fix: capture output from quick-completing commands in screen isolation (issue #96)

[konard]

Summary

Fixes #96: agent --version (and other quick-completing commands) run through screen isolation produced no output, even though the command succeeded with exit code 0.

Root Cause

GNU Screen's -L log capture uses an internal FILE* buffer with a 10-second default flush interval (log_flush = 10). For commands that complete in milliseconds (like agent --version), the buffer is never flushed to the log file before the session terminates — so the output is silently lost.

From GNU Screen source (src/ansi.c):

static void WLogString(Window *win, char *buf, size_t len)
{
    logfwrite(win->w_log, buf, len);  // writes to FILE* buffer
    if (!log_flush)
        logfflush(win->w_log);        // only flushes if log_flush == 0 (default: 10s)
}

Fix

Native logging path (screen ≥ 4.5.1): Create a temporary screenrc with logfile flush 0 and pass it via screen -c. This sets log_flush = 0, forcing screen to flush after every write — eliminating the 10-second delay.

Tee fallback path (screen < 4.5.1, e.g. macOS bundled 4.0.3): Add a 50ms retry when the log file appears empty right after session completion, handling the TOCTOU race where tee may not have flushed to disk before screen -ls shows the session as gone.

Before (output missing)

$ --isolated screen -- agent --version
│ ...
$ agent --version


✓
│ exit      0

After (output captured)

$ --isolated screen -- agent --version
│ ...
$ agent --version

0.13.2

✓
│ exit      0

Changes

File	Change
js/src/lib/isolation.js	Create temp screenrc with logfile flush 0 for native logging; add retry for empty log; clean up both temp files
js/test/isolation.test.js	Add regression test: node --version output must be captured in screen isolation (issue #96)
rust/src/lib/isolation.rs	Equivalent fix for Rust implementation
rust/src/lib/isolation_tests.rs	Add Rust regression test for version-flag commands (issue #96)
docs/case-studies/issue-96/	Deep-dive case study with timeline, root cause, and solutions evaluated
js/.changeset/fix-screen-log-flush-quick-commands.md	Patch changeset
experiments/	Diagnostic scripts used to reproduce and verify the fix

Test Plan

• All existing JS tests pass (67 pass, 0 fail)
• New regression test: should capture output from version-flag commands (issue #96) passes — captures v20.20.1 from node --version
• Rust compilation and tests pass (cargo check, cargo test, cargo fmt --check, cargo clippy)
• ESLint and Prettier pass on changed files
• Manual verification: node js/src/bin/cli.js --isolated screen -- agent --version shows version output

🤖 Generated with Claude Code

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Public pricing estimate: $5.235641
• Calculated by Anthropic: $4.730235 USD
• Difference: $-0.505406 (-9.65%)
  📎 Log file uploaded as Gist (2884KB)
  🔗 View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

🔄 Auto-restart triggered (attempt 1)

Reason: CI failures detected

Starting new session to address the issues.

---

Auto-restart-until-mergeable mode is active. Will continue until PR becomes mergeable.

[konard]

🔄 Auto-restart-until-mergeable Log (iteration 1)

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Public pricing estimate: $1.668288
• Calculated by Anthropic: $1.779365 USD
• Difference: $0.111077 (+6.66%)
  📎 Log file uploaded as Gist (4262KB)
  🔗 View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• All CI checks have passed
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag