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

'Run this as a task on <model>' becomes a reading helper; chat misreports it

Aberta
  • Algo errado?
#1,569 0 comentários 0 reações 0 responsáveis Ver no GitHub

Mantenedores costumam responder em até 1 dia

Ninguém assumiu esta issue ainda.

Avaliação

Dificuldade
4/5
Tempo estimado
3-5 dias
Facilidade para iniciantes
68/100
Tipo de issue
Bug
Clareza
Claramente especificada
Status de atividade
Ativa
Stack de tecnologia
go
Domínio
cli, testing-qa

Direção de pesquisa

Start with internal/session/readhandoff.go, especially the hand-off admission around lines 150-160 and sweepTitle around lines 324-329, then reproduce the message using the isolated profile and Python repo described. Compare the resulting tasks.json and usage.jsonl with the chat report; done means the request produces a named task proposal or clear refusal, avoids unbranched edits, and reports the task's actual model and usage.

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

Descrição

area:session bug sev:critical

Seen on: dev 837b2b0.

Behaviour

Run this as a task on glm 5.3 flash: implement parse and humanize … made no proposal and no crew line. A ◆ reading: Run this as a task on glm 5.3 flash … helper ran on z-ai/glm-5.3 ($0.032) and edited durations.py in place (uncommitted, no branch). The chat then said Task ran on glm 5.3 flash … Time: 11 seconds. Cost: $0.055 (estimate $0.203), while glm-5.3-flash spent only $0.003 in that workspace. The same words in another window made a real task on kimi-k3.

A request for a task on a named model should produce a task proposal (or a clear refusal), and the chat's report of model, time and cost must match what actually ran.

Replication

  1. Build dev 837b2b0 (git checkout 837b2b0 && make build, binary bin/codeaf), or install the dev build with curl -fsSL https://agentfield.ai/get/devaf | bash.
  2. Use an isolated profile: export HOME=$(mktemp -d), export OPENROUTER_API_KEY, and keep the default model (~deepseek/deepseek-v4-flash-latest, crew on auto).
  3. On a busy machine set task.max_load to 0 (/settings, Tasks) so the busy-machine gate does not hold tasks.
  4. Make a small Python repo: R=$(mktemp -d) && cd "$R" && git init -q && printf 'def parse(s):\n raise NotImplementedError\n\ndef humanize(n):\n raise NotImplementedError\n' > durations.py && git add -A && git commit -qm init.
  5. codeaf, then send: Run this as a task on glm 5.3 flash: implement parse and humanize in durations.py (parse "1h30m" to seconds, humanize seconds back) and add tests.
  6. Look for a ◆ reading: row instead of a proposal card; read git status (edits in place, no branch) and compare the chat's Cost: line with codeaf logs for the workspace.

Evidence

  • Screen: ◆ reading: Run this as a task on glm 5.3 flash …; chat reply Task ran on glm 5.3 flash … Time: 11 seconds. Cost: $0.055 (estimate $0.203).
  • The session's tasks.json: node 1 titled reading: …, model z-ai/glm-5.3; usage.jsonl shows glm-5.3-flash at $0.003.
  • internal/session/readhandoff.go:150-160 (hand-off admission) and :324-329 (sweepTitle gives reading: <first line>).

Guessed cause

A guess from reading the code, not a confirmed diagnosis. As in the reading-helper issue filed beside this one: the read hand-off fires before propose_task, the quick task runs on the reader's model with a write-capable belt, and the chat's summary is composed by the model rather than from the helper's real model and cost.

Acceptance

  • e2e: the message above produces a task proposal whose card names glm-5.3-flash (or the chat refuses in one line); no file changes outside a task branch.
  • e2e: any cost or model the chat quotes for a task matches the task's usage rows.

Found while writing the public docs; manual text differences are in #1545.

🤖 Generated with Claude Code

Linguagem predominante
Go
Estrelas
115
Forks
14
Merge médio
9h 37min
PRs com merge (30d)
755

Preparar o ambiente

Ainda não verificamos os arquivos de configuração deste projeto. 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 Agent-Field/CodeAF

Todas as issues de Agent-Field/CodeAF

Issues semelhantes

Mais issues de Go

Receba novas issues na sua caixa de entrada

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