Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@services/workspace_resolver.py`:
- Around line 337-345: The code currently calls json.loads(row["value"]) into cd
and then passes cd to _determine_project_for_conversation which assumes a dict;
guard against non-dict JSON by checking the type of cd after json.loads (e.g.,
isinstance(cd, dict)); if cd is not a dict, log a warning referencing cid and
the actual type/value and continue (so assemble_workspace_tabs/alias inference
won't crash), then only call _determine_project_for_conversation(cd, ...) when
cd is a dict.

In `@services/workspace_tabs.py`:
- Around line 127-131: The code currently swallows JSON decode failures inside
_try_loads_kv_value and only logs SchemaError later (e.g., in the Bubble parse
path that logs "Failed to parse Bubble from bubbleId:%s"), so malformed JSON is
silently skipped; fix by surfacing JSON decoding errors at the bubble/composer
call sites: replace the single call to _try_loads_kv_value(...) with an explicit
json.loads(...) wrapped in try/except JSONDecodeError (or change
_try_loads_kv_value to return the decode exception) and when a JSONDecodeError
occurs call _logger.warning including the context keys (bubbleId or
composerData), the raw value, and the exception, then skip processing; apply the
same pattern for the composerData parse site referenced in the comment so decode
errors are logged for both locations.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@services/workspace_tabs.py`:
- Around line 117-126: The json.loads(row["value"]) call in
services/workspace_tabs.py can raise TypeError when row["value"] is None (NULL)
and ValueError for other invalid inputs, but the except only handles
json.JSONDecodeError; update the parsing logic around parsed =
json.loads(row["value"]) (the block that logs "Failed to decode Bubble...") to
also catch TypeError and ValueError (or alternatively guard with an explicit
None check before calling json.loads) and include the same warning/logging
behavior and continue, ensuring the failure mode matches the previous
_try_loads_kv_value behavior.
- Around line 195-204: The JSON parse for composer data can raise TypeError when
row["value"] is None; update the parsing in the try/except around
json.loads(row["value"]) (the block referencing composer_id and row["value"]) to
guard against None by either checking row["value"] is not None before calling
json.loads or by expanding the except to catch TypeError alongside
json.JSONDecodeError, and keep the existing _logger.warning and continue
behavior when parsing fails.

In `@tests/test_parse_failure_logging.py`:
- Line 140: The test unpacks a returned tuple from assemble_workspace_tabs into
payload and status but never uses payload, triggering Ruff RUF059; change the
unused variable to a conventional ignored name (e.g., _payload or _) in the test
calls to assemble_workspace_tabs (the invocations around payload, status =
assemble_workspace_tabs(...) at the two occurrences) so only the used status is
kept and the linter warning is resolved.

---

Outside diff comments:
In `@services/workspace_resolver.py`:
- Around line 337-344: The current except block catches a bare Exception after
calling json.loads(row["value"]); replace it with a specific exception tuple to
only handle JSONDecodeError, TypeError, and ValueError (e.g., except
(json.JSONDecodeError, TypeError, ValueError) as e) so only relevant decode/type
errors are caught, and keep the existing _logger.warning(..., cid, e) and
continue logic intact; ensure json.JSONDecodeError is referenced correctly (or
imported from json.decoder) if your Python version requires it.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@services/workspace_tabs.py`:
- Around line 117-118: Replace the silent skip for rows with a None payload by
emitting a warning that includes identifying row metadata: instead of just "if
row['value'] is None: continue", call the module logger (e.g., logger.warning or
self.logger.warning) and include row.get("key") or other identifying fields and
the full row (or its id) in the message, then continue; apply the same change to
the second occurrence around the other None-check so malformed NULL KV rows are
logged for observability.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Outside diff comments:
In `@services/workspace_tabs.py`:
- Around line 123-126: The JSON decoding in services/workspace_tabs.py currently
only catches json.JSONDecodeError around json.loads(row["value"]); change this
to use the existing _try_loads_kv_value helper (which already handles
JSONDecodeError/TypeError/ValueError) or expand the except clause to also catch
TypeError and ValueError so non-string SQLite payloads don't raise and break
/tabs; update both Bubble and Composer json.loads sites and ensure the warning
log still includes context (row key/id) as before.