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 `@docs/docs/jobs.md`:
- Line 271: Remove the leading shell prompt from the command example: replace
the snippet "$ rq worker --serializer json" with a plain command line "rq worker
--serializer json" so it no longer triggers markdownlint MD014; update the code
block containing that exact string accordingly.

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 `@docs/docs/workers.md`:
- Around line 251-253: Remove the leading shell prompt character from the
example command in the Markdown fenced code block so it shows the raw command;
specifically update the console block that contains "rq worker --serializer
json" (currently prefixed with "$") to remove the "$" prefix and leave just the
command "rq worker --serializer json".

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 `@rq/worker/base.py`:
- Around line 1192-1193: The code checks results[7] after building a pipeline
which is brittle; instead capture the pipeline index for the job heartbeat
before adding it and use that dynamic index to inspect the job heartbeat result
and decide whether to call self.connection.delete(job.key). Concretely: record
the pipeline length (e.g., prior_len = len(pipeline.commands) or the return
position) immediately before calling job.heartbeat()/pipeline.hset for the job,
then after executing the pipeline read results[prior_len] (or the stored index)
rather than results[7]; update the references around heartbeat(),
execution.heartbeat(), job.heartbeat(), results and the
self.connection.delete(job.key) call and add a short comment explaining the
indexing approach.

In `@rq/worker/worker_classes.py`:
- Around line 191-198: The spawned "-c" Python source currently embeds values
via raw string interpolation (self.key, job.id, queue.name) which can break on
quotes/newlines and risk injection; update the code that builds the "-c" string
to escape these values using a Python literal-safe representation (e.g. use
repr() or json.dumps on self.key, job.id, queue.name) before inserting them, and
replace the direct interpolations used with Worker.find_by_key, Job.fetch, and
Queue(...) so the child process receives safely quoted literals.

---

Outside diff comments:
In `@rq/queue.py`:
- Around line 920-983: The change to Queue.parse_args removed the named return
type, breaking public API consumers that expect attribute access (e.g.,
parsed.job_id); restore a named return by reintroducing a QueueArgs NamedTuple
(or dataclass) at module scope with fields matching the returned values (f,
timeout, description, result_ttl, ttl, failure_ttl, depends_on, job_id,
at_front, meta, retry, repeat, on_success, on_failure, on_stopped, pipeline,
unique, args, kwargs) and change Queue.parse_args to return QueueArgs(...)
instead of a bare tuple; alternatively, if you intend to break the public
contract, mark parse_args private, but prefer the QueueArgs approach to preserve
compatibility.

---

Nitpick comments:
In `@rq/worker_registration.py`:
- Around line 73-78: Replace the "# type: ignore" by explicitly narrowing None
for the fallback connection in get_keys(): instead of "redis = connection  #
type: ignore", branch on "if connection is not None" and assign redis =
connection there, otherwise handle the missing connection case (raise a clear
exception or return an empty result) so nullability is checked explicitly; keep
the existing queue branch that sets redis_key = WORKERS_BY_QUEUE_KEY %
queue.name and the fallback redis_key = REDIS_WORKER_KEYS, and reference the
variables queue, connection, WORKERS_BY_QUEUE_KEY and REDIS_WORKER_KEYS to
locate the change.

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 `@rq/serializers.py`:
- Around line 61-64: Replace the runtime assert with an explicit type check:
instead of "assert not isinstance(serializer, str)" perform an if
isinstance(serializer, str): raise TypeError(...) to validate the public
argument named serializer before the existing isinstance(serializer, Serializer)
check; keep the subsequent NotImplementedError for non-Serializer objects but
use the explicit check to ensure user-provided string serializers produce a
clear, stable error rather than an AssertionError that can be stripped by -O.