← ClaudeAtlas

terminal-stresslisted

Diagnose and reproduce TUICommander terminal truncation, duplicated history, missing output, Claude/Ink full-frame repaint pollution, slash-parser pressure, and false busy/idle reports. Use when a live tab looks corrupted, rows appear partial or duplicated, an agent is marked idle while working, or terminal integrity changes need repeatable PTY stress validation.
sstraus/tuicommander · ★ 172 · Data & Documents · score 87
Install: claude install-skill sstraus/tuicommander
# Diagnose terminal integrity Treat raw PTY bytes, canonical grid rows, clean log output, agent transcript, and lifecycle state as separate evidence. Never call a visual anomaly database corruption until those layers have been compared. ## Preserve volatile evidence first 1. Identify the exact session and confirm its cwd/display name. 2. Capture the raw ring before making additional requests; it rotates under sustained output. 3. Capture dimensions and canonical rows immediately afterwards. 4. Record lifecycle fields (`shell_state`, `agent_state`, background work, and current task) at the same time. 5. Read logs and transcripts only after the volatile terminal evidence is safe. Use the repository capture helper: ```bash python3 tests/terminal-stress/capture.py \ --session SESSION_ID \ --base-url http://127.0.0.1:9876 \ -o /tmp/tuic-terminal-capture ``` Port `9876` is the live orchestrator. Use it only for read-only capture of the session Boss identified. Never run synthetic workloads or create throwaway sessions there. The capture must contain: - `raw.bin`: pre-transform PTY bytes from the flight recorder. - `canonical.txt`: rows returned by the backend grid. - `meta.json`: dimensions, row range, scroll state, and capture time. Do not commit an unsanitized live capture. It may contain prompts, paths, credentials, or repository data. ## Compare the layers Use a distinctive visible string such as a banner, table header, or allegedly duplicated sentence. C