-465: The two source lines containing multiple sentences should
be split so each sentence occupies its own source line: break the line beginning
"Bot tokens for Telegram (`getUpdates`), Discord (gateway), and Slack (Socket
Mode) only allow one active consumer per token. If two NemoClaw sandboxes..."
into separate lines for each sentence, and likewise break the line starting "To
fix, remove the messaging channel from the other sandbox (`nemoclaw remove
<other-sandbox>` or rerun onboarding on it with the channel disabled). Current
NemoClaw warns at `nemoclaw onboard` time when another sandbox already has the
same channel enabled, but sandboxes created before that check was added may
still be in a conflict loop." into individual-sentence lines; ensure punctuation
is preserved and wrap each sentence on its own source line to satisfy the "one
sentence per line" guideline.

Verify each finding against the current code and only fix it if needed.

Duplicate comments:
In `@src/nemoclaw.ts`:
- Around line 1057-1065: Wrap the body of backfillAndFindOverlaps (which
requires ./lib/messaging-conflict and ./lib/onboard and calls
backfillMessagingChannels(registry, { providerExists: providerExistsInGateway })
and findAllOverlaps(registry)) in a try/catch so any exception
(require/probe/registry errors) is caught and the function returns an empty
array instead of throwing; log or noop the error as appropriate and ensure the
function signature and return type remain the same (still returning the result
of findAllOverlaps(registry) on success, [] on any failure).

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/lib/inventory-commands.ts`:
- Around line 119-151: The code reads the sandboxes snapshot into `sandboxes`
before calling `deps.backfillAndFindOverlaps()`, so any updates to a sandbox's
`messagingChannels` by that callback are not seen when computing `defaultEntry`
later; after `deps.backfillAndFindOverlaps()` returns (inside the if-block where
it's called), refresh or reassign `sandboxes` (or re-fetch via the appropriate
dependency) so the subsequent `const defaultEntry = sandboxes.find((sb) =>
sb.name === defaultSandbox)` and the
`deps.checkMessagingBridgeHealth(defaultSandbox, channels)` call use the
up-to-date sandbox state; specifically update the `sandboxes` variable or
recompute `defaultEntry` after backfill before running the bridge health checks.

In `@src/lib/messaging-conflict.ts`:
- Around line 17-20: The probe currently collapses execution failures into a
false "provider absent" signal; update the ConflictProbe.providerExists contract
and its callers so failures are distinct from a definitive false: change
providerExists to return boolean | "error" (or similar tri-state) and update
backfillMessagingChannels to treat the "error" result as an execution error
(skip or propagate and do not persist messagingChannels: []), not as provider
absent; also update providerExistsInGateway and other callers to return "error"
when the openshell/gateway command fails instead of returning false (this
applies to the providerExists usage around backfillMessagingChannels and the
related logic in the 51-76 region).

In `@src/lib/onboard.ts`:
- Around line 2702-2728: Compute the actual configured channels
(activeMessagingChannels) before the conflict check and use that instead of the
raw UI selection (enabledChannels) when calling findChannelConflicts and when
persisting to the registry; specifically, move the logic that derives
activeMessagingChannels (the filter that removes channels with missing/empty
tokens or that were skipped) above the block that calls
backfillMessagingChannels and findChannelConflicts, then pass
activeMessagingChannels into findChannelConflicts(sandboxName,
activeMessagingChannels, registry) and use activeMessagingChannels for the later
registry write (instead of enabledChannels) so warnings, aborts, and persisted
state all reflect the real attached providers.

---

Nitpick comments:
In `@docs/reference/troubleshooting.md`:
- Around line 449-465: The paragraph that starts with "Bot tokens for Telegram
(`getUpdates`), Discord (gateway), and Slack (Socket Mode) only allow one active
consumer per token." and the paragraph that ends with "Current NemoClaw warns at
`nemoclaw onboard` time when another sandbox already has the same channel
enabled, but sandboxes created before that check was added may still be in a
conflict loop." should be rewritten so each sentence occupies its own source
line (one sentence per line), preserving the code blocks and inline commands
like `nemoclaw status`, `openshell term <sandbox-name>`, `tail -f
/tmp/gateway.log`, and `nemoclaw <other-sandbox> destroy` exactly as-is. Ensure
no sentences are merged on the same source line anywhere in those paragraphs.

Verify each finding against the current code and only fix it if needed.

Duplicate comments:
In `@src/nemoclaw.ts`:
- Around line 1071-1074: The providerExists function currently treats any
non-zero exit from captureOpenshell(["provider","get", name]) as "absent", which
hides transport/auth/gateway errors and lets backfillMessagingChannels persist
messagingChannels: []; modify providerExists to inspect the failing result from
captureOpenshell (the returned object from captureOpenshell) and only return
"absent" when the output indicates a true NotFound/does-not-exist case; for any
other non-NotFound error (transport, auth, gateway, unexpected status) return
"error" so callers like backfillMessagingChannels can avoid persisting an empty
messagingChannels array and surface the failure instead.