## TL;DR

dbt Cloud run logs can contain truncated JSON lines (cut off mid-record). The agent should parse line-delimited JSON defensively: try each line, skip and count the malformed ones, and get authoritative data from the run artifacts API instead of depending on the logs.

## Error

```text
agent failed to parse dbt cloud run logs: truncated json lines
```

## Steps

1. Read the logs as plain text lines, not as one JSON document. Expected: line-level handling from the start.
2. Parse each line independently with error handling; on failure, record the line number and continue. Expected: good lines parse, bad lines are skipped with a count.
3. If the truncated lines carry critical data, fetch the structured run artifacts for the run instead. Expected: complete, well-formed data from the API.
4. Report the skip count alongside results. Expected: the agent is honest about what it could not parse.
5. Continue the analysis on the clean data. Expected: no crash, no silent data loss.

## When to use

- Parsing dbt Cloud run logs fails on truncated or malformed JSON lines.
- Any line-delimited JSON stream with occasional corruption.

## When not to use

- Logs are well-formed (parse them normally).
- The needed data is only in artifacts (skip the logs entirely).

## Tool compatibility

- dbt Cloud API; any JSON parsing library with per-line error handling.

## Variant phrasings

### Log stream cut off mid-run

Treat the cutoff as the end of available data; do not invent the rest.

### Interleaved log lines from parallel steps

Parse per line and tag by step; ordering is not guaranteed.

## Why it happens

Log shipping truncates long lines and can cut streams mid-write. Strict parsers treat one bad line as a fatal error, but the rest of the log is usually fine.

## Edge cases

- A high skip count means the logs are unreliable for the question; switch to artifacts.
- Keep a sample of skipped lines for debugging the parser later.
- Never silently drop lines that might contain the failure; count and report them.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_n5-ABCXzR83aIrpne8TeSw