gsd-debug

Systematic debugging with resumable checkpoints lets you debug complex software issues without losing context across session resets. You persist investigation progress as checkpoint-based sessions, allowing you to list in-progress runs by slug and continue debugging later after reviewing evidence or reproducing steps.

Root cause analysis during incident investigation uses a structured scientific-method debugging loop that orchestrates symptom collection, hypothesis investigation, and root-cause reporting. It uses dedicated subagents to manage a checkpoint continuation loop and perform isolated investigations without applying fixes.

To resume an interrupted debugging session, parse the continue subcommand with the specific run's slug identifier. The debug session manager retrieves the persisted execution context from the configured workflow markdown path, allowing you to pick up the investigation exactly where you left off.

Yes, you can run diagnostics on environment-sensitive bugs without applying fixes by using the diagnose flag. This mode focuses solely on gathering symptoms, investigating hypotheses, and optionally producing a root-cause report, ensuring the investigation remains purely diagnostic and non-destructive.

Limitations of checkpoint-based workflows for software diagnostics include the requirement to parse specific subcommands and flags for status checks, and the inability to automatically apply fixes. The workflow is strictly designed for investigation and root-cause reporting, not for resolving the identified code issues.

Agent orchestration supports isolated subagents for managing debug runs by separating the roles of session management and investigation. A dedicated subagent manages the checkpoint and continuation loop, while another performs the actual investigation, maintaining isolation throughout the diagnostic process.