Skip to content

test(e2e): human scenarios for the 0.15.42 release - #82

Open
savass33 wants to merge 9 commits into
mainfrom
test/human-e2e-scenarios
Open

savass33 wants to merge 9 commits into
mainfrom
test/human-e2e-scenarios

Conversation

@savass33

@savass33 savass33 commented Nov 4, 2026

Copy link
Copy Markdown

O que é

Testes de terminal que agem como uma pessoa usando a CLI. Cobrem chaves próprias da Anthropic e da OpenAI, logins antigos, a troca de modelo quando o limite acaba, sessões de versões anteriores e o auto mode, tanto pelo roteador quanto na conta do próprio usuário.

Provedores e chaveiro são falsos, então nenhum desses testes toca conta real. A API real da OpenAI e a da Anthropic não são exercitadas aqui.

Também registra no UPSTREAM_SYNC.md os pontos adicionados nesta release.

Ordem de merge

Por último. Os cenários dependem de #74, #75, #76, #77, #78, #79, #80 e #81. Esta branch já contém #74 e #79, então o diff só fica limpo depois que elas entrarem.

Como rodar

bun scripts/check-cli-quality.ts --suite
bun scripts/check-cli-quality.ts --terminal
node --test --test-concurrency=1 --test-reporter=spec scripts/e2e/human/limit-switch.e2e.mjs

Rode um arquivo por vez, com a máquina livre. Mais detalhes em scripts/e2e/human/README.md.

Problema conhecido

the /model picker is legible está marcado como todo: o renderizador deixa letras antigas na lista do /model (200K contextc· via sua chavesAnthropic). src/ink não foi alterado nesta release, então a 0.15.41 deve ter o mesmo problema.

🤖 Generated with Claude Code

savass33 and others added 9 commits November 3, 2026 16:25
Verboo mode skipped the TrustDialog and still marked every folder as
trusted, so project hooks, apiKeyHelper, MCP headersHelper, statusLine,
project env and plugin marketplaces declared in a freshly cloned repo ran
without a prompt. Port upstream's provider-free gate (setupScreenGates,
OpenClaude Twigpine#1864) and show the trust prompt once per folder for every
provider. Declining exits before anything runs; headless -p is unchanged.

The dialog no longer links to the Claude Code security guide. The PTY e2e
harness gains trusted/projectFiles/settingSources options and covers both
the decline and accept paths with a SessionStart hook.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… PRs

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Terminal sessions that act like a person: own Anthropic and OpenAI keys, retired
logins, the switch when a limit runs out, sessions from older releases, and auto
mode on the router and on the user's own account. Fake providers and a fake
keyring keep them away from real accounts.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rios

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Chained Request objects follow the caller's signal through weak references, so a
garbage collection during a hung request lost the abort and the classifier
scenario hung 2 times in 4. It also relaxes a timing check measured from the
request's arrival, while the CLI's clock starts before it is sent.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant