Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

[Bug]: add/flush returns 500 every time when an extraction reply quotes code with an unmatched `}` (everalgo brace-counting JSON extractor)

Aperta
  • Qualcosa non va?
#515 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
48/100
Tipo di issue
Bug
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
python
Ambito
ai, backend

Direzione di ricerca

Start by checking how EverOS pins everalgo and trace the episode extraction path from src/everos to the affected helper in everalgo-user-memory; the issue lists eight other duplicated helpers across three packages. Reproduce with the supplied valid JSON example, then review the suggested raw_decode fix and its unit tests. Done means the full JSON string parses correctly and the fix covers the affected helpers in the pinned stable line.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Area

src/everos

What happened?

POST /memory/add (and /memory/flush) returns HTTP 500 INTERNAL_ERROR for some conversations every time it is retried. The episode extractor's LLM reply is valid JSON, but EverOS fails to parse it:

episode_extract_retry error='Unterminated string starting at: line 3 column 14 ...'   (x2)
unhandled_exception exception_type=JSONDecodeError

The cause is _extract_json_object in the pinned everalgo packages. It finds the first balanced {...} by counting { and } characters, but it does not skip braces inside JSON string values. When the model writes code into a string (common for coding-agent conversations, e.g. "... closed the callback with `});` ..."), the extra } closes the object early. json.loads then gets a truncated string and raises JSONDecodeError. The failure is determined by the content, so retries never help and the conversation can never be stored.

Expected: a valid JSON reply is parsed regardless of what its string values contain.

Steps to reproduce

Minimal, no server or LLM needed (everalgo-user-memory 0.4.0, as pinned by EverOS 1.4.2):

import json
from everalgo.user_memory.episode import _extract_json_object

reply = '{"title": "Fix", "content": "Closed the handler with `});` and redeployed.", "summary": "Fix."}'
json.loads(reply)                        # OK, the reply is valid JSON
json.loads(_extract_json_object(reply))  # JSONDecodeError: Unterminated string starting at: line 1 column 29

End to end: send /memory/add a conversation in which the assistant pastes JavaScript, so that the model's episode content quotes a snippet like }); with more } than {.

Affected code

The same helper is duplicated in 9 modules. In EverAlgo at the everalgo-user-memory/v0.4.0 tag (the version EverOS 1.4.2 pins):

  • everalgo-user-memory: user_memory/episode.py:111, atomic_fact.py:154, foresight.py:112, profile.py:233
  • everalgo-agent-memory: agent_memory/case.py:742, profile.py:516, skill.py:325, skill_ops.py:193
  • everalgo-rank: rank/rerank.py:105

The latest stable releases (everalgo-user-memory 0.7.0, everalgo-agent-memory 0.5.1) still have it. EverAlgo main no longer does: the strict LLM-contract rewrite (EverAlgo ae523ec, shipped only in the 0.9.0rc / 0.6.0rc pre-releases) replaced these helpers with model_validate_json. So EverOS stays affected until it moves to a stable everalgo release that includes that rewrite, or until the fix is backported to the pinned line.

Suggested fix for the stable line: read exactly one JSON value from the first { with json.JSONDecoder().raw_decode(text, start) and keep the existing error messages. A backport with tests is ready.

Related: the default LLM timeout is too short for slower endpoints

Separately, in the same incident, about a third of the extraction calls to a slower OpenAI-compatible endpoint (~50 output tokens/s) were cancelled at exactly 60 s. Any episode or atomic-fact reply over ~3,000 tokens could never finish, and the synchronous add/flush returned 500 (LLMError: Request timed out) after the client's retries. Setting EVEROS_LLM__TIMEOUT_SECONDS=180 fixed it. That setting exists (see #333) but is missing from .env.example and the configuration docs, next to the embedding and rerank timeouts that are listed. Suggestion: document it there, and mention that one add can take up to (retries + 1) x timeout, so that reverse-proxy and client timeouts can be sized to match.

Environment
EverOS 1.4.2 (PyPI), Python 3.12, Linux
everalgo-user-memory 0.4.0, everalgo-agent-memory 0.4.0, everalgo-rank 0.4.1, everalgo-core 0.3.0
LLM: OpenAI-compatible gateway
Logs or screenshots
episode_extract_retry error='Unterminated string starting at: line 3 column 14 (char 67)'
episode_extract_retry error='Unterminated string starting at: line 3 column 14 (char 82)'
unhandled_exception exception_type=JSONDecodeError   -> HTTP 500 INTERNAL_ERROR

A tested backport of the raw_decode fix for all 9 modules (with unit tests) is ready: which maintenance branch of EverAlgo would you like the backport PR against?

Lingua principale
Python
Stelle
13.3k
Fork
926
Merge medio
17h 8m
PR unite (30g)
32

Preparare l'ambiente

Questo progetto non fornisce container di sviluppo, Dockerfile né guida per i contributori, quindi l'ambiente è a tuo carico: parti dal suo README e consulta la nostra guida al primo contributo per i passaggi generali.

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di EverMind-AI/EverOS

Tutte le issue di EverMind-AI/EverOS

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.