## TL;DR

JSONL gives you structured logs that stay greppable with plain text tools.
A consistent schema across agents makes fleet-wide queries trivial.
Applies to multi-agent systems, cron fleets, and pipeline workers.

## The query

```text
JSONL logging conventions for agent fleets
```

## Use this when

- Multiple agents write logs you need to analyze.
- You want structure that stays greppable with simple tools.
- One shared schema beats per-agent formats.

## Not for

- One agent, occasional debugging (plain text logs are fine).
- You already ship to a structured logging backend (follow its schema instead).
- The logs contain secrets (fix that first; no format helps).

## Steps

1. Define the required fields: timestamp, agent name, run id, event, and message.
   Expected output: A schema every agent follows.
2. Write one JSON object per line, no pretty-printing, no multi-line values.
   Expected output: Logs that grep, jq, and simple scripts all parse.
3. Include the run id and agent name on every line, not just at startup.
   Expected output: The ability to filter a full run from interleaved fleet logs.
4. Rotate the files daily and keep a simple index of which file covers which date.
   Expected output: Bounded disk usage and fast lookups by date.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_K353neLm5wHmXGPm-ZGMiA