dsh-debug-mode

Debugging AI-agent-involved front-end or Electron projects is often a game of scattered evidence: browser consoles, Electron logs, and Node server output live in separate places, the agent can only guess from a pasted snapshot, and the temporary console.log lines you added to hunt the bug tend to linger in the code base. dsh-debug-mode collapses the entire evidence chain into a single structured JSONL stream. One to three surgical probes capture only the fields directly tied to the current hypothesis, so after each reproduction the agent reads fresh runtime evidence and can prove or reject the failing branch instead of scrolling through thousands of unstructured lines.\n\nRunning /debug activates a Host-enforced runtime phase that requires the agent to create a session, start a log transport, and place marked probes before asking for a reproduction. Once evidence arrives, the agent either applies the smallest supported fix or opens a new evidence round; it cannot silently fall back to console-only output or skip verification. Clicking Fixed triggers automatic cleanup: probe removal, ingest-process shutdown, and local log deletion, leaving no stray instrumentation behind. Four bundled helpers for session bootstrapping, JSONL ingestion, evidence summarization, and leftover-instrumentation detection can also be invoked independently for custom instrumentation or CI pipelines.\n\nThe plugin is especially useful for hunting down UI state desynchronization, worker-boundary communication anomalies, async races, and serialization failures—bugs that static code inspection alone cannot prove. It is a good fit for developers who want the agent to own the full evidence-to-fix-to-verification-to-cleanup loop, and for engineering teams that need a consistent JSONL evidence format reusable in CI workflows.