HTTP 400 missing messages[N].reasoning_content after context compaction (long sessions)

Ouverte
  • Quelque chose ne va pas ?
#1,121 1 commentaire 0 réactions 0 personnes assignées Voir sur GitHub

Les mainteneurs répondent en général sous 1 jour

Personne n'a encore pris cette issue.

Évaluation

Difficulté
4/5
Temps estimé
3-5 jours
Accessibilité débutants
55/100
Type d'issue
Bug
Clarté
Plutôt claire
Activité
Active
Stack technique
ai-infra-agents, typescript
Domaine
ai-infra-agents, backend

Piste de recherche

Start by locating the harness message serializer and the compaction/history rebuild path; the issue does not name specific files or tests. Reproduce with a model that requires reasoning_content, then inspect assistant messages after compaction. Done means serialized history retains required reasoning on every assistant message, with a regression test covering compacted turns.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Description

Summary

After a long multi-turn session with auto-compaction, sending the next message fails with HTTP 400:

Missing required parameter: 'messages[50].reasoning_content'. This model needs the reasoning of every earlier assistant message...

The turn is blocked. continue sometimes works for 1–2 turns, but the error recurs. Two trace IDs from the same session are included below.

Expected Behavior

The harness should replay reasoning_content for every earlier assistant message when the selected model requires it. If context compaction drops reasoning, the compacted history should replace old assistant turns with a summary entry that does not require reasoning, rather than leaving assistant entries without the field.

Actual Behavior

After compaction, the rebuilt request omits reasoning_content on older assistant messages. The provider rejects the whole request at messages[50]. No tool call or assistant response executes; the user must retry or start a new session.

Steps to reproduce the issue
  1. Use a model that requires reasoning_content on all assistant history.
  2. Run a long session (40+ assistant turns with tool calls).
  3. Trigger auto-compaction (or /compact).
  4. Send the next user message.
  5. Observe HTTP 400 at messages[N].reasoning_content.
Command Code Version

0.1.60 (desktop, macOS; bundled @commandcode/harness 0.1.0)

Operating System

macOS

Terminal/IDE

Command Code desktop app (macOS)

Shell

zsh

Session file (optional)

Not attached — the session contains project files. Trace IDs from the two occurrences:

  • c36659e1be4cb4330a058117255da812
  • df284ce38687a51028ea4de40acfdd48
Fix prompt (optional)

In the harness message serializer, when rebuilding history after compaction, preserve reasoning_content on every role: assistant message that originally had it. If a message is summarized away, replace it with a system/summary entry instead of an assistant entry lacking reasoning. Add a regression test: build 60 assistant turns with reasoning, compact, serialize, assert every assistant message still has reasoning_content. Include message_index, message_role, provider, and model in the 400 error payload.

Additional context

Exact error text:

Error: 400 Missing required parameter: 'messages[50].reasoning_content'. This model needs the reasoning of every earlier assistant message: send back the 'reasoning_content' returned with that message. If r...

Type "continue" to try again. If the issue persists, contact support: https://commandcode.ai/discord
Trace ID: c36659e1be4cb4330a058117255da812

Second occurrence in the same session:

Error: 400 Missing required parameter: 'messages[50].reasoning_content'. This model needs the reasoning of every earlier assistant message: send back the 'reasoning_content' returned with that message. If r...

Type "continue" to try again. If the issue persists, contact support: https://commandcode.ai/discord
Trace ID: df284ce38687a51028ea4de40acfdd48

Suspected cause: compaction/history rebuild keeps {role, content, tool_calls} but drops reasoning_content; the provider validates every assistant message, not just the last one.

Workaround: type continue, or start a new session with a summary.

Langage dominant
Aucune donnée de langage
Étoiles
4.2k
Forks
377
Merge moyen
1 j 3 h
PR mergées (30 j)
6

Préparer son environnement

Ce projet ne fournit ni conteneur de développement, ni Dockerfile, ni guide de contribution : l'installation est à votre charge. Commencez par son README, et consultez notre guide de la première contribution pour les étapes générales.

Par où commencer

  1. Lisez l'issue en entier, puis le guide de contribution du projet.
  2. Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
  3. Forkez le dépôt et travaillez sur une branche.
  4. Ouvrez une pull request qui référence le numéro de l'issue.

Autres issues de CommandCodeAI/command-code

Toutes les issues de CommandCodeAI/command-code

Issues similaires

Plus d'issues AI Infra & Agents

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.