backends/advanced/src/advanced_omi_backend/workers/transcription_jobs.py (3)

88-90: Remove extraneous f prefix — no placeholders in this string.

Static analysis (Ruff F541) correctly flags this. Same issue recurs on Lines 1022, 1059, and 1071.

🧹 Proposed fix

-            logger.info(
-                f"🎤 Speaker recognition disabled, using original speaker labels"
-            )
+            logger.info(
+                "🎤 Speaker recognition disabled, using original speaker labels"
+            )

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/workers/transcription_jobs.py`
around lines 88 - 90, The logged message uses an unnecessary f-string prefix (no
placeholders) which triggers Ruff F541; remove the leading f from the string
literal in the logger.info call(s) inside transcription_jobs.py (the calls that
currently log "🎤 Speaker recognition disabled, using original speaker labels"
and the other similar messages flagged) so they become plain string arguments to
logger.info; locate the logger.info(...) invocations in the transcription job
code and edit each message to drop the f prefix.

---

341-343: logger.exception already attaches the traceback — {e} in the message is redundant.

Ruff TRY401 flags this. Using logger.exception("⚠️ Error triggering transcript plugins in batch mode") (without {e}) is cleaner since the exception details are auto-appended.

🧹 Proposed fix

-            logger.exception(
-                f"⚠️ Error triggering transcript plugins in batch mode: {e}"
-            )
+            logger.exception(
+                "⚠️ Error triggering transcript plugins in batch mode"
+            )

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/workers/transcription_jobs.py`
around lines 341 - 343, The logger.exception call currently interpolates the
exception object into the message (logger.exception(f"⚠️ Error triggering
transcript plugins in batch mode: {e}")), which is redundant because
logger.exception already appends the traceback; update the call to remove the
{e} interpolation and use logger.exception("⚠️ Error triggering transcript
plugins in batch mode") so the traceback is still recorded without duplicating
the exception text (modify the logger.exception invocation where variable e is
referenced).

---

1018-1022: Remove extraneous f prefixes from strings with no placeholders (Lines 1022, 1059, 1071).

Same Ruff F541 issue as Line 89.

🧹 Proposed fix for all three

-            logger.info(f"✅ Session ended without speech (grace period expired)")
+            logger.info("✅ Session ended without speech (grace period expired)")

-                    logger.error(
-                        f"❌ Session failed - transcription service unavailable"
-                    )
+                    logger.error(
+                        "❌ Session failed - transcription service unavailable"
+                    )

-                    logger.error(
-                        f"❌ Session failed - check transcription service configuration"
-                    )
+                    logger.error(
+                        "❌ Session failed - check transcription service configuration"
+                    )

Also applies to: 1058-1060, 1070-1072

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/workers/transcription_jobs.py`
around lines 1018 - 1022, Several logger.info calls in transcription_jobs.py use
f-strings with no placeholders (triggering Ruff F541); remove the unnecessary
leading "f" so they are plain string literals. Locate the logger.info calls that
reference session_closed_at and final_check_grace_period (the message "✅ Session
ended without speech (grace period expired)") and the other similar messages
later in the same function (the logger.info invocations near the checks using
session_closed_at and any following "no speech" messages), and change them from
f"... " to regular "..." strings.

backends/advanced/src/advanced_omi_backend/services/audio_stream/producer.py (1)

142-151: Silent except Exception: pass hides Redis/parse failures.

If hget fails (connection error, timeout) or json.loads raises, the fallback to default values is completely silent. Operators lose visibility into why the end signal was emitted with potentially wrong audio format parameters, which could affect worker flush behavior.

🔧 Proposed fix

-        except Exception:
-            pass  # Fall back to defaults
+        except Exception as exc:
+            logger.warning(
+                f"⚠️ Could not retrieve audio_format for {session_id}, "
+                f"falling back to defaults: {exc}"
+            )

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/services/audio_stream/producer.py`
around lines 142 - 151, The try/except around building session_key and parsing
audio_format (the block that reads audio_format_raw via self.redis_client.hget,
json.loads, and extracts sample_rate/channels/sample_width) currently swallows
all exceptions; change it to catch Exception as e and log the failure (include
session_id/session_key and the exception details) before falling back to
defaults so Redis/parse errors are visible in logs; ensure the log call
references the same symbols (session_id, session_key, audio_format_raw) and
still allows the code to continue using the default sample_rate, channels, and
sample_width.

backends/advanced/src/advanced_omi_backend/services/transcription/__init__.py (2)

210-213: Redundant list entries after .strip().

v.strip().lower() can never produce a string with trailing whitespace, so "token " and "bearer " in the check list are dead entries. Simplify to ["token", "bearer"] to avoid confusion.

♻️ Suggested simplification

-                if k.lower() == "authorization" and (
-                    not v
-                    or v.strip().lower() in ["token", "token ", "bearer", "bearer "]
-                ):
+                if k.lower() == "authorization" and (
+                    not v
+                    or v.strip().lower() in ("token", "bearer")
+                ):

🤖 Prompt for AI Agents

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

In
`@backends/advanced/src/advanced_omi_backend/services/transcription/__init__.py`
around lines 210 - 213, The conditional that checks an Authorization header uses
redundant entries with trailing spaces because v.strip().lower() can never
include trailing whitespace; update the check inside the Authorization branch
(where k.lower() == "authorization" and v is tested) to replace the list
["token", "token ", "bearer", "bearer "] with the simplified ["token", "bearer"]
so the condition reads something like checking v.strip().lower() in ["token",
"bearer"].

---

614-616: provider_name parameter is accepted but never used.

Callers passing a provider_name may expect it to influence provider selection, but it's silently ignored. Consider removing it (breaking change) or adding a deprecation warning, or prefixing with _ to signal it's intentionally unused.

🤖 Prompt for AI Agents

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

In
`@backends/advanced/src/advanced_omi_backend/services/transcription/__init__.py`
around lines 614 - 616, The parameter provider_name on
get_transcription_provider is accepted but never used; to avoid confusion,
rename it to _provider_name (to signal intentional unused) and add a runtime
deprecation warning when a non-None value is passed: import warnings and call
warnings.warn(...) with DeprecationWarning inside get_transcription_provider,
explaining that provider_name is ignored and will be removed in a future
release; keep existing logic unchanged and reference get_transcription_provider
and the renamed _provider_name so callers see the intent.

backends/advanced/src/advanced_omi_backend/routers/modules/queue_routes.py (3)

212-213: Unused Conversation import in two functions

from advanced_omi_backend.models.conversation import Conversation is imported inside both get_jobs_by_client (Line 213) and fetch_client_jobs inside get_dashboard_data (Line 1195), but Conversation is never referenced in either function body. Safe to remove both.

Also applies to: 1191-1195

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/routers/modules/queue_routes.py`
around lines 212 - 213, Remove the unused local import of Conversation in both
functions: inside get_jobs_by_client and inside the nested fetch_client_jobs
within get_dashboard_data; locate the lines that read "from
advanced_omi_backend.models.conversation import Conversation" in those functions
and delete them (leaving the existing get_queue import intact) so no unused
Conversation import remains.

---

192-194: Exception chaining and logging.exception missing across except blocks

Ruff flags multiple patterns throughout the file (B904, TRY400):

1. raise ... from e — six raise HTTPException(...) calls inside except blocks (Lines 192–194, 344–346, 486–488, 893–895, 1026–1028, 1416–1418) suppress the original exception's traceback. Add from e to all of them, e.g.:

   -        raise HTTPException(
   -            status_code=404, detail=f"Job not found or could not be canceled: {str(e)}"
   -        )
   +        raise HTTPException(
   +            status_code=404, detail=f"Job not found or could not be canceled: {e!s}"
   +        ) from e

2. logger.exception vs logger.error — logger.error(...) inside except blocks (e.g., Lines 191, 343, 446, 485, 604, 834, 892, 962) should be logger.exception(...) so the traceback is automatically captured without passing exc_info=True.

Also applies to: 344-346, 486-488, 893-895, 1026-1028, 1416-1418

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/routers/modules/queue_routes.py`
around lines 192 - 194, Update all except blocks in queue_routes.py that
currently log with logger.error(...) and re-raise HTTPException(...) to instead
call logger.exception(...) and re-raise the HTTPException using exception
chaining (raise HTTPException(...) from e). Specifically, replace
logger.error(...) calls in the same except handlers as the six re-raises and
change each "raise HTTPException(...)" to "raise HTTPException(...) from e" so
the original traceback is preserved; keep the caught exception variable name (e)
consistent. This applies to the handlers that contain the HTTPException raises
around the blocks shown (the six occurrences) and the corresponding logger calls
in those except blocks.

---

1067-1163: Duplicate client-job-fetching logic between get_jobs_by_client and fetch_client_jobs

fetch_client_jobs (Lines 1191–1308) inside get_dashboard_data reimplements most of get_jobs_by_client (Lines 197–346) with two silent behavioural differences:

• It only checks queued / started / finished / failed registries, omitting canceled / deferred / scheduled that get_jobs_by_client covers.
• It does not recurse over dependent jobs (process_job_and_dependents), so the dashboard view for an expanded client will be incomplete.

get_dashboard_data already calls other endpoint logic (e.g., session_controller.get_streaming_status). Consider extracting the shared lookup into a reusable helper, or calling the existing get_jobs_by_client logic directly, to keep both views consistent.

Also applies to: 1191-1308

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/routers/modules/queue_routes.py`
around lines 1067 - 1163, Duplicate job-fetching logic: fetch_client_jobs inside
get_dashboard_data reimplements get_jobs_by_client but omits
cancelled/deferred/scheduled registries and dependent-job recursion; refactor to
reuse the existing logic. Replace the inline fetch_client_jobs implementation
with a call to get_jobs_by_client (or extract the shared lookup into a helper
used by both) so the dashboard uses the same registry set
(queued/started/finished/failed/canceled/deferred/scheduled) and invokes
process_job_and_dependents for each root job to include dependent jobs; ensure
permission checks (current_user.is_superuser) and redis Job.fetch usage remain
intact and propagate meta/fields the same way as get_jobs_by_client.

wizard.sh (1)

2-2: Same source failure guard needed as other scripts

Same issue as status.sh — without set -e or || exit 1, a failing check_uv.sh (via return or missing file) won't abort the script.

🔧 Suggested fix

-source "$(dirname "$0")/scripts/check_uv.sh"
+source "$(dirname "$0")/scripts/check_uv.sh" || exit 1

🤖 Prompt for AI Agents

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

In `@wizard.sh` at line 2, The script currently sources check_uv.sh without a
failure guard, so a failing or missing check_uv.sh won't abort wizard.sh; update
wizard.sh to guard the source call (or enable strict failure) — either add a
top-level set -e in wizard.sh or change the source invocation of
scripts/check_uv.sh to include a failure check (e.g., source "$(dirname
"$0")/scripts/check_uv.sh" || exit 1) so that a failing check_uv.sh causes
wizard.sh to exit; locate the source line in wizard.sh and apply one of these
guards.

wizard.py (2)

384-393: Unused result assignment; move success print to else block (ruff F841, TRY300)

With check=True, result is only assigned when the subprocess succeeds, but it's never read. The success message is also more idiomatically placed in an else clause.

🔧 Proposed fix

-        result = subprocess.run(
+        subprocess.run(
             cmd,
             cwd=service["path"],
             check=True,
             timeout=300,  # 5 minute timeout for service setup
         )
-
-        console.print(f"✅ {service_name} setup completed")
-        return True
-
-    except FileNotFoundError as e:
+    except FileNotFoundError as e:

Then add the success message in an else block:

     except Exception as e:
         console.print(f"❌ {service_name} setup failed: {e}")
         return False
+    else:
+        console.print(f"✅ {service_name} setup completed")
+        return True

🤖 Prompt for AI Agents

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

In `@wizard.py` around lines 384 - 393, The subprocess.run call assigns an unused
variable result and the success message should be in a try/else pattern; update
the try block that calls subprocess.run(cmd, cwd=service["path"], check=True,
timeout=300) to not assign the unused result (or assign to _), and move
console.print(f"✅ {service_name} setup completed") and the return True into an
else: clause paired with that try so the success path executes only when no
exception is raised (keep existing except handlers unchanged).

---

399-401: Multiple f-strings have no placeholders — remove extraneous f prefix (ruff F541)

Ruff flags the following lines (among others) as f-strings without any interpolation:

• Lines 401, 406, 407 (in FileNotFoundError/TimeoutExpired handlers)
• Lines 414, 415 (in CalledProcessError handler)
• Line 738 ("\n🔊 [bold cyan]Streaming[/bold cyan]")
• Lines 1183, 1207, 1210, 1212, 1215, 1218 (in main())

🔧 Example fixes (apply the same pattern to all flagged lines)

-        console.print(f"[yellow]   And that 'uv' is installed and on your PATH[/yellow]")
+        console.print("[yellow]   And that 'uv' is installed and on your PATH[/yellow]")

-        console.print(f"[yellow]   Configuration may be partially written.[/yellow]")
+        console.print("[yellow]   Configuration may be partially written.[/yellow]")

-        console.print(f"[yellow]   To retry just this service:[/yellow]")
+        console.print("[yellow]   To retry just this service:[/yellow]")

# … same pattern for lines 414, 415, 738, 1183, 1207, 1210, 1212, 1215, 1218

🤖 Prompt for AI Agents

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

In `@wizard.py` around lines 399 - 401, Several print statements and exception
handlers use f-strings without placeholders (ruff F541); locate the
console.print calls and exception blocks that mention FileNotFoundError,
TimeoutExpired, and CalledProcessError as well as the main() prints (e.g., the
"\n🔊 [bold cyan]Streaming[/bold cyan]" literal) and remove the unnecessary
leading "f" from those string literals so they are plain strings instead of
f-strings; update every occurrence noted in the review (the console.print in the
uv PATH message, the exception handlers around lines referencing
FileNotFoundError/TimeoutExpired/CalledProcessError, and the main() output
lines) to eliminate the extraneous f prefix.

Makefile (1)

125-131: pre-commit may not be on PATH immediately after uv tool install

uv tool install pre-commit places the binary in uv's tool bin directory (typically ~/.local/bin or $UV_TOOL_BIN_DIR). If that directory isn't on the user's PATH, the subsequent @pre-commit install lines will fail with "command not found". Using uvx (or uv tool run) sidesteps the PATH dependency entirely.

🔧 Suggested alternative

-	`@uv` tool install pre-commit
+	`@uv` tool install pre-commit --quiet
 	`@echo` ""
 	`@echo` "🔧 Installing git hooks..."
-	`@pre-commit` install --hook-type pre-push
-	`@pre-commit` install --hook-type pre-commit
+	`@uv` tool run pre-commit install --hook-type pre-push
+	`@uv` tool run pre-commit install --hook-type pre-commit

🤖 Prompt for AI Agents

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

In `@Makefile` around lines 125 - 131, The Makefile runs `uv tool install
pre-commit` then calls `pre-commit install` which can fail if the uv tool bin
dir isn't on PATH; update the Makefile so you invoke the installed tool via uv's
runner (e.g., `uvx` or `uv tool run`) instead of calling `pre-commit` directly.
Replace the two `@pre-commit install --hook-type ...` invocations with calls
that use `uvx`/`uv tool run` (or the uv-provided runner) to run `pre-commit
install` so the binary is executed from uv's tool directory regardless of PATH.
Ensure both hook-type installs (pre-push and pre-commit) are updated to use the
same approach and keep the surrounding echo/verbosity behavior unchanged.

stop.sh (1)

2-2: Same source failure guard needed as other scripts

🔧 Suggested fix

-source "$(dirname "$0")/scripts/check_uv.sh"
+source "$(dirname "$0")/scripts/check_uv.sh" || exit 1

🤖 Prompt for AI Agents

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

In `@stop.sh` at line 2, The script uses the bare source invocation source
"$(dirname "$0")/scripts/check_uv.sh" without a failure guard; update that line
to guard against missing or failing include by either checking the file exists
before sourcing or appending a fallback exit on failure (e.g., test -f
"$(dirname "$0")/scripts/check_uv.sh" && source ... || { echo "Failed to load
check_uv.sh"; exit 1; }), ensuring the sourced command string ("$(dirname
\"$0\")/scripts/check_uv.sh") is the same symbol you modify.

restart.sh (1)

2-2: Same source failure guard needed as other scripts

🔧 Suggested fix

-source "$(dirname "$0")/scripts/check_uv.sh"
+source "$(dirname "$0")/scripts/check_uv.sh" || exit 1

🤖 Prompt for AI Agents

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

In `@restart.sh` at line 2, The source invocation for scripts/check_uv.sh in
restart.sh should be guarded like the other scripts: first verify the file
exists and is readable, then source it and if sourcing returns a non-zero status
emit a clear error and exit non-zero; update the line that calls source
"$(dirname "$0")/scripts/check_uv.sh" to perform these checks and fail-fast on
missing/unreadable file or on a non-zero return from the sourced script.

start.sh (1)

2-2: Same source failure guard needed as other scripts

Without set -e or || exit 1, a non-zero return from check_uv.sh silently falls through to uv run.

🔧 Suggested fix

-source "$(dirname "$0")/scripts/check_uv.sh"
+source "$(dirname "$0")/scripts/check_uv.sh" || exit 1

🤖 Prompt for AI Agents

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

In `@start.sh` at line 2, The sourced check script can return non-zero but
currently allows execution to continue to uv run; make the start.sh invocation
of check_uv.sh fail-fast so a non-zero exit aborts start.sh (either add a
top-level set -e in start.sh or change the source line that calls check_uv.sh to
enforce exit on failure, e.g. ensure the result of source "$(dirname
"$0")/scripts/check_uv.sh" triggers an immediate exit when non-zero) and keep
the uv run invocation guarded so it never runs if check_uv.sh fails.

backends/advanced/.env.template (1)

65-70: Swap GALILEO_PROJECT and GALILEO_LOG_STREAM to restore alphabetical key ordering.

dotenv-linter flags GALILEO_LOG_STREAM (line 68) as appearing after GALILEO_PROJECT (line 67), violating the section's key order convention.

🔧 Proposed fix

 # Galileo (OTEL-based LLM observability)
 GALILEO_API_KEY=
-GALILEO_PROJECT=chronicle
 GALILEO_LOG_STREAM=default
+GALILEO_PROJECT=chronicle
 # GALILEO_CONSOLE_URL=https://app.galileo.ai  # Default; override for self-hosted

🤖 Prompt for AI Agents

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

In `@backends/advanced/.env.template` around lines 65 - 70, Swap the two
environment keys so they are alphabetically ordered: place GALILEO_LOG_STREAM
before GALILEO_PROJECT in the Galileo (OTEL-based LLM observability) block;
update the sequence involving GALILEO_API_KEY, GALILEO_LOG_STREAM,
GALILEO_PROJECT (and keep GALILEO_CONSOLE_URL commented as-is) to satisfy
dotenv-linter key ordering rules.

backends/advanced/src/advanced_omi_backend/clients/audio_stream_client.py (1)

628-645: send_button_event has no audio_started guard.

send_chunks_from_file checks session.audio_started before proceeding, but send_button_event has no equivalent guard. If the button event is sent before audio-start is acknowledged (e.g., immediately after start_stream with a very tight race), the server may not have a session context yet.

Consider adding:

if not session.audio_started:
    raise RuntimeError("Stream not started; cannot send button event")

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/clients/audio_stream_client.py`
around lines 628 - 645, send_button_event must guard against sending before
audio is acknowledged: inside send_button_event, after retrieving session via
self._sessions.get(stream_id) and before scheduling the coroutine, check
session.audio_started (the same guard used in send_chunks_from_file) and raise a
RuntimeError like "Stream not started; cannot send button event" if false; this
prevents races when start_stream returns before the server has established
session context and mirrors the protection already implemented in
send_chunks_from_file.

backends/advanced/src/advanced_omi_backend/prompt_registry.py (1)

97-100: Silent except: pass masks real LangFuse errors during seeding.

If client.get_prompt() fails due to a network timeout or auth error (not just "not found"), the code silently falls through and attempts create_prompt, which will then also fail — losing the root cause. Log the swallowed exception at DEBUG level so it's diagnosable without polluting startup logs.

🔧 Proposed fix

-                try:
-                    existing = client.get_prompt(prompt_id)
-                except Exception:
-                    pass  # Prompt doesn't exist yet
+                try:
+                    existing = client.get_prompt(prompt_id)
+                except Exception as fetch_exc:
+                    logger.debug(
+                        f"Prompt '{prompt_id}' not found in LangFuse (will create): {fetch_exc}"
+                    )

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/prompt_registry.py` around lines
97 - 100, The try/except around client.get_prompt(prompt_id) currently swallows
all exceptions; change the except to capture the exception (except Exception as
e) and log it at DEBUG with context (e.g., include prompt_id and the exception)
before continuing so real LangFuse/network/auth errors are diagnosable; keep the
control flow that falls through to client.create_prompt if appropriate, but
ensure the logged message uses the module's logger (or existing logger variable)
and includes exc_info/stack trace for troubleshooting.

backends/advanced/src/advanced_omi_backend/controllers/websocket_controller.py (2)

1534-1534: Unpack _ for unused payload in PCM control mode

Ruff RUF059: payload is unpacked from parse_wyoming_protocol but never used in the control-mode branch. Rename it to signal intentional discard:

-                    header, payload = await parse_wyoming_protocol(ws)
+                    header, _ = await parse_wyoming_protocol(ws)

🤖 Prompt for AI Agents

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

In
`@backends/advanced/src/advanced_omi_backend/controllers/websocket_controller.py`
at line 1534, The unpacked value named payload from parse_wyoming_protocol is
unused in the PCM control-mode branch; rename it to _ (or _payload) to signal
intentional discard—update the unpacking line where header, payload = await
parse_wyoming_protocol(ws) in the websocket_controller's PCM control-mode
handling to header, _ = await parse_wyoming_protocol(ws) (or header, _payload)
so linters like Ruff RUF059 no longer flag an unused variable.

---

547-549: Unused user_id / user_email parameters (Ruff ARG001)

• _finalize_streaming_session (line 548): both user_id and user_email are accepted but never referenced in the body.
• _process_rolling_batch (line 1077): user_email unused.
• _process_batch_audio_complete (line 1176): user_email unused.

Remove the unused parameters from each signature and all call sites to keep the API surface clean.

Also applies to: 1077-1078, 1176-1177

🤖 Prompt for AI Agents

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

In
`@backends/advanced/src/advanced_omi_backend/controllers/websocket_controller.py`
around lines 547 - 549, Remove the unused user_id and user_email parameters from
the function signatures of _finalize_streaming_session and remove user_email
from _process_rolling_batch and _process_batch_audio_complete; then update every
call site to stop passing those arguments (search for invocations of
_finalize_streaming_session, _process_rolling_batch, and
_process_batch_audio_complete and remove the corresponding arguments), and
adjust any type hints/docstrings accordingly so the functions and their callers
remain consistent.

backends/advanced/src/advanced_omi_backend/workers/rq_worker_entry.py (1)

26-31: Silent except Exception: pass obscures OTEL init failures (Ruff S110)

init_otel already guards its own ImportError and generic exceptions internally. The only errors that escape to this outer except are unexpected module-load failures. Swallowing them with pass makes diagnosis impossible. A logger.debug with exc_info=True costs nothing at INFO log level and keeps workers unblocked:

♻️ Proposed fix

     try:
         from advanced_omi_backend.observability.otel_setup import init_otel
         init_otel()
-    except Exception:
-        pass  # Optional — don't block workers
+    except Exception:
+        logger.debug("OTEL initialization skipped", exc_info=True)

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/workers/rq_worker_entry.py` around
lines 26 - 31, The outer try/except around init_otel currently swallows all
exceptions with "except Exception: pass"; change it to log the unexpected
exception instead of silencing it: import or use the module logger and in the
except block call logger.debug("init_otel failed (non-fatal)", exc_info=True)
(or an existing logger variable) so unexpected module-load failures from
init_otel are recorded while keeping workers unblocked; reference the init_otel
import/initialization and replace the bare except in that block.

backends/advanced/src/advanced_omi_backend/observability/otel_setup.py (3)

10-13: lru_cache makes is_galileo_enabled permanently sticky.

The cached result never invalidates, so if the env var is set after first invocation (e.g., in tests), the check returns stale data. Acceptable for production, but consider adding a note or a _reset helper for test harnesses.

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/observability/otel_setup.py`
around lines 10 - 13, The is_galileo_enabled function is decorated with
`@lru_cache`(maxsize=1) which makes its result permanent and causes stale behavior
if GALILEO_API_KEY is set after the first call (common in tests); remove the
`@lru_cache` decorator or replace it with a controlled caching pattern and add a
companion reset helper (e.g., is_galileo_enabled_reset or
clear_is_galileo_cache) so tests can invalidate the cached value, and update the
function docstring to note caching behavior if you keep caching.

---

104-105: Use logger.exception to preserve the traceback.

logger.error(f"Failed to initialize OTEL: {e}") loses the stack trace, making debugging harder. Since this is a catch-all except Exception, the traceback is especially valuable.

Proposed fix

     except Exception as e:
-        logger.error(f"Failed to initialize OTEL: {e}")
+        logger.exception(f"Failed to initialize OTEL: {e}")

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/observability/otel_setup.py`
around lines 104 - 105, In the except Exception block in otel_setup.py (the
handler that currently does "except Exception as e" and calls logger.error),
replace the logger.error call with logger.exception so the full traceback is
preserved; locate the except block in the OTEL initialization function (where
logger is used) and change the logging call to use logger.exception("Failed to
initialize OTEL") while keeping the exception variable to ensure the stack trace
is emitted.

---

16-43: _session_token global is not safe under concurrent async tasks.

If multiple async tasks (or threads) call set_galileo_session concurrently, the global _session_token is overwritten and clear_galileo_session can only reset the last one. Since the current usage is in RQ workers (sequential per process), this isn't an active bug, but it will bite if this is ever called from concurrent FastAPI request handlers or async tasks.

Consider using a contextvars.ContextVar to store the token, or document the single-task assumption.

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/observability/otel_setup.py`
around lines 16 - 43, The global _session_token used by set_galileo_session and
clear_galileo_session is unsafe for concurrent async tasks because multiple
calls can overwrite the token; replace the module-level global with a
context-local storage (use contextvars.ContextVar) to store the session token
per execution context, update set_galileo_session to set the ContextVar and
clear_galileo_session to reset it via the token returned by ContextVar.set (or
use ContextVar.get/clear semantics), and keep the ImportError guards and
is_galileo_enabled() check intact so behavior is unchanged for single-task RQ
workers.

backends/advanced/src/advanced_omi_backend/plugins/router.py (1)

141-145: Mutable default in ConditionResult NamedTuple.

extra: Dict[str, Any] = {} uses a shared mutable default. Current usage never mutates it in-place, so this is safe today, but it's a latent footgun. Not blocking, just noting the risk.

🤖 Prompt for AI Agents

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

In `@backends/advanced/src/advanced_omi_backend/plugins/router.py` around lines
141 - 145, ConditionResult's field extra uses a shared mutable default ({})
which is a latent bug; change the type to Optional[Dict[str, Any]] with a
default of None (e.g., extra: Optional[Dict[str, Any]] = None) and update any
consumers of ConditionResult (places that reference ConditionResult.extra) to
treat None as an empty dict (e.g., extra or {} or copy if mutation is needed).
This avoids a shared mutable default while keeping existing semantics.