Session resume fails on corrupted session file (SDK less tolerant than CLI)

[ChrisRomp]

Summary

The bridge fails to resume a session when the session file contains a corrupted JSON line. The CLI handles the same file gracefully (presumably skipping or repairing the bad line), but the SDK's session.resume treats it as fatal.

Steps to Reproduce

1. Have a long-running session that accumulates a large session file (~20k+ lines)
2. Session file gets corrupted (likely from a truncated write -- e.g., bridge restart mid-write)
3. Bridge restarts and attempts to resume the session
4. /resume <session-id> also fails with the same error

Expected Behavior

Session resumes successfully (as the CLI does), either by:

• Skipping/repairing the corrupted line
• Truncating the session history to the last valid checkpoint
• Gracefully degrading with a warning rather than refusing to load

Actual Behavior

Resume fails with:

Request session.resume failed with message: Session file is corrupted
(line 20712: SyntaxError: Unterminated string in JSON at position 8287 (line 1 column 8288))

The bridge auto-resume path handles this by creating a new session (expected fallback), but /resume surfaces the error to the user with no recovery option.

Component

session-manager / SDK (@github/copilot-sdk)

Version

0689632

Platform

macOS (Darwin), Node.js v25.6.1, SDK 0.3.0, CLI 1.0.36

Additional Context

• Session ID: 4357b1a4-56c5-4340-b3dd-4f236b3647a4
• The CLI (copilot binary) can load the same session file successfully
• Root cause is likely the SDK doing strict JSON parsing line-by-line vs the CLI having error recovery
• This may be an upstream SDK issue rather than something fixable in bridge code
• Potential bridge-side mitigation: detect corruption on /resume and offer to truncate the session file to the last valid line

Reported By

Agent (automated)