agent's tail-log parser failed when JSON log format changed mid-stream

TL;DR: Only JSON-parse tail lines that start with { after trimming, and pass everything else through as plain log text. wrangler tail --format json mixes human-readable banner lines into the same stream, and the JSON schema drifts between wrangler versions, so a parser that assumes every line is JSON dies on the first banner.

agent's tail-log parser failed when JSON log format changed mid-stream

Steps

1. Run wrangler tail --format json and capture the first 20 lines raw, without parsing. Expected: a mix of JSON objects and plain-text lines like connection banners.

2. Change the parser to attempt JSON parsing only on lines that start with { after trimming whitespace; forward all other lines as plain log text. Expected: banner lines no longer crash the parser; real log events still parse.

3. Sniff the format at startup: read the first 5 non-empty lines, record whether JSON mode is actually active, and emit a warning if the stream is in pretty mode instead. Expected: the agent detects the wrong mode and adjusts, instead of failing silently.

4. Pin the wrangler version in CI and in the agent's environment (e.g. a fixed wrangler entry in package.json). Expected: wrangler --version prints the same version in every environment.

5. Add a regression fixture: a saved 50-line sample of mixed tail output that the parser must handle in tests. Expected: format drift turns the test red before it breaks a production run.

Use this when

• A parser consumes wrangler tail output and crashes on non-JSON lines
• Tail output parsing broke after a wrangler upgrade
• An agent needs structured fields (level, message, timestamp) from worker logs
• JSON and pretty lines arrive interleaved on the same stream

Not for this skill when

• The tail session itself fails to connect (see the wrangler tail authentication skill instead)
• Logs are missing entirely rather than malformed
• You are parsing wrangler dev console output, which has a different format

Variant phrasings

• wrangler tail JSON parse error on banner lines
• tail log parser chokes when wrangler prints non-JSON output
• agent fails to parse wrangler tail --format json stream
• wrangler tail output format changed between versions

Why it happens

wrangler tail writes connection banners, reconnect notices, and human-readable status lines to the same stdout as the JSON log events. A strict line-delimited-JSON parser treats the first banner as corrupt input and throws. Separately, wrangler occasionally renames or reshapes fields in the JSON events, so field-exact parsers break on upgrade even when every line is valid JSON.

Edge cases

• Reconnect banners can appear mid-stream after a network blip, not just at startup; the line filter must run for the whole session.
• Extremely long log lines can be truncated by the terminal or pipe; guard the JSON parse with try/catch regardless.
• If you need guaranteed structure, consider shipping logs to a sink (e.g. a logging service) instead of scraping tail output.
• Pretty mode is the default; always pass --format json explicitly rather than assuming it.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstD1m5hZGgnh9yN8DcIsedg