Receber por email
← Voltar para o catálogo

termscope

Controle e inspecione aplicativos de terminal programaticamente para testes de TUI, regressão visual, depuração de problemas de renderização e automação de ferramentas CLI interativas. Ele oferece captura de snapshots, entrada de teclado/texto, busca de texto e um protocolo de sessão JSON-lines.

18estrelas
Atualizado há 6 meses

Ver no GitHub ↗ Licença: MPL-2.0

Como adicionar

/plugin marketplace add mwunsch/termscope

O comando exato pode variar conforme o repositório. Confira o README no GitHub.

Para o autor da skill

Cole no README do seu repo

Mostra que sua skill está catalogada na Skillteca, gera backlink e tráfego rastreável.

[![Skillteca](https://www.skillteca.com.br/api/badge/termscope/svg)](https://www.skillteca.com.br/skills/termscope)

Skills relacionadas

Ver todas de Desenvolvimento →

Alerta por categoria

Receba novas skills de Desenvolvimento toda segunda

1 email curto, só com as skills novas de Desenvolvimento. 4 minutos de leitura, sem spam, cancela com 1 clique.

Você confirma o email no primeiro envio. Sem spam. Cancela com 1 clique.

Ler descrição completa

termscope

Drive and inspect terminal applications — Playwright for the terminal.

Prerequisites

termscope must be installed and on $PATH. Verify:

termscope --version

If not installed:

curl -fsSL https://raw.githubusercontent.com/mwunsch/termscope/main/install.sh | sh

When to use termscope

  • Inspect the visual state of a TUI application
  • Debug rendering issues by capturing terminal snapshots
  • Run visual regression tests in CI
  • Automate interactive terminal applications (fill prompts, navigate menus)
  • Verify that a CLI tool produces expected output with styling

One-shot mode

Snapshot

Capture the terminal state of a command:

# Plain text (default) — best for reading in context
termscope snapshot -- htop

# JSON — structured data for programmatic processing
termscope snapshot --format json -- my-app

# With custom terminal size
termscope snapshot --cols 120 --rows 40 -- btop

# Write to file
termscope snapshot --format svg -o screenshot.svg -- my-tui

Exec (linear sequence)

Chain actions left-to-right:

# Wait for prompt, type, press enter, snapshot
termscope exec \
  --wait-for-text "Search:" \
  --type "hello" \
  --press RET \
  --wait-idle 200 \
  --snapshot \
  -- my-tui

# Assert text is present (exit 0/1 for CI)
termscope exec --expect "Error" -- my-app

Exec flags are processed in order. Each flag is a step.

Session mode (agent driving)

Start a persistent session:

termscope session -- vim test.txt

Send JSON-line requests on stdin, receive JSON-line responses on stdout. Stderr is for diagnostics only — never protocol data.

Protocol

Request format: {"id": N, "method": "name", "params": {...}}

Response format: {"id": N, "result": {...}} or {"id": N, "error": {"code": "...", "message": "..."}}

Methods

snapshot — Capture terminal state

{"id":1,"method":"snapshot"}
{"id":1,"method":"snapshot","params":{"format":"json"}}

type — Send text characters

{"id":2,"method":"type","params":{"text":"hello world"}}

press — Send a key using Emacs notation

{"id":3,"method":"press","params":{"key":"RET"}}
{"id":3,"method":"press","params":{"key":"C-c"}}
{"id":3,"method":"press","params":{"key":"C-x C-s"}}

wait_for_text — Block until text appears

{"id":4,"method":"wait_for_text","params":{"pattern":"Ready","timeout":5000}}

wait_for_idle — Wait for output to settle

{"id":5,"method":"wait_for_idle","params":{"duration":200}}

query — Get terminal metadata

{"id":6,"method":"query"}
// Returns: cols, rows, cursor, cursor_style, cursor_visible, title, alt_screen

resize — Change terminal dimensions

{"id":7,"method":"resize","params":{"cols":120,"rows":40}}

close — End the session

{"id":8,"method":"close"}
// Returns: exit_code

Session lifecycle

  • Session starts when termscope session -- <cmd> launches
  • Session ends when: (a) agent sends close, (b) child exits, or (c) termscope receives SIGTERM
  • If child exits: {"event":"child_exit","exit_code":N} then EOF
  • Errors do NOT end the session — the agent decides what to do

Key notation

Emacs-style key notation:

| Notation | Meaning | |---|---| | C-c | Ctrl+C | | C-x | Ctrl+X | | M-x | Alt+X | | C-M-a | Ctrl+Alt+A | | RET | Enter | | TAB | Tab | | ESC | Escape | | SPC | Space | | DEL | Backspace | | <delete> | Forward delete | | <up> <down> <left> <right> | Arrow keys | | <home> <end> | Home/End | | <prior> <next> | Page Up/Down | | <f1> … <f12> | Function keys |

Key sequences (space-separated): C-x C-s means Ctrl+X then Ctrl+S.

Output formats

  • text (default) — Numbered lines with header. Best for LLM token efficiency.
  • spans — Text + per-line style runs. For understanding UI structure.
  • json — Structured JSON with metadata and lines array.
  • html — Styled <pre> with <span> elements.
  • svg — Visual screenshot as SVG.

Common patterns

Navigate a list

{"id":1,"method":"wait_for_text","params":{"pattern":"Select:"}}
{"id":2,"method":"press","params":{"key":"<down>"}}
{"id":3,"method":"press","params":{"key":"<down>"}}
{"id":4,"method":"press","params":{"key":"RET"}}

Fill a text field

{"id":1,"method":"wait_for_text","params":{"pattern":"Name:"}}
{"id":2,"method":"type","params":{"text":"John Doe"}}
{"id":3,"method":"press","params":{"key":"TAB"}}

Wait for a prompt then respond

{"id":1,"method":"wait_for_text","params":{"pattern":"Continue? [y/n]"}}
{"id":2,"method":"type","params":{"text":"y"}}
{"id":3,"method":"press","params":{"key":"RET"}}

Check for errors

{"id":1,"method":"snapshot","params":{"format":"json"}}
// Check if any line contains "Error" or "fatal"

Test at multiple sizes

{"id":1,"method":"resize","params":{"cols":80,"rows":24}}
{"id":2,"method":"wait_for_idle","params":{"duration":200}}
{"id":3,"method":"snapshot"}
{"id":4,"method":"resize","params":{"cols":40,"rows":12}}
{"id":5,"method":"wait_for_idle","params":{"duration":200}}
{"id":6,"method":"snapshot"}
Compartilhar X LinkedIn