--quiet does not suppress printer output despite its help text

• Status: Done
• Kind: Bug
• Authors: jp
• Date: 2026-09-04
• Label: client=cli
• Label: package=jp_printer
• Label: type=bug

--quiet is documented as "Suppress all output, including errors" (crates/jp_cli/src/lib.rs:144-146), but it only turns tracing off: configure_logging sets the level filter to OFF and drops the stderr log layer (lib.rs:1344-1346, lib.rs:1403-1404). Nothing else consults the flag.

So jp -q query still prints the assistant response, tool headers, retry notices, and the non-standard-finish notice. The run's error message is the sharpest contradiction: run writes it with eprintln! regardless of the flag (lib.rs:444-452), which is exactly what the help text promises not to do.

The single exception is jp c grep, which reads ctx.term.args.quiet itself (cmd/conversation/grep.rs:218-244) to short-circuit to an exit status. That is grep's own -q semantics, not a general rule.

The docs disagree with each other too

• RFD 072 (Discussion) says plugin output "goes through the protocol as print commands, which JP routes through its printer", giving plugins "automatic support for --quiet" (lines 127-131, 649-653). The printer has no such support.
• D15 defines -q as tracing-only (line 122), which matches the code, and lists graduated levels (-qq = no chrome) as future work spanning the Printer, ChatResponseRenderer, and interactivity systems (lines 223-225).
• D32 defines -q as "Suppresses chrome. Does not affect tracing." (line 128), the inverse of what is implemented.

Where the fix goes

The Printer is the only place that sees every emission, so the flag belongs there rather than at each of its call sites. Printer::sink() already discards everything (crates/jp_printer/src/printer.rs:182), which makes the crude version a one-line change in run_inner. It is crude because it is all-or-nothing: it would swallow command data on stdout too, and jp -q c path printing nothing is not what a script wants.

Deciding what -q covers is the actual work:

• chrome only (stderr), leaving data on stdout, per RFD 048's channel split
• everything, matching today's help text
• graduated levels, per D15's future-work note

Whichever is chosen, the help text and the behaviour need to end up saying the same thing.

Why it is filed rather than fixed in place

Raised while reviewing PR #1074, which adds a chrome line announcing a run that left the cwd's workspace. Guarding that one line on !quiet would have made it the only printer emission in the CLI honouring the flag while everything around it ignores it.

Severity

Contained and visible. A user asking for silence gets output; nothing is lost or corrupted, and the exit status is unaffected.