Hacktoberfest 2026: as issues que os mantenedores marcaram para outubro, abertas e boas para iniciantes. Ver issues do Hacktoberfest

Compaction has no visual feedback — /compact and auto-compaction appear to hang

Aberta
  • Algo errado?
#928 0 comentários 0 reações 0 responsáveis Ver no GitHub

Ninguém assumiu esta issue ainda.

Avaliação

Dificuldade
3/5
Tempo estimado
1-2 dias
Facilidade para iniciantes
68/100
Tipo de issue
Bug
Clareza
Razoavelmente clara
Status de atividade
Ativa
Domínio
cli

Direção de pesquisa

Start by tracing the TUI handling for the harness's compaction_start and compaction_done events, then follow the manual /compact entry point and the automatic compaction path. Done means both paths show an in-progress indication, a completion result using the available token data, and a neutral finished or failure message when result data is absent.

Escrita pelo modelo de indexação a partir do texto da issue.

Descrição

Summary

Compaction runs with no visual feedback. Every stage of it is already recorded on disk — the compaction_start / compaction_done events fire, the transcript gets a compaction entry, and the session meta.json gets cumulative stats — but the TUI renders nothing while it runs and nothing when it finishes.

Because compaction is a summarization model call over the whole context, it is not instant (this session compacted 446,626 tokens). With no indicator, the session looks hung for the duration, and after it returns there is no way to tell it happened, how much was reclaimed, or that /compact did anything at all.

This affects both paths: manual /compact and automatic compaction near the context limit.

Expected Behavior
  • While compacting: some visible indication that work is in progress — a status/spinner line, ideally with the token count being summarized, so the session doesn't read as frozen.
  • When finished: a feed line reporting the result, e.g. tokens before → after and tokens saved. /compact is an explicit user action, so it should always confirm what it did.
  • On failure/cancel: a visible message rather than a silent no-op.

The data needed for all of this is already produced by the harness (see evidence below) — this looks like a rendering gap rather than a missing-instrumentation problem.

Actual Behavior
  • No progress indicator during compaction. The TUI is unchanged while the summarization call is in flight.
  • No completion line. Nothing reports tokens saved, before/after counts, or event count.
  • /compact produces no visible result, so there is no way to distinguish "compacted successfully" from "ignored my command".
Steps to reproduce the issue
  1. Start a session and build up a large context (or let it approach the limit).
  2. Run /compact (or wait for automatic compaction to trigger).
  3. Observe: nothing renders while compaction runs.
  4. Observe: nothing renders when it completes — no token counts, no confirmation.
Command Code Version

1.65.2

Operating System

macOS

Terminal/IDE

Ghostty

Shell

zsh

Session file (optional)

Not attached — the affected session contains private data (personal bookmarks and account context) and the template notes it holds the whole session. Happy to provide a redacted extract, or a minimal reproduction session, on request.

Fix prompt (optional)

Compaction currently renders nothing. Surface it in the feed. The harness already emits compaction_start and compaction_done (the latter with tokensSaved, only when > 0), and the session meta.json already carries {count, tokensSaved, lastTokensSaved, lastCompactedAt} — so this is a rendering change, not new instrumentation. On compaction_start, render a status/spinner line indicating history is being summarized. On compaction_done, replace it with a completed line showing tokens saved and the running session total. Make sure the manual /compact path renders the same result so the command visibly confirms what it did. If tokensSaved is 0 or the event doesn't arrive, fall back to a neutral "compaction finished" line rather than leaving the spinner up.

Additional context

Compaction is frequent and does substantial work, all of it invisible. Aggregated across the 10 sessions on this machine that have compacted:

Metric Value
Sessions with compaction 10
Total compaction events 33
Total tokens reclaimed 16,665,573

Largest single session: 9 events, 5,423,527 tokens reclaimed.

Affected session (this one):

// ~/.commandcode/projects/<project>/6672a74d-….meta.json
"compaction": {
  "count": 1,
  "tokensSaved": 351429,
  "lastTokensSaved": 351429,
  "lastCompactedAt": "2026-09-25T02:38:11.873Z"
}

That compaction saved 351,429 tokens (446,626 before → ≈95,197 after). The user-visible result of that was nothing.

The transcript records it as a first-class entry:

// <session>.jsonl — entry type "compaction"
{
  "type": "compaction",
  "id": "002d875a",
  "parentId": "8e7d9c82",
  "timestamp": "2026-09-25T02:38:11.871Z",
  "firstKeptEntryId": "bbf7600d",
  "tokensBefore": 446626
}

Entry-type counts in that transcript: message 429, session 1, compaction 1.

The events the harness emits (from the mods API reference):

compaction_start / compaction_done | compaction mod | tokensSaved (done, only when > 0)

(subagent_start/subagent_stop are SubagentStart/Stop; compaction_start/compaction_done are Pre/PostCompact; notice is Notification.)

So compaction_start and compaction_done are both available and compaction_done already carries tokensSaved.

Related gap — script hooks cannot see compaction. The documented hook events are PreToolUse, PostToolUse, Stop, and SessionStart. There is no PreCompact/PostCompact for settings.json hooks (the mods API has them, script hooks do not). That means users who want their own compaction indicator or log currently have no hook-level way to do it, which makes the built-in rendering the only route. Worth considering as a follow-up.

Note on the docs: product-help.md cites the project as https://github.com/commandcode/command-code, which 404s (the commandcode org has 0 public repos). The live repo is CommandCodeAI/command-code. Minor, but it is where the in-product /feedback flow points users.

Linguagem predominante
Sem dados de linguagem
Estrelas
4k
Forks
357
Nenhum PR com merge em 30d

Preparar o ambiente

Este projeto não oferece contêiner de desenvolvimento, Dockerfile nem guia de contribuição, então a configuração fica por sua conta: comece pelo README e veja nosso guia da primeira contribuição para os passos gerais.

Primeiros passos

  1. Leia a issue inteira e depois o guia de contribuição do projeto.
  2. Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
  3. Faça um fork do repositório e trabalhe em uma branch.
  4. Abra um pull request que referencie o número da issue.

Mais de CommandCodeAI/command-code

Todas as issues de CommandCodeAI/command-code

Issues semelhantes

Mais issues de CLI

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.