Logs and diagnostics
Start by identifying whether the issue is installation, service startup, entry readiness, or an existing Run.
For external MCP, use doctor --execution agent_ide. Offline checks inspect saved configuration and installation integrity without importing Workflow code or calling models. Online checks ask the current host for the selected entry's readiness.
Doctor can finish successfully while reporting attention or not_checked. Inspect individual checks in text or JSON. available readiness does not validate a remote model response.
Read retained logs
Logs work offline. --follow observes new files after restart without starting a service. Run ID and --service are mutually exclusive. Stop following with Ctrl+C.
For progress during a command, add --info or --debug. Default commands do not display service logs. Business JSON remains on stdout and progress goes to stderr; explicit logs --json emits JSONL on stdout.
Each service instance keeps one .intloom/logs/LOG-<timestamp>.jsonl file. Logs contain identifiers, events, timings, usage, and classified errors rather than original intents, answers, prompts, raw Tool payloads, or credentials. They are diagnostics, not recoverable execution state.
Common situations
Logging failures
A log write failure rejects new work and stops active execution at boundaries. Status reports logError, and stop remains available. Already committed effects are preserved. Retention removes closed files older than 14 days or to make room within a 1 GiB directory budget; the active service file does not rotate.
See Error reference for error categories and Recovery for safe restart behavior.